# Helpfeel: full content corpus > The complete clean-Markdown body of every Helpfeel knowledge-base, documentation, blog, and case-study article, concatenated for AI ingestion. Each article is delimited by its title and a Source: URL. Only publicly indexable content is included; gated, paid, and draft pages are intentionally excluded. --- ## AI Agent Operations: Run, Measure, and Improve Workflows Source: https://en.helpfeel.com/ai-agent-operations # AI Agent Operations: Run, Measure, and Improve Workflows AI agent operations is the work that starts where the demo ends: running, measuring, and improving an agent once its actions have real consequences. Your prototype ran. Who owns the first bad handoff? In practice, AI agent operations connects each agent to a named owner, a defined outcome, clear boundaries, visible evidence, and a controlled process for change. That is the operating system for AI work after launch. ## What does AI agent operations mean? An operated agent sits inside an owned business workflow. Its technical execution matters, but so do the customer outcome, the handoffs, the exceptions, and the decisions people make after reviewing its work. The practice covers five jobs: 1. Map the work from trigger to verified outcome. 2. Define what the agent can do, what it must show, and when it must stop. 3. Name one person who owns the workflow outcome. 4. Measure the result and the path used to produce it. 5. Turn failures and human corrections into controlled improvements. The controls will vary by workflow. A read-only research agent needs different limits from an agent that can issue refunds, publish content, or update a customer record. But the principle doesn't change: **more authority requires stronger evidence, tighter boundaries, and clearer human decisions.** The [NIST AI Risk Management Framework](https://airc.nist.gov/airmf-resources/airmf/5-sec-core) organizes AI risk work into four connected functions: govern, map, measure, and manage. It also calls for clear roles, ongoing monitoring, periodic review, and assigned responsibility for deployed systems. AI agent operations turns those ideas into daily decisions around a specific workflow. ## How are workflows, execution loops, and operating loops different? These terms describe three levels of work. Keeping them separate prevents a team from improving an agent's output while the business process around it keeps breaking. | Level | What it describes | Main question | Owner | | -------------------- | ---------------------------------------------------------------------------------------------- | ----------------------------------------- | ------------------------- | | Business workflow | The path from a real trigger to a verified customer or business outcome | Did the work produce the intended result? | Business or process owner | | Agent execution loop | The agent's sequence of observing, deciding, using tools, and checking progress during one run | Did the agent take a sound path? | Agent engineering team | | Operating loop | The recurring cycle used to run, review, and improve the deployed workflow | What should change before the next cycle? | Loop owner | Google Cloud's guide to [production-ready AI agents](https://cloud.google.com/blog/products/ai-machine-learning/a-devs-guide-to-production-ready-ai-agents) recommends evaluating the full trajectory of an agent's decisions and actions, then using staged rollouts before granting wider production access. We use **AI loop** as a short name for the operating loop. It includes the agent, the person reviewing its work, the evidence they use, the exceptions they handle, and the changes they approve. ## How do you map a workflow before assigning it to AI agents? Map the current work in enough detail that another person could run it and know when it is complete. "handle refunds" isn't a workflow. It hides the inputs, decisions, systems, exceptions, and proof that make the work reliable. Start with one bounded outcome. Then record: 1. **Trigger:** What event starts the work? 2. **Inputs:** What facts, records, and permissions are required? 3. **Steps:** What actions happen, and in what order? 4. **Decisions:** Which rules are fixed, and which require judgment? 5. **Systems:** Which tools and sources can the agent use? 6. **Handoffs:** When does work move to another agent or a person? 7. **Exceptions:** What conditions leave the normal path? 8. **Completion evidence:** What proves the intended result occurred? 9. **Readback:** Which live system confirms that a write or action persisted? 10. **Measures:** Which customer and business outcomes should improve? The result is an executable description of the work, detailed enough to run and verify. For customer service teams, software can automate routing and repeatable steps. AI agent operations adds the ownership, evidence, exception handling, and improvement cycle required when that software can interpret context and choose actions. ## What is an AI agent operating contract? An AI agent operating contract is a plain-language record of what an agent is allowed and expected to do. It gives the business owner, technical team, and reviewer one shared definition of safe and useful operation. A good operating contract covers: - The intended customer and business outcome - Required inputs and trusted sources - Allowed tools and permission limits - Expected outputs and evidence - Quality and completion tests - Human approval points - Stop conditions - Exception and escalation paths - Readback and rollback requirements - Review frequency - The person with authority to approve changes The operating contract defines the business agreement around the agent's behavior. The prompt guides how the agent behaves inside that agreement. It should also state what the agent must never do. A content agent might draft a page but require human approval before public release. A finance agent might match records but never move money. A support agent might resolve routine questions but transfer account closures or high-risk complaints to a person. ## Who owns an AI agent workflow? A named person must own the workflow outcome. Technical teams can own models, infrastructure, and reliability. Risk teams can set policy, and subject experts can review difficult cases. Business ownership still belongs to one accountable person. We call that person the **loop owner**. The loop owner has six jobs: 1. Define the intended customer and business outcome. 2. Own the normal path and the exception path. 3. Set permissions, stop conditions, and human gates with the technical and risk teams. 4. Review evidence, quality, cost, and downstream impact. 5. Turn failures and human corrections into an improvement queue. 6. Approve the next change and verify its effect. The loop owner needs the authority and context to decide whether the workflow is useful, safe, and ready to expand. Model debugging and prompt work can stay with the technical specialists. Several people may advise on the work. Name one accountable owner anyway. That person decides whether the loop continues, changes, expands, or stops. If ownership still resolves to a team, ask again: which person can make those decisions? ## Where should humans review or stop an AI agent? Put human decisions where errors carry a high cost, where the agent lacks authority, or where the work depends on context the system cannot reliably access. Design that review into the workflow before an incident demands it. Use four control types: | Control | Use it when | Example | | ---------- | ------------------------------------------------------------ | --------------------------------------------------------- | | Approval | The agent prepares an action that a person must authorize | Public claims, pricing, contract language | | Exception | The normal path cannot continue with available rules or data | Conflicting records, missing owner, unclear policy | | Escalation | The issue requires more authority, empathy, or expertise | High-risk complaint, security concern, unusual refund | | Stop | Continuing could create harm or an invalid result | Failed readback, missing permission, unsafe tool response | Every control needs a named recipient and a time expectation. "Send to a human" doesn't tell you which person or team receives the case, what evidence they get, or what decision puts the case back into operation. ## How should you launch an AI agent workflow? Launch in stages. Each stage should expose a different class of failure before the agent gets more authority. 1. **Historical test:** Run past cases with known outcomes. 2. **Sandbox:** Use synthetic or isolated data and no live side effects. 3. **Shadow mode:** Run beside the current process without changing live state. 4. **Narrow production:** Limit volume, users, permissions, or case types. 5. **Controlled expansion:** Increase authority only after a defined review. Shadow mode is especially useful because you can compare the agent with current work under real conditions. When the two disagree, classify the failure before choosing a fix. The cause could be the model, the prompt, missing knowledge, a broken workflow, an unclear policy, bad source data, or a failed integration. Each one calls for different work. Google Cloud recommends a similar path from sandbox to limited rollout and then production. The gate between stages is what matters. Expansion should be a decision backed by evidence, not a date on a launch plan. ## What should you measure for an AI agent? Measure the outcome, the work, and the operating cost together. One success rate cannot tell you whether the agent is helping the business. Use six groups of measures: | Measure | What to ask | | ---------------- | ------------------------------------------------------------------------------------ | | Customer outcome | Did the customer complete the task? Was the answer useful and timely? | | Business outcome | Did the workflow improve the target result without creating work elsewhere? | | Task quality | Was the output correct, complete, grounded, and compliant? | | Reliability | Did the agent use the right tools, recover from errors, and complete readback? | | Human work | How often did people review, correct, approve, or take over? How long did that take? | | Cost and speed | What did each completed outcome cost, and how long did it take? | Then look downstream. Check whether the agent's output created repeat work, unusable records, or broken follow-on steps. A green task score can hide a very human mess. Microsoft describes AI observability as the ability to monitor, understand, and troubleshoot AI systems through evaluation, monitoring, and tracing. Its [observability guidance](https://learn.microsoft.com/en-us/azure/foundry/concepts/observability) includes task completion, tool-call accuracy, latency, errors, quality scores, and post-production monitoring. Your business measures need to sit beside those technical signals. ## How do AI workflows improve without losing control? A useful improvement item names the failure class, its effect, the proposed change, the owner, and the test that will show whether the change worked. Turn that evidence into one approved change at a time. Use a simple improvement cycle: 1. Collect failed cases, exceptions, corrections, and outcome changes. 2. Classify the cause before proposing a fix. 3. Rank improvements by effect and frequency. 4. Approve one bounded change. 5. Test it against historical and current cases. 6. Release it through the correct gate. 7. Compare the new evidence with the baseline. 8. Keep, revise, or reverse the change. Human corrections are valuable evidence. They become useful when your team records why each correction was needed, then connects it to a lasting change in the workflow, knowledge, policy, tools, or agent instructions. ## How does AI agent operations change human work? As agents take on more execution, people gain broader responsibility for judgment, relationships, system quality, and improvement. The work doesn't disappear. The bottleneck moves. In customer service, routine questions can move to AI while service professionals focus on complex issues, customer relationships, knowledge quality, and root causes. This is also how AI agent operations connects to The Loop for customer support. AI agents handle repeatable interactions. Human service agents review what the system couldn't complete, curate useful data, improve knowledge, and drive business action. Each cycle should make the next one better. ## Related reading - [Loop Ownership 101](/blog/loop-ownership-101): run the first exception, weekly review, and first month of an AI loop. - [Customer service workflow software](/customer-service-software/customer-service-workflow-software): automate routing and repeatable steps. - [Voice of customer analytics](/voc-analytics): turn customer interactions into ranked signals for improvement. - [Customer support](/customer-support): connect repeatable work, human judgment, and self-service. - [The Helpfeel platform](/platform): see how Helpfeel combines AI agents, knowledge, and continuous improvement. ## Frequently asked questions ### What is AI agent operations? AI agent operations is the practice of running, measuring, and improving AI agents inside real business workflows. It gives each workflow a human owner, an operating contract, clear limits, useful measures, and a controlled process for change. ### Who should own an AI agent? A named business owner should own the workflow outcome and improvement decisions. Technical teams can own the agent platform, but the person accountable for the work must decide its boundaries, exceptions, measures, and next change. ### What should you measure for an AI agent? Measure task completion, output quality, customer and business outcomes, exceptions, human review, cost, and change over time. A fast agent that creates downstream work or customer problems is not performing well. ### What is an AI loop? An AI loop is the operating cycle around an agent workflow: run the work, review evidence, classify failures, approve one improvement, and verify its effect. It is separate from the agent's internal reasoning loop. AI agent operations works when every run leaves evidence, every exception has a destination, and every change has an owner. That is the job after launch: give the agent only the authority the evidence has earned, and give one person the power to decide what happens next. --- ## AI Help Center: The Complete Guide Source: https://en.helpfeel.com/ai-help-center # AI Help Center: The Complete Guide An AI help center is a self-service support platform that uses artificial intelligence to surface answers and resolve customer questions before they reach a human agent. It combines a knowledge base with an AI layer that reads what customers ask, retrieves the right article, and delivers the answer in conversational language. Strong AI help centers reduce ticket volume, cut support costs, and free your team to focus on work that needs a human. This guide covers what an AI help center is, how it differs from a traditional help center, how to build one, and how the core technologies work together. Each section links to a full guide if you want to go deeper. ## What is an AI help center? An AI help center is a knowledge base powered by artificial intelligence that answers customer questions automatically. It includes a searchable library of help articles, an AI agent that retrieves the right answer when a customer asks a question, and routing logic that escalates complex inquiries to a human when the AI cannot resolve them. Traditional help centers require customers to search for answers manually. AI help centers let customers ask in plain language and get an instant, accurate response. The AI layer understands intent, matches it to the right content, and returns the answer without requiring the customer to know which article to look for. The result is faster resolution, higher self-service rates, and fewer tickets reaching your team. According to Freshworks, conversational AI is projected to save $80 billion in contact center labor costs by 2026. ## What is the difference between a help center and an AI help center? A traditional help center is a static library of articles. Customers search by keyword, browse categories, or scroll through FAQs. If they find the right article, they get an answer. If they don't, they open a ticket. An AI help center adds an intelligent layer on top. Customers type or speak their question in natural language. The AI reads the intent, searches the knowledge base, and returns the answer in conversational format. If the knowledge base has no good answer or the inquiry is too complex, the AI routes the customer to a human agent with full context. The difference shows up in resolution speed and self-service rates. Traditional help centers resolve inquiries only when customers successfully navigate to the right article. AI help centers resolve inquiries by understanding the question and delivering the answer directly. For more on the technology behind this, see the [conversational AI for customer service guide](/ai-help-center/conversational-ai-for-customer-service). ## How does an AI help center work? An AI help center works in three steps: understand the question, retrieve the answer, and deliver it or escalate it. First, the AI reads what the customer typed or said and identifies the intent. This uses natural language processing to understand that "how do I reset my password" and "I forgot my login" are the same question. Second, the AI searches the knowledge base for the article that answers that question. It ranks results by relevance, pulls the most accurate answer, and formats it in conversational language. Third, the AI either resolves the inquiry or routes it to a human. If the answer is complete and the customer is satisfied, the inquiry closes. If the AI cannot find a good answer, or if the customer asks a follow-up that requires judgment, the inquiry escalates to an agent with context intact. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The platform handles the content work, the AI layer, and the handoff logic in one system. ## What are the benefits of an AI help center? AI help centers reduce support costs, improve resolution speed, and scale without adding headcount. **Cost reduction:** According to industry benchmarks compiled by Kodif, AI implementations reduce cost per contact by 20 to 30 percent and cut overall contact center operational costs by 30 percent. Teams spend less on repetitive inquiries and more on complex work that drives value. **Faster resolution:** Customers get answers instantly instead of waiting in a queue. According to a survey by Intercom, 53 percent of customer support professionals reported noticeably shorter response and resolution times as the top benefit of AI tools. **Higher self-service rates:** AI help centers resolve inquiries before customers open a ticket. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. **Scalability:** Self-service scales without adding staff. Every percentage point increase in self-service means fewer tickets per customer, which means you can serve more customers with the same team. For a detailed breakdown of how to measure these outcomes, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## What technologies power an AI help center? AI help centers combine four core technologies: a knowledge base, natural language processing, generative AI, and routing logic. **Knowledge base:** A searchable library of help articles that covers every common question customers ask. The knowledge base is the foundation. If the content is incomplete or unclear, the AI cannot resolve inquiries accurately. **Natural language processing (NLP):** The technology that lets the AI understand customer questions in plain language, identify intent, and match it to the right article. NLP handles variations in phrasing so "how do I cancel" and "I want to stop my subscription" return the same answer. **Generative AI:** The layer that drafts conversational responses, summarizes long articles, and adapts answers to context. Generative AI turns static knowledge base content into dynamic, personalized responses. **Routing logic:** The rules that determine when the AI escalates an inquiry to a human. Strong routing logic ensures customers reach an agent when the AI cannot help, when the question requires judgment, or when the customer asks to speak with a person. Each technology works together. The knowledge base holds the answers, NLP finds the right one, generative AI delivers it conversationally, and routing logic ensures nothing falls through the cracks. For a deeper look at how generative AI fits into customer service workflows, see the [generative AI in customer service guide](/ai-help-center/generative-ai-in-customer-service). ## How do you build an AI help center? Build an AI help center in four steps: audit your knowledge base, choose an AI platform, deploy the AI agent, and measure results. ### 1. Audit and organize your knowledge base Pull three months of customer inquiry data and group it by question type. Identify the repetitive questions that account for most of your volume. For each question, write or update one clear, complete article. Use the question itself as the heading and answer it in the first sentence. If your knowledge base already exists, audit it. Articles that customers search for but don't read, or that generate follow-up questions, aren't helping. Update them or remove them. ### 2. Choose an AI help center platform Look for a platform that includes the knowledge base, the AI agent, and the measurement tools in one system. The AI should be able to search your content, answer questions conversationally, and route unresolved inquiries to a human with full context. Helpfeel runs this as a managed service. We write the articles, deploy the AI agent, watch what customers search for, and update the knowledge base on a regular cadence so it stays current without pulling time from your team. ### 3. Deploy the AI agent where customers ask questions Your AI agent should work everywhere customers look for help: your help center, chat widget, contact forms, and email. Deploy it across all channels so customers get the same fast, accurate answers no matter where they start. Make sure the agent knows when to stop. If a question requires judgment, if the customer is frustrated, or if the knowledge base has no good answer, the agent should escalate immediately. ### 4. Measure self-service rate, containment rate, and satisfaction Track three metrics together. Self-service rate shows what share of inquiries the AI resolves. Containment rate shows whether customers find the answer and don't contact you again. Satisfaction shows whether the resolution actually helped. If self-service climbs but satisfaction drops, the AI is answering questions but not resolving them. Go back and improve the content. For more on tracking and interpreting these metrics, read the [containment rate guide](/customer-support-metrics-and-kpis/containment-rate-complete-guide). ## The six parts of a strong AI help center strategy | If you want to | Read | | ---------------------------------------------------- | ------------------------------------------------------------------------------------------------ | | Understand what an AI help center is | [What is an AI help center](/ai-help-center/what-is-an-ai-help-center) | | Build a knowledge base that customers use | [Self-service knowledge base](/ai-help-center/self-service-knowledge-base) | | Design a self-service system that resolves inquiries | [Customer self-service](/ai-help-center/customer-self-service) | | Deploy conversational AI that understands intent | [Conversational AI for customer service](/ai-help-center/conversational-ai-for-customer-service) | | Use generative AI to draft answers | [Generative AI in customer service](/ai-help-center/generative-ai-in-customer-service) | | Choose AI help desk software | [AI help desk software](/ai-help-center/ai-help-desk-software) | ## How do you measure AI help center success? Measure success by tracking whether the AI help center reduces ticket volume while keeping satisfaction steady or higher. **Self-service rate** tells you what percentage of inquiries resolve without reaching an agent. A healthy AI help center resolves the majority of routine questions automatically. **Containment rate** tells you whether customers find the answer and stop there, or whether they contact you again through another channel. High containment means the AI is actually resolving inquiries, not just responding to them. **Customer satisfaction (CSAT)** tells you whether the experience is helping or frustrating customers. If satisfaction drops as self-service climbs, the AI is getting in the way instead of helping. Track all three together. A self-service rate of 80 percent means nothing if satisfaction is falling or customers are switching to phone to get around the AI. For detailed measurement guidance, read the [self-service rate guide](/customer-support-metrics-and-kpis) and the [containment rate guide](/customer-support-metrics-and-kpis/containment-rate-complete-guide). ## What is the role of a managed knowledge base? An AI help center is only as good as the knowledge base behind it. If the content is incomplete, outdated, or unclear, the AI will return incomplete, outdated, or unclear answers. Most AI help center projects stall because the knowledge base degrades faster than the team can maintain it. A managed knowledge base solves this. Someone owns the content, audits it on a schedule, and closes gaps as they appear. The AI stays accurate because the content stays current. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates so the knowledge base improves over time without pulling time from your team. For more on building a knowledge base that works, see the [self-service knowledge base guide](/ai-help-center/self-service-knowledge-base). ## How do AI help centers fit into a broader customer support strategy? AI help centers sit at the front of the support operation. They resolve the repetitive, high-volume inquiries automatically so your team can focus on the work that requires empathy, judgment, and problem-solving. The structure looks like this: customers ask a question, the AI help center tries to resolve it, and anything the AI cannot handle routes to a human agent. The AI carries the repetitive load. The agents handle the complex cases, the escalations, and the conversations that need a person. This keeps cost per contact stable even as total inquiry volume grows, because the AI absorbs the volume increase. It also keeps your team focused on work that matters instead of answering the same five questions all day. For a broader view of how AI help centers fit into the full support operation, read the [customer support guide](/customer-support) and the [customer support automation guide](/customer-support/how-to-automate-customer-support). AI help centers also intersect with customer service workflows beyond technical support. For more on the difference and where they overlap, see the [customer service software guide](/customer-service-software). ## Frequently asked questions ### What is an AI help center? An AI help center is a self-service support platform that uses artificial intelligence to surface relevant answers, resolve customer questions, and route inquiries without a human agent. It combines a knowledge base with an AI layer that understands what customers ask and retrieves the right answer. ### How does an AI help center work? An AI help center reads the customer's question, searches the knowledge base for the answer, and returns it in conversational language. If the AI cannot resolve the inquiry, it routes the customer to a human agent with context intact. ### What is the ROI of an AI help center? AI help centers reduce support costs by automating repetitive inquiries. According to industry data, implementations reduce cost per contact by 20 to 30 percent and save contact centers up to 30 percent in operational costs through lower ticket volume and faster resolution times. ### What is the difference between a help center and an AI help center? A traditional help center is a searchable library of articles. An AI help center adds an intelligent layer that understands questions in natural language, retrieves answers automatically, and resolves inquiries before customers open a ticket. ## Go deeper An AI help center is the foundation of a scalable, sustainable support operation. Helpfeel runs the full system for you: a managed, AI-ready knowledge base plus an AI agent that resolves inquiries before they reach your team. [See how the done-for-you model works](/product). --- ## Why AI Support Gives Confident Wrong Answers Source: https://en.helpfeel.com/ai-help-center/why-ai-support-gives-wrong-answers # Why AI Support Gives Confident Wrong Answers An AI support layer is only as reliable as the knowledge it retrieves from. A model reasoning over a stale, contradictory, or incomplete corpus does not fail loudly. It produces fluent, confident answers that are wrong, at machine scale. The output looks like competence while the system quietly amplifies every gap in its source. This is a central question for any [AI help center](/ai-help-center). This piece situates where confident wrong answers come from and why they are hard to catch. It does not describe the way out. ## Where is the problem set? Not by malice, but by sequence. Leadership issues a mandate for AI in customer service. A team moves quickly and deploys retrieval over the knowledge base that already exists. The quality of that knowledge base is assumed, not verified, because the attention is on the model, which is the exciting part. The structural mismatch: the existing knowledge base was written for humans, who bring judgment and skepticism and route around a gap. A retrieval model brings none of that. It treats a stale article as ground truth with the same confidence it treats a correct one. ## Why does one error become a pattern? Trace a single wrong answer through both worlds: | | Human agent | AI layer | | ------------------ | ----------------------------- | ------------------------------------------ | | Scope of one error | One customer, one interaction | Every customer who asks a similar question | | Detection | Caught in the next exchange | Found late, by downstream effects | | Correction speed | Immediate | After a pattern emerges in complaints | A single point of decay in the source is no longer a single bad interaction. It is a systematic error pattern broadcast at the full speed of the automated channel. The leverage that makes AI support attractive is what turns one weak article into a fleet of identical wrong answers. ## Why does the trust damage land twice? An unverified foundation erodes trust on two fronts at once. Customers who catch an AI error stop trusting the channel, and often the brand behind it. Internal stakeholders who catch it stop trusting the initiative, so a program meant to demonstrate progress starts demonstrating risk. A tool that fails loudly gets fixed. A tool that fails confidently gets trusted, right up until it does not. Analysts have already forecast that service quality will dip as organizations wrestle with AI deployment complexity before the gains arrive. That dip is less a modeling problem than a foundation problem wearing a modeling costume. ## What does this force into view? The question shifts from why do we not have AI to why is our AI wrong. That second question is far more expensive, and it lands on whoever sponsored the initiative. It forces attention onto the step that was skipped: verifying, and continuously re-verifying, the knowledge the model stands on. The clarity is that an AI layer does not create trustworthy answers. It inherits the trustworthiness of its source and amplifies it, in whichever direction that source points. ## Frequently asked questions ### Why does AI support give wrong answers? Because an AI layer inherits the reliability of the knowledge it retrieves from. If the underlying knowledge base is stale, incomplete, or contradictory, the model reproduces those flaws as fluent, confident answers. ### Why are AI support errors hard to catch? The answers are fluent and confident, so a casual reviewer cannot tell a wrong one from a right one without independently knowing the correct answer. Failures are found late, by their downstream effects, after trust is already spent. ### Why is a confident wrong answer worse at scale? A human error is one bad interaction. An AI error built on a weak source repeats across every similar question at machine speed, turning one weak article into a systematic pattern of identical wrong answers. Helpfeel exists because of this exact problem. [See how we think about it](/platform). --- ## AI Help Desk Software: Features, Evaluation, and Use Cases Source: https://en.helpfeel.com/ai-help-center/ai-help-desk-software # AI Help Desk Software: Features, Evaluation, and Use Cases AI help desk software is a support platform that uses artificial intelligence to resolve customer and employee requests without requiring an agent for every interaction. It automatically answers common questions, routes tickets based on intent and priority, and gives agents tools to handle complex issues faster. When set up right, it reduces repetitive workload and shortens resolution time. This guide explains what AI help desk software does, what features matter, and how to evaluate platforms. For the broader strategy around building a help center with AI at the center, start with the [AI help center guide](/ai-help-center), which covers content structure, self-service design, and how the AI layer fits into the full customer experience. ## What is AI help desk software? AI help desk software is a ticketing and support platform that combines artificial intelligence with traditional help desk functions: intake, routing, tracking, and resolution. The AI layer handles two jobs. First, it resolves routine inquiries automatically by searching a knowledge base or executing simple actions like password resets. Second, it assists human agents by suggesting answers, summarizing conversations, and pulling relevant context from past tickets. According to Gartner, over 40 percent of initial support interactions are now handled by AI. The software works across channels: email, live chat, SMS, phone, contact forms, and social media. When a request comes in, the AI reads the question, determines intent, and either resolves it immediately or routes it to the right agent with context already attached. Traditional ticketing systems organize work but require a person to answer every request. AI help desk software adds a resolution layer. The distinction is automation that reduces volume, not just organization that distributes it. ## What features should AI help desk software include? Not all platforms that claim AI actually automate resolution. Some add sentiment analysis or ticket tagging but still send every request to an agent. Here's what the software should do if the AI label is accurate. **Core features to require:** - **AI-powered ticket routing:** The system reads the request, classifies intent, assigns priority, and routes to the correct team or agent. This happens before a human sees the ticket. - **Automated resolution for common questions:** The AI searches a knowledge base, answers the question in conversational language, and closes the request if the customer is satisfied. No agent involved. - **Knowledge base integration:** The software connects to a structured knowledge base and pulls answers from it. If the knowledge base is incomplete or unclear, the AI can't resolve requests accurately. - **Agent assist tools:** When an agent picks up a ticket, the AI suggests answers, summarizes the issue, and surfaces related articles or past tickets. This cuts handle time. - **Multi-channel support:** The AI works the same way whether the customer asks via chat, email, phone transcript, or contact form. Channels shouldn't require separate configurations. - **Handoff logic:** The AI knows when it can't help. If the question is too complex, if the customer is frustrated, or if the knowledge base has no answer, the system escalates to a person immediately and passes full context. **Features that add value but aren't required:** - Sentiment analysis to flag negative tone and prioritize those tickets - Workflow automation for approval chains or multi-step processes - Self-service portals where customers can track their own tickets - Reporting dashboards that show resolution rate, response time, and AI vs. human handle percentage The best AI help desk platforms resolve the majority of routine requests and give agents time to focus on work that requires judgment or empathy. If the platform requires an agent to touch every ticket, the AI isn't doing its job. ## How do you evaluate AI help desk software? Evaluation comes down to six areas. Score each one based on your team's size, volume, and support model. | Evaluation criterion | What to test | Why it matters | | ------------------------------- | -------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Channel coverage** | Does the platform handle all the places customers contact you? Email, chat, phone, SMS, social? | If it doesn't cover a channel, you'll need a separate tool or manual workaround. That fragments your data and adds overhead. | | **Handoff quality** | When the AI can't help, does it route the ticket cleanly to an agent with full context? | Poor handoffs force the customer to repeat themselves and frustrate your team. Test this with ambiguous or emotional requests. | | **Knowledge base integration** | Does the platform require you to build and maintain the knowledge base, or is that part of the service? | Most teams underestimate the work required to keep a knowledge base current. If you don't have dedicated content resources, look for a platform that manages this for you. | | **Agent workflow tools** | Does the AI give your agents suggested answers, ticket summaries, and relevant context when they pick up a ticket? | These tools cut handle time and reduce the cognitive load on your team. Without them, agents are just using an AI-labeled ticketing system. | | **Measurement and reporting** | Can you see self-service rate, containment rate, ticket volume reduction, and CSAT split by AI vs. human resolution? | You need data to know if the automation is working. Platforms that only report ticket volume or response time aren't showing you the full picture. | | **Content and setup ownership** | Does the vendor handle onboarding, content migration, and ongoing updates, or is that your team's job? | Implementation work is where most AI help desk projects stall. If you don't have bandwidth to write articles and tune the AI, choose a done-for-you model. | Compare platforms by running a pilot with real traffic. Send a sample of your inquiry volume through the system and measure how many requests get resolved without an agent, how many get escalated cleanly, and whether your team's workload actually drops. Platforms like Zendesk, Intercom, and Freshdesk offer strong ticketing workflows and AI add-ons. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that resolves inquiries across every channel, so your team handles less repetitive volume. The platform includes the content work, AI layer, and measurement tools in one package. ## What support scenarios is AI help desk software built for? AI help desk software works best in environments with high inquiry volume and a significant portion of repetitive questions. It's a fit for customer-facing support teams, internal IT help desks, and employee service centers. **Customer support use cases:** - Order status, tracking, and delivery updates - Account access, password resets, and login help - Billing questions, refund policies, and payment status - Product feature explanations and how-to guidance - Return, exchange, and warranty instructions **Internal IT and employee service use cases:** - Software access requests and provisioning - VPN, device, and account troubleshooting - Policy lookups for HR, facilities, and compliance - Onboarding and offboarding workflows - Expense, PTO, and benefits questions The common thread is that these questions show up in volume, have clear answers, and don't require judgment. A customer asking "where is my order" can get an instant answer. An employee asking for an exception to a travel policy needs to talk to a person. AI help desk software is not a replacement for human agents. It's the layer that handles routine volume so your team can focus on complex requests, escalations, and the conversations that require empathy or problem-solving. According to Forrester, machine learning technologies can boost IT service desk capacity by up to 30 percent. That capacity gain comes from automation carrying the repetitive load. For more context on how AI agents fit into the broader support stack, see the [AI customer support software guide](/customer-support/ai-customer-support-software) and the [customer service software overview](/customer-service-software). ## What to watch for when implementing AI help desk software Implementation is where most projects fail. The software gets deployed, the team gets trained, but the knowledge base isn't ready, the AI returns weak answers, and customers start routing around the system to get to a person faster. Here's what to check before you go live. **Knowledge base readiness:** The AI is only as good as the content it pulls from. If your knowledge base is incomplete, outdated, or written in internal jargon, the AI will return incomplete or confusing answers. Audit your content before launch. Write articles in plain language, structure them with the question as the heading, and answer in the first sentence. **Escalation rules:** Define clear triggers for when the AI should stop and hand off to an agent. Examples: the customer uses language that signals frustration, the question falls outside the knowledge base, the request requires account-level changes, or the conversation has gone back and forth more than three times without resolution. Test these rules with real traffic before you scale. **Agent training:** Your team needs to understand what the AI can and can't do. Train them on how to review AI-generated answers, when to override the suggestion, and how to flag content gaps so the knowledge base improves over time. Agents are your feedback loop. **Measurement cadence:** Set up a weekly review of self-service rate, containment rate, and CSAT. If any of those metrics drop, dig into the data. Low containment usually means the AI is answering but not resolving. Low CSAT means customers are frustrated by the experience. Fix the content or the escalation logic, then measure again. Most platforms require you to own this work. Helpfeel runs it for you. We handle content migration, write new articles based on what your customers are asking, flag gaps in real time, and update the knowledge base on a regular cadence. The AI keeps working because the content keeps improving, and your team doesn't carry the maintenance load. ## Frequently asked questions ### What is AI help desk software? AI help desk software is a platform that uses artificial intelligence to handle support requests. It automatically responds to common questions, routes tickets to the right team, and gives agents tools to resolve issues faster. It works across email, chat, contact forms, and phone. ### What features should AI help desk software include? Look for AI-powered ticket routing, automated responses to common questions, knowledge base integration, agent assist tools, multi-channel support, and handoff logic that escalates complex issues to a person. The best platforms combine automation with clear escalation rules. ### How do you evaluate AI help desk software? Evaluate on channel coverage, handoff quality when the AI can't help, knowledge base integration, agent workflow tools, measurement and reporting, and whether the platform handles the content work or requires you to manage it yourself. Match the platform to your support volume and team size. ### What's the difference between AI help desk software and traditional ticketing systems? Traditional ticketing systems organize and route requests but require an agent to answer every one. AI help desk software adds a layer that can resolve requests automatically, give agents suggested answers, and handle repetitive volume without human involvement. The difference is automation, not just organization. ## See how the done-for-you model works AI help desk software only delivers results if the knowledge base behind it stays current and the escalation logic is tuned to your volume. Helpfeel handles the content work, the AI layer, and the ongoing maintenance in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Conversational AI for Customer Service Source: https://en.helpfeel.com/ai-help-center/conversational-ai-for-customer-service # Conversational AI for Customer Service Conversational AI for customer service uses natural language processing to understand customer questions and respond in conversational language, handling inquiries across chat, email, help centers, and contact forms. It learns from interactions, improves over time, and resolves repetitive volume so your team can focus on work that needs a human. This guide covers what conversational AI is, how it differs from scripted chatbots, where it fits in your support stack, and what results you can expect. For a broader look at how AI improves self-service, see the [AI help center guide](/ai-help-center), which covers the full system that supports conversational AI. ## What is conversational AI for customer service? Conversational AI for customer service is technology that uses natural language processing and machine learning to understand customer inquiries and respond in natural language. It reads or listens to a question, identifies the intent, searches your knowledge base for the answer, and generates a response that sounds conversational rather than scripted. The system improves as it handles more inquiries. It learns which answers resolve questions, which phrases signal frustration, and when to route a conversation to a person. According to Polaris Market Research, the AI customer service market reached $12.06 billion in 2024, driven by businesses shifting repetitive volume to AI while keeping human agents focused on complex requests. Conversational AI works across every channel where customers ask questions: help centers, chat widgets, email, contact forms, and voice systems. The goal is to resolve inquiries before they become tickets, so customers get answers faster and agents handle less repetitive work. ## How does conversational AI work? Conversational AI combines three technologies to understand and respond to customer questions. 1. **Natural language processing (NLP):** Reads or listens to the customer's question and converts it into data the system can analyze. NLP identifies intent, extracts key information, and understands variations in phrasing, so the system knows that "Where's my order?" and "I need to track my shipment" mean the same thing. 2. **Machine learning:** Pulls the relevant answer from your knowledge base or data source. The AI agent searches for content that matches the customer's intent, ranks results by relevance, and selects the best answer. 3. **Natural language generation (NLG):** Converts the answer into conversational language that sounds natural to the customer. Instead of returning a raw knowledge article, the system generates a response that fits the question and the conversation. These three steps happen in seconds. The customer types a question, the AI agent processes it, searches your knowledge base, and returns a clear answer in the same channel where the question was asked. ## How is conversational AI different from a chatbot? Conversational AI and chatbots both handle customer questions, but they work differently. A chatbot follows a script. You program it with specific phrases and responses, and it can only handle the exact scenarios you've built. If a customer asks a question the chatbot hasn't been trained on, it fails or loops back to a menu. Conversational AI understands intent and context. It doesn't need a script for every variation of a question. It uses natural language processing to interpret what the customer is asking, even if the phrasing is new, and pulls the answer from your knowledge base. If a customer asks, "Can I return this after 30 days?" the AI agent understands the intent and answers based on your return policy, even if the exact question wasn't pre-programmed. | Feature | Chatbot (rule-based) | Conversational AI | | --------------------- | --------------------------------- | ---------------------------------------------------- | | How it works | Follows pre-programmed scripts | Uses NLP to understand intent and generate responses | | Handles new questions | Only responds to scripted phrases | Understands variations and infers meaning | | Learns over time | Requires manual updates | Improves with machine learning and more data | | Best for | Simple, narrow use cases | High-volume, varied inquiries across channels | Conversational AI fits where chatbots break down: when customers ask the same question in different ways, when inquiries span multiple topics, and when you need the system to improve without rebuilding it every month. ## What customer service tasks should conversational AI handle? Conversational AI works best on repetitive, high-volume questions that have clear answers but show up in different phrasings. These are the inquiries your team answers the same way every time, but that take time away from complex work. **Tasks conversational AI handles well:** - Order status, tracking, and delivery timelines - Password resets and account access - Return, exchange, and refund policies - Product features, compatibility, and specifications - Hours, locations, and availability - Billing questions that need a simple lookup - Common how-to questions **Tasks that still need a human:** - Complaints, escalations, or dissatisfaction - Requests that require judgment or exceptions - Conversations tied to emotion (cancellations, financial hardship) - Complex troubleshooting that requires back-and-forth diagnosis According to Statista, 82% of consumers in 2024 said they would use a chatbot instead of waiting on hold for a customer representative. Customers prefer instant answers for routine questions. Conversational AI gives them that speed, and it gives your team time for the conversations that require empathy or problem-solving. For more on how AI agents fit into the broader support workflow, see the [AI customer support agent guide](/customer-support/ai-customer-support-agent). ## How conversational AI fits into your support stack Conversational AI sits at the front of your support operation. It's the first system customers interact with when they visit your help center, open a chat widget, or submit a contact form. The AI agent resolves as many inquiries as it can, and routes unresolved questions to the right person on your team. This structure keeps repetitive volume out of your queue and puts human agents on the work that needs judgment. A customer asking "What's your return policy?" gets an instant answer from the AI agent. A customer asking to return an item after the return window closed gets routed to an agent who can evaluate the situation and decide. Conversational AI also surfaces gaps in your knowledge base. If customers repeatedly ask a question the AI can't answer, that signals missing content. Close the gap, and the AI agent starts resolving that question automatically. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions. The platform handles the content work, the AI layer, and the measurement, so your team can focus on conversations that need a human. For more on building the knowledge foundation that conversational AI depends on, see the [self-service knowledge base guide](/ai-help-center/self-service-knowledge-base). ## What results can you expect from conversational AI? Conversational AI reduces ticket volume, speeds up resolution time, and keeps satisfaction steady or higher when it's set up right. Teams see the strongest results when they pair the AI agent with a managed knowledge base that stays current. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. Those results come from three things: a knowledge base that answers the majority of customer questions, an AI agent that retrieves and delivers those answers in conversational language, and a team that closes content gaps as they appear. According to Master of Code, 52% of contact centers have invested in conversational AI and 44% plan to adopt it. The teams that see results measure self-service rate, containment rate, and customer satisfaction together. If volume drops but satisfaction also falls, the AI agent is answering questions but not resolving them. That's a signal to improve the content or the routing logic. For a detailed breakdown of how to track self-service performance, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## How conversational AI connects to generative AI in customer service Conversational AI and generative AI are related but not the same. Conversational AI is the broader category. It includes any system that understands and responds to customer questions in natural language. Generative AI is a type of conversational AI that creates responses dynamically instead of pulling from pre-written content. Generative AI models can answer questions your knowledge base doesn't explicitly cover by synthesizing information from multiple sources. This makes them especially useful for complex or edge-case questions where a single article doesn't exist. However, generative AI requires careful guardrails to ensure accuracy, tone, and brand consistency. For a deeper look at how generative AI works in customer service and where it fits alongside retrieval-based systems, see the [generative AI in customer service guide](/ai-help-center/generative-ai-in-customer-service). ## Frequently asked questions ### What is conversational AI for customer service? Conversational AI for customer service uses natural language processing to understand customer questions and respond in conversational language. It handles inquiries across chat, email, help centers, and contact forms, learning from interactions to improve over time. ### How is conversational AI different from a chatbot? Chatbots follow scripted rules and can only respond to exact phrases. Conversational AI understands intent, context, and variations in language. It can answer questions it hasn't seen before by pulling from a knowledge base and generating responses in natural language. ### What customer service tasks can conversational AI handle? Conversational AI handles repetitive inquiries like order status, password resets, return policies, product questions, and account lookups. It works best on high-volume questions with clear answers. Complex or emotional conversations still need a human. ### Does conversational AI replace customer service teams? No. Conversational AI resolves repetitive inquiries so your team has time for complex requests that need judgment, empathy, or authority. It augments your team by removing the work that doesn't require a human. ## See how Helpfeel combines conversational AI with a managed knowledge base Conversational AI only works if the knowledge base behind it stays current and complete. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a person. [See how the done-for-you model works](/product). --- ## Customer Self-Service: Definition, Benefits & How to Improve Source: https://en.helpfeel.com/ai-help-center/customer-self-service # Customer Self-Service: Definition, Benefits & How to Improve Customer self-service lets people find answers and resolve their own issues using knowledge bases, AI agents, and help centers, without waiting for a support rep. When it works, customers get faster answers and your team handles less repetitive volume. This guide covers what self-service is, why it works, how to raise your self-service rate, and when to route customers to a person. Start with the [AI help center guide](/ai-help-center) for the full picture of how self-service fits into modern support strategy, or jump to [how to scale customer support](/customer-support/how-to-scale-customer-support) for the operational framework. ## What is customer self-service? Customer self-service is any system that lets customers find answers, complete tasks, or resolve issues on their own, without involving a support agent. The most common forms are knowledge bases, AI help centers, chatbots, FAQs, account portals, and automated troubleshooting tools. According to Higher Logic, 92% of consumers say they would use an online knowledge base for self-support if it were available. According to American Express, 60% of customers prefer self-service tools over talking to a live agent for simple tasks. People want fast answers. They don't want to wait in a queue for information they could find in 30 seconds. Self-service works when the answer is clear and repeatable. A customer looking for return instructions, order status, password reset steps, or product specs can get an instant answer from a knowledge base or AI agent. A customer with a billing dispute or a complex technical problem needs a person. The goal of self-service is not to eliminate human support. It's to carry the routine volume so your team has time for the conversations that require judgment, empathy, or problem-solving. ## Why customer self-service works Self-service works because it's faster than email, cheaper than phone support, and available 24/7. It gives customers control. They can search, read, and act at their own pace without explaining their issue to a stranger. It also scales. A support team of five people can only handle so many tickets per day. A knowledge base served by an AI agent can handle thousands of inquiries at the same time, with no wait time and no drop in quality. **Here's why businesses invest in it:** | Benefit | What it does | | -------------------------------- | --------------------------------------------------------------------------- | | Reduces repetitive ticket volume | AI agents answer routine questions so your team focuses on complex requests | | Scales without adding headcount | Self-service handles more volume as you grow, no new hires needed | | Works around the clock | Customers in every time zone get answers, even when your team is off | | Lowers support costs | Fewer tickets per month means lower cost per resolution | | Improves response time | Instant answers beat email or chat wait times | | Increases customer satisfaction | Fast, accurate self-service raises satisfaction if the content is good | According to Salesforce, 80% of high-performing service organizations offer a self-service solution, compared to only 56% of low performers. Self-service separates the teams that scale from the teams that burn out trying to keep up. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. ## How customer self-service works Customer self-service has three parts: a knowledge base with answers, an AI agent or search tool that surfaces the right answer, and a routing system that escalates to a human when needed. ### 1. A knowledge base with clear, complete answers Your knowledge base holds the answers to every question customers ask regularly. Each article should answer one question fully, in plain language, with no jargon or assumptions about what the reader already knows. The article should start with the answer in the first sentence, then add detail, steps, or context below. This structure works for both AI agents and human readers. ### 2. An AI agent or search tool that finds the right answer fast Customers don't want to browse categories or guess which article might help. They want to ask a question and get an answer. An AI agent reads the question, searches the knowledge base, and returns the answer in conversational language. The best AI agents surface the answer in seconds, adapt to how people phrase questions, and guide customers to the next step if needed. They don't make customers hunt. ### 3. A routing system that escalates complex questions to a person Not every question can be resolved through self-service. Some are outside the knowledge base. Some require authority to make exceptions. Some customers are frustrated and need empathy, not automation. Your system should recognize when to stop and hand off the conversation to a human. A good rule: if the AI agent can't find a confident answer, or if the customer asks to speak to a person, route immediately. > "Self-service really is service if it is done right." (Gina Williams, Midland Radio, on the CX Heroes podcast) ## How to improve your self-service rate Your self-service rate is the percentage of customer inquiries resolved without an agent. A higher rate means your AI agent and knowledge base are doing the work. A lower rate means customers are bypassing self-service and going straight to your team. Follow these steps to raise your self-service rate: ### 1. Track what customers search for and where they get stuck Use analytics to see which questions customers ask, which articles they read, and where they abandon the help center to contact support. These gaps tell you what to fix first. If customers search for a topic and find nothing, write an article. If they read an article but still contact support, the article isn't clear or complete. Rewrite it. ### 2. Write knowledge articles that answer the question in the first sentence Each article should pass this test: a customer who reads only the first sentence should know the answer. Then add the steps, details, or context below. Use the actual question as the heading. If customers ask "How do I reset my password," the heading should be "How do I reset my password," not "Password Management." ### 3. Deploy an AI agent that surfaces answers in every channel Customers ask questions in different places: your help center, chat widget, contact form, email. Your AI agent should work in all of them, so customers get the same fast, accurate answer no matter where they start. Helpfeel includes an AI agent that works across your help center, search bar, chat, and contact forms. The platform also handles the content work, so your knowledge base stays current without pulling time from your team. ### 4. Close content gaps on a regular schedule Your knowledge base is never finished. Products change, policies update, and new questions appear. Set a schedule to review search data, flag articles that aren't working, and write new content for gaps. A managed knowledge base handles this for you. Helpfeel tracks what customers search for, identifies missing or unclear content, and ships updates on a regular cadence. For a detailed breakdown of how to track and interpret your self-service rate, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## What to measure Track these three metrics together: - **Self-service rate:** percentage of inquiries resolved without an agent - **Containment rate:** percentage of customers who find an answer and don't contact you again - **Customer satisfaction (CSAT):** whether customers are happy with the experience If your self-service rate climbs but satisfaction drops, you're not resolving inquiries. You're frustrating customers. If containment rate is low, customers are reading articles but still need help. That means your content isn't clear or complete. All three metrics should move in the right direction. If one slips, fix the content or the routing logic. ## When to route to a human Self-service handles routine questions with clear answers. It does not handle edge cases, complaints, complex technical problems, or conversations that require empathy. **Route to a human when:** - The customer asks to speak to a person - The question requires judgment or authority to make exceptions - The AI agent can't find a confident answer - The customer is frustrated or escalating - The issue is tied to emotion (cancellations, refunds, dissatisfaction) Your AI agent should recognize these signals and hand off immediately. A customer who wants to talk to a person should never have to fight the automation to get there. ## Frequently asked questions ### What is customer self-service? Customer self-service lets people find answers and resolve issues on their own using knowledge bases, AI agents, help centers, and automated tools, without waiting for a support agent. It works for routine questions that have clear, repeatable answers. ### What are the benefits of customer self-service? Self-service gives customers faster answers, works 24/7, and reduces repetitive support volume so teams can focus on complex requests. It lowers costs, scales without adding headcount, and improves satisfaction when done well. ### How do you improve self-service rate? Improve your self-service rate by writing clear knowledge articles that answer each question fully, deploying an AI agent that surfaces the right answer fast, and measuring what customers search for so you can close content gaps. ### When should you route customers to a human agent? Route to a human when the question requires judgment, empathy, or authority to make exceptions. Examples: complaints, cancellations, complex technical issues, or anything outside your knowledge base. Self-service handles routine volume, not edge cases. ## See how the done-for-you model works Customer self-service only works if the knowledge base behind it stays current and the AI agent surfaces the right answer every time. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Generative AI in Customer Service: Guide Source: https://en.helpfeel.com/ai-help-center/generative-ai-in-customer-service # Generative AI in Customer Service: Guide Generative AI in customer service creates contextually relevant responses in real time using large language models trained on text. It answers customer questions, summarizes conversations, personalizes recommendations, and drafts replies without needing pre-scripted answers. When grounded in accurate data, generative AI resolves inquiries faster than rule-based systems. When ungrounded, it risks hallucination, producing confident but incorrect answers that damage trust. This guide covers what generative AI is, how it's used in customer service, the risks to watch for, and how a managed knowledge base keeps it accurate. For the broader context on AI in support, start with the [AI help center guide](/ai-help-center), which covers the full stack of AI tools, from search to agents to analytics. ## What is generative AI in customer service? Generative AI in customer service is technology that uses large language models to create responses, recommendations, and summaries based on the context of a customer inquiry. Unlike rule-based systems that select from pre-written scripts, generative AI composes original answers in real time. The technology reads a customer's question, pulls relevant context from your knowledge base or inquiry history, and generates a conversational response. According to Salesforce, generative AI creates "customized responses, recommendations, and solutions in real-time," making interactions feel more natural than older chatbot scripts. Generative AI works well for repetitive inquiries that need slight variations. A customer asking "Where is my order?" gets a response tailored to their specific order status, not a generic tracking link. A customer asking "How do I reset my password?" gets step-by-step instructions written for their account type. The power is in the personalization and speed. The risk is in accuracy. ## How is generative AI used in customer service? Generative AI handles tasks that traditionally required an agent to read, interpret, and respond. The most common use cases fall into four categories: answering questions, assisting agents, summarizing interactions, and closing content gaps. ### 1. Answering customer questions in self-service Generative AI agents respond to customer inquiries in chat, email, and help centers. The AI reads the question, searches a knowledge base for the answer, and returns a conversational response. When the knowledge base is accurate and complete, this resolves inquiries faster than routing to a person. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The AI agent surfaces the right answer from the knowledge base and delivers it in context. ### 2. Assisting agents with drafts and summaries Generative AI can draft replies for agents, summarize long conversations, suggest next steps, and translate inquiries in real time. According to Gartner, AI agents "collaborate with other AI agents and human agents as needed to orchestrate the steps to resolve a customer issue." This cuts the time agents spend typing, searching, and documenting. An agent handling a billing dispute gets a summary of the customer's recent interactions and a draft reply pulling the relevant policy. The agent reviews, edits, and sends. ### 3. Personalizing recommendations and responses Generative AI tailors responses based on customer history, preferences, and behavior. A customer who frequently orders a specific product sees recommendations for complementary items. A customer contacting support after a service outage gets a proactive explanation and discount offer. According to IBM, generative AI "allows companies to move beyond simple answers and deliver proactive suggestions, tailored recommendations and even solve customer issues before they happen." ### 4. Identifying trends and closing content gaps Generative AI analyzes inquiry patterns to surface common questions that lack good answers. It flags content gaps, drafts new articles to fill them, and suggests updates to outdated material. This turns customer inquiries into feedback that improves the knowledge base continuously. Deloitte notes that AI can "handle more tedious documentation and reporting tasks, freeing contact center agents to address more complex situations that require a human touch." ## What are the risks of generative AI in customer service? The biggest risk is hallucination, when the AI generates confident but incorrect information. In customer service, hallucination means the AI invents product details, fabricates policies, or gives wrong answers that look convincing. According to a 2024 McKinsey survey, 44% of organizations using generative AI reported at least one negative consequence, with inaccuracy cited as the most common problem. In customer service, inaccurate answers create escalations, erode trust, and force agents to clean up mistakes the AI made. Hallucination happens when the AI lacks access to accurate source data or when the knowledge base contains outdated, incomplete, or conflicting information. The model fills the gap by predicting what sounds plausible, not what's correct. ### Other risks to watch | Risk | What it means | How to mitigate | | ------------------ | ----------------------------------------------------------------------------- | --------------------------------------------------------------------- | | Inaccurate answers | AI invents information or misinterprets the knowledge base | Ground the AI in a managed knowledge base with regular content audits | | Tone drift | AI response feels robotic, overly formal, or inconsistent with your brand | Set tone guidelines and review AI-generated responses before launch | | Over-automation | AI tries to handle complex or emotional inquiries it should route to a person | Define clear guardrails for what the AI should and shouldn't answer | | Bias in responses | AI reproduces biases from training data or poorly written content | Monitor outputs for bias and review content for inclusive language | The common thread is lack of oversight. Generative AI needs guardrails, grounding, and measurement to work well in production. ## How do you prevent AI hallucination in customer service? Prevent hallucination by grounding the AI in a managed knowledge base that's accurate, complete, and regularly updated. The AI should retrieve answers from verified content, not generate them from scratch. **Four strategies to reduce hallucination:** 1. **Ground the AI in a single source of truth.** Every answer the AI gives should trace back to a specific article in your knowledge base. If the knowledge base has no answer, the AI should say so and route the inquiry to a person. 2. **Set clear guardrails for what the AI can answer.** Define which question types the AI handles and which require a human. Complex, emotional, or high-stakes inquiries should route immediately. 3. **Monitor outputs for accuracy.** Review a sample of AI-generated responses regularly. Flag incorrect answers and update the knowledge base to close the gap. 4. **Use a managed knowledge base that stays current.** If your knowledge base falls out of date, the AI will give outdated answers or hallucinate to fill the gap. A managed knowledge base has someone responsible for keeping content accurate. Helpfeel runs this work for you. We maintain the knowledge base, monitor what the AI retrieves, flag content gaps, and update articles on a regular cadence so the AI always pulls from accurate, current source material. For more on grounding AI in a knowledge base, see the [self-service knowledge base guide](/ai-help-center/self-service-knowledge-base). ## How generative AI compares to conversational AI Generative AI and conversational AI overlap but serve different purposes. Conversational AI is the broader category that includes any system designed to hold a conversation with a customer: rule-based chatbots, voice assistants, and generative AI agents. Generative AI is a subset of conversational AI that uses large language models to create responses dynamically instead of selecting from pre-written scripts. It's more flexible and feels more natural, but it requires grounding in accurate data to avoid hallucination. For a deeper look at conversational systems, read the [conversational AI for customer service guide](/ai-help-center/conversational-ai-for-customer-service). ## How to know if generative AI is working Generative AI is working when it resolves inquiries accurately, keeps customer satisfaction steady or higher, and reduces repetitive volume for your team. You should see self-service rate climb, ticket volume drop, and containment rate stay high. Generative AI is failing when it creates follow-up inquiries because the answers were wrong, unclear, or incomplete. You'll see ticket volume drop but satisfaction scores fall, or customers switching to phone and chat to get around the AI. Track three metrics together: - **Self-service rate:** percentage of inquiries resolved without an agent - **Containment rate:** percentage of customers who find an answer and don't contact you again - **Customer satisfaction:** whether customers are happy with the resolution If all three stay healthy, the AI is working. If satisfaction drops while self-service climbs, the AI is answering questions but not resolving them. For a detailed breakdown of how to measure self-service success, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## When to use generative AI in customer service Use generative AI when you have high inquiry volume, a strong knowledge base, and the capacity to monitor outputs for accuracy. Generative AI shines in scenarios where customers ask similar questions in different ways, where personalization matters, and where speed is critical. **Good fit for generative AI:** - High-volume, low-complexity inquiries - Questions that need slight personalization (order status, account details) - Multilingual support that requires real-time translation - Agent assistance (drafting replies, summarizing conversations) **Not a good fit yet:** - Complex troubleshooting that requires judgment - High-stakes decisions (refunds, cancellations, exceptions) - Emotional or sensitive conversations - Industries where incorrect answers carry legal or safety risk If your knowledge base is incomplete or outdated, fix that first. Generative AI amplifies the quality of your content. If the content's bad, the AI will be bad. For more on AI help center strategy, read the [AI customer support agent guide](/customer-support/ai-customer-support-agent) and the [AI help center overview](/ai-help-center). ## Frequently asked questions ### What is generative AI in customer service? Generative AI in customer service creates contextually relevant, real-time responses to customer questions using large language models trained on text. It answers questions, summarizes conversations, drafts replies, and personalizes recommendations without needing pre-scripted responses. ### What are common use cases for generative AI in customer service? Common use cases include answering customer questions in chat, summarizing conversations for agents, drafting follow-up emails, personalizing product recommendations, translating inquiries in real time, and identifying trends across inquiry data to close content gaps. ### What is AI hallucination in customer service? AI hallucination is when a generative AI model produces confident but incorrect information. In customer service, this means the AI might invent product details, fabricate policies, or give wrong answers that look convincing, eroding trust and creating escalations. ### How do you prevent AI hallucination in customer service? Prevent hallucination by grounding the AI in a managed knowledge base, setting clear guardrails for what the AI can answer, monitoring outputs for accuracy, and routing ambiguous or high-stakes questions to a human agent immediately. ## See how the managed model works Generative AI only works well when it's grounded in accurate, current content. Helpfeel is a done-for-you platform: we manage the knowledge base, monitor what the AI retrieves, flag gaps, and ship updates so the AI always pulls from verified source material. [See how the done-for-you model works](/product). --- ## Self-Service Knowledge Base: The Complete Guide Source: https://en.helpfeel.com/ai-help-center/self-service-knowledge-base # Self-Service Knowledge Base: The Complete Guide A self-service knowledge base is a searchable collection of answers that lets customers resolve their own questions without opening a ticket or calling support. It includes product documentation, troubleshooting guides, policy explanations, and how-to instructions organized so customers can find what they need quickly. When it's built right, a knowledge base resolves most inquiries before they reach your team, reduces ticket volume, and keeps satisfaction high. This guide covers what a self-service knowledge base is, what makes one effective, how to build it, and how to measure whether it's working. For a broader view of how knowledge bases fit into an AI help center strategy, start with the [AI help center guide](/ai-help-center). ## What is a self-service knowledge base? A self-service knowledge base is a centralized collection of information that customers use to find answers on their own. According to Atlassian, it's "a centralized, organized collection of information about a product, service, department, or topic" built to reduce service requests. Unlike a static FAQ page, a knowledge base is searchable, interconnected, and built to handle complex questions. It includes articles on common issues, step-by-step instructions, policy details, and product explanations. Customers can search for a specific question, browse by topic, or follow links between related articles. The goal is to make finding the answer faster and easier than opening a ticket. When customers can resolve their own questions, they get help immediately and your support team has time for the work that needs a human. ## What are the benefits of a self-service knowledge base? A self-service knowledge base reduces support volume, speeds up resolution, and keeps customers in control of their own experience. The benefits show up in both customer satisfaction and team capacity. **For customers:** - Instant answers without waiting for a reply - Access to help 24/7, across time zones - The ability to solve problems without explaining context to an agent - Consistent answers across every channel **For support teams:** - Fewer repetitive tickets - More time for complex or high-value inquiries - Lower cost per contact as self-service absorbs volume - A single source of truth that keeps answers consistent According to Document360, clients see an average 45% increase in customer satisfaction after adopting self-service. Desku research shows that customers who use self-service options are 73% more likely to stay loyal. > "Self-service really is service if it is done right." (Gina Williams, Midland Radio, on the CX Heroes podcast) The key phrase is "if it is done right." A knowledge base that answers questions resolves inquiries. A knowledge base that's incomplete, hard to search, or filled with outdated content creates frustration and drives customers to other channels. ## What should a self-service knowledge base include? A strong knowledge base includes the questions customers ask most often, answered clearly and completely. Start with the inquiries that show up in volume and build from there. **Core content to include:** - Product or service documentation - Troubleshooting steps for common issues - How-to guides for key features or tasks - Policy explanations (returns, refunds, billing, privacy) - Account management help (password resets, login issues, profile updates) - Definitions or terminology customers need to understand your product **Content to leave out:** - Internal process documentation not relevant to customers - Marketing copy that doesn't answer a question - Duplicate articles that create confusion - Outdated content that no longer reflects your product or policies Each article should answer one question completely. Use the question itself as the heading, answer it in the first sentence, then add detail below. Avoid jargon, assume the reader knows nothing about your product, and include screenshots or examples where they help. For detailed guidance on structuring articles, see the [customer self-service guide](/ai-help-center/customer-self-service). ## How do you build a self-service knowledge base in five steps? Building a self-service knowledge base follows a clear sequence. Each step builds on the one before it. ### 1. Map the questions your customers ask most often Pull three to six months of support ticket data from your helpdesk or contact center platform. Group inquiries by question type, not by tag or department. You want to see the actual questions customers ask. You'll find that a small number of question types account for most of your volume. Those are your starting point. Write articles for the top 20 to 30 questions first, then expand from there. ### 2. Write clear, complete answers for each question For each high-volume question, write one article that answers it fully. Start with the question as the heading. Answer it in the first sentence. Add details, steps, examples, or screenshots below. Each article should stand on its own. No references to "see above," no assumptions about what the reader already knows, no jargon without definitions. If a customer lands on this article from search, they should be able to resolve their question without reading anything else. ### 3. Organize articles so customers can navigate or search Structure your knowledge base so customers can either search for a specific question or browse by topic. Group related articles into categories. Use consistent naming. Link related articles to each other so customers can move from one question to the next. Make sure your search works. Test it with the exact phrases customers use in support tickets. If the right article doesn't surface in the first three results, either improve the article title and keywords or write a new article that matches the search query better. ### 4. Deploy the knowledge base where customers ask questions Put your knowledge base in every place customers look for help. That includes your website, your product UI, your chat widget, your contact forms, and your email signatures. The easier it is to find, the more customers will use it. If you're using an AI agent or chatbot, connect it to your knowledge base so it can retrieve and serve the right article automatically. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. ### 5. Measure what's working and close the gaps Track three metrics together: self-service rate (the percentage of inquiries resolved without an agent), containment rate (the percentage of customers who find an answer and don't contact you again), and customer satisfaction. Pull the search queries that returned no useful result. Those are your content gaps. Write articles for them and measure whether the new content reduces ticket volume for that question type. For a detailed breakdown of how to track and interpret self-service metrics, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## What makes a self-service knowledge base effective? | Feature | Why it matters | | --------------------------------------------- | -------------------------------------------------------------------------------------------------- | | Clear, plain-language answers | Customers understand the answer on first read and don't need to contact support for clarity. | | Strong search that surfaces the right article | The right article shows up in the first few results, so customers don't give up and open a ticket. | | Articles that answer one question completely | Customers resolve their inquiry without needing to read three other articles. | | Regular updates to keep content current | Outdated articles create confusion and drive customers to contact support instead. | | Mobile-friendly design | Customers can find answers on any device, wherever they are. | An effective knowledge base resolves inquiries. An ineffective one makes customers work harder to find help, which sends them to other channels instead. ## The role of a managed knowledge base A knowledge base is only as good as the content inside it. If articles go stale, if new product features go undocumented, or if customers search for something and find nothing, the knowledge base stops working. This is why most knowledge base projects stall after launch. Writing the initial articles is the easy part. Keeping them current, closing gaps as they appear, and measuring which content isn't resolving inquiries requires ongoing attention. A managed knowledge base solves this. Someone owns the content, reviews it on a schedule, watches what customers search for, and ships updates as your product and policies change. The knowledge base stays current without pulling time from your support team. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. The knowledge base stays current without your team having to manage it. ## How a self-service knowledge base fits into your support strategy A knowledge base is the foundation of a sustainable support operation. It handles the repetitive volume so your team focuses on the work that needs a human. Most inquiries should resolve in the knowledge base or through an AI agent that retrieves the right article. The ones that reach a person should be the ones that need judgment, empathy, or a decision only a human can make. This structure keeps cost per contact stable even as total volume grows, because the repetitive volume goes straight to self-service. For detailed strategies on automating and scaling support, see the [customer support guide](/customer-support) and the [guide to reducing support tickets](/customer-support/how-to-reduce-support-tickets). ## Frequently asked questions ### What is a self-service knowledge base? A self-service knowledge base is a searchable collection of articles, guides, and answers that customers use to resolve their own questions without contacting support. It includes product help, troubleshooting steps, policy explanations, and how-to instructions organized so customers can find what they need quickly. ### What makes a good self-service knowledge base? A good knowledge base answers the questions customers actually ask, uses clear language, surfaces the right article in search results, and resolves the inquiry without needing a follow-up. It stays current, works across devices, and makes self-service faster than contacting support. ### How do you build a self-service knowledge base? Build a self-service knowledge base by mapping your high-volume support questions, writing clear answers for each one, organizing articles so customers can navigate or search easily, and measuring which questions still generate tickets so you can close the gaps. ### What percentage of customers use self-service? According to research compiled by Document360, 91% of customers would use a knowledge base if it met their needs. Desku reports that customers who use self-service options are 73% more likely to stay loyal. A well-built knowledge base becomes the first place customers go for help. ## See how the managed model works A self-service knowledge base only works if the content inside it stays current and complete. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## What Is an AI Help Center? Source: https://en.helpfeel.com/ai-help-center/what-is-an-ai-help-center # What Is an AI Help Center? An AI help center is a self-service support platform that uses artificial intelligence to search your knowledge base, retrieve relevant articles, and answer customer questions automatically, without routing to an agent. It combines a searchable knowledge base with conversational AI so customers get the right answer in seconds, not after reading through five articles. This page covers what an AI help center is, how it works, and how it differs from traditional help centers. For a full view of how AI help centers fit into a broader support strategy, see the [AI help center guide](/ai-help-center). ## What is an AI help center? An AI help center is a customer support platform that uses AI to understand what a customer is asking, search a knowledge base for the answer, and serve that answer conversationally. Instead of making customers search for articles, read through pages, and piece together an answer, the AI does the work for them. The category sits between traditional knowledge bases, where customers read articles manually, and fully autonomous AI agents, where bots handle entire conversations and resolve tickets without human intervention. Most AI help centers combine both layers: the AI serves answers from the knowledge base, and when it cannot resolve the inquiry, it routes the customer to an agent. According to Klaviyo, an AI helpdesk is a customer support system that automates ticket management and other tasks so teams can provide personalized customer service. HappySupport defines AI help center software as a platform that uses artificial intelligence to surface relevant articles, generate draft answers, route tickets, and resolve customer queries without a human agent. The key difference is that a traditional help center waits for the customer to find the right article. An AI help center brings the answer to the customer. ## How does an AI help center work? An AI help center has three parts: a knowledge base with structured, searchable articles; an AI layer that reads customer questions and retrieves the right content; and a routing system that hands off unresolved inquiries to a person. Here is the full sequence: 1. **A customer asks a question** in the help center, chat widget, contact form, or email. 2. **The AI reads the question** and searches the knowledge base for content that answers it. 3. **The AI serves the answer** in conversational language, often with a link to the full article. 4. **If the AI cannot find a good answer**, or if the customer asks a follow-up question outside the knowledge base, the inquiry routes to an agent. Every question the AI resolves is one fewer ticket your team handles. Every question the AI cannot answer becomes a signal that the knowledge base is missing content, so you can fill the gap. ## What is the difference between an AI help center and a traditional help center? Traditional help centers are static repositories. Customers search for a keyword, browse a list of articles, click one, read it, and decide whether it answered their question. If it did not, they search again or open a ticket. AI help centers are interactive. Customers ask a question in plain language. The AI searches the knowledge base, pulls the relevant content, and returns a direct answer. If the customer needs more detail, the AI can link to the full article. If the answer is not in the knowledge base, the AI routes the inquiry to an agent immediately. | Traditional help center | AI help center | | ------------------------------------------ | ----------------------------------------------------- | | Customer searches and reads articles | AI searches and delivers the answer | | Requires keyword matching and browsing | Understands questions in natural language | | Customer assembles the answer from content | AI assembles the answer from the knowledge base | | No routing logic if the answer is missing | Routes unresolved inquiries to an agent automatically | | Scales by adding more articles | Scales by improving AI accuracy and content coverage | The difference shows up in resolution speed and self-service rate. Traditional help centers answer questions when customers find the right article. AI help centers answer questions when customers ask. ## What are the benefits of an AI help center? AI help centers resolve routine inquiries automatically, reduce ticket volume, and give support teams capacity to focus on complex requests. The benefits break into three categories: speed for customers, capacity for teams, and scalability for the business. **For customers:** - Instant answers, no queue time. - 24/7 availability without staffing overnight shifts. - Conversational interaction instead of reading through long articles. **For support teams:** - Fewer repetitive tickets, more time for work that needs judgment. - Automatic routing for inquiries the AI cannot resolve. - Real-time visibility into which questions customers ask most often, so you can improve content. **For the business:** - Support costs stay stable as inquiry volume grows. - You can serve more customers without hiring linearly. - Higher self-service rates mean fewer escalations and faster resolution. According to Master of Code, key advantages of AI in customer service include improved customer service (69%), decreased wait times (55%), and streamlined workflows (54%). Other benefits are enhanced customer satisfaction (48%) and better use of data and analytics (41%). McKinsey reports that AI deployments reduce total interactions by 40 to 50%, and AI is projected to cut $80 billion in contact center labor costs by 2026. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. The teams that hit those numbers measure self-service rate, containment rate, and customer satisfaction together, and adjust the content when any of those metrics slip. ## What types of questions can an AI help center answer? An AI help center answers the questions that show up in volume and follow a pattern: order status, password resets, return steps, billing lookups, policy questions, hours, locations, and feature explanations. These are the inquiries your team answers the same way every time. **Good fit for AI help centers:** - Where is my order? - How do I reset my password? - What is your return policy? - How do I upgrade my plan? - Do you ship to Canada? - How do I use this feature? **Not a good fit yet:** - Complaints or escalations. - Requests that require judgment or exception handling. - Conversations tied to emotion (cancellations, dissatisfaction, loss). - Anything your team answers differently based on context. The AI help center should carry the repetitive load so your team has capacity for the work that matters. If a question type requires your agents to think before answering, it is not ready to automate. See the [customer support automation guide](/customer-support/how-to-automate-customer-support) for a full playbook on what to automate first. ## How do you measure whether an AI help center is working? Measure three things: self-service rate, containment rate, and customer satisfaction. Together, they show whether the AI help center is resolving inquiries or just blocking customers. - **Self-service rate** is the percentage of inquiries resolved without an agent. This tells you how much volume the AI is carrying. - **Containment rate** is the percentage of customers who find an answer and do not contact you again. This tells you whether the answers are actually working. - **Customer satisfaction** tells you whether customers are happy with the resolution. If self-service rate climbs but satisfaction drops, the AI is answering questions but not resolving them. That means the knowledge base has gaps or the answers are unclear. Go back and improve the content. For a detailed breakdown of how to track and interpret these metrics, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## What is the role of the knowledge base in an AI help center? The knowledge base is the engine. The AI layer is only as good as the content it pulls from. If your knowledge base is incomplete, outdated, or unclear, the AI will return incomplete, outdated, or unclear answers. This is why most AI help center projects stall after launch. Teams deploy the AI, see initial results, and then watch performance degrade as the knowledge base falls out of date. The AI cannot fix bad content. It can only surface what exists. A strong AI help center includes a managed knowledge base where someone owns the content, reviews it on a schedule, and closes gaps as they appear. The AI keeps working because the content keeps improving. For a deeper look at how self-service knowledge bases work, see the [self-service knowledge base guide](/ai-help-center/self-service-knowledge-base). ## How does an AI help center reduce support costs? An AI help center reduces support costs by resolving the high-volume, low-complexity inquiries that take up most of your team's time. Every question the AI answers is one fewer ticket an agent handles, which means you can serve more customers with the same team size. The cost structure shifts from variable (more volume = more agents) to fixed (more volume = same infrastructure). Traditional support scales linearly. AI help centers scale logarithmically. As your customer base grows, the AI handles the repetitive load, and your team focuses on the complex inquiries that need judgment. This does not mean you eliminate headcount. It means the next hire you were going to make to keep up with volume is a hire you do not need to make. The team you have can serve more customers, and the capacity they gain goes toward improving the experience instead of just keeping up. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. We handle the content work, the AI layer, and the measurement tools in one package, so your team can focus on the conversations that need a human. ## Frequently asked questions ### What is an AI help center? An AI help center is a self-service support platform that uses artificial intelligence to search your knowledge base, retrieve relevant articles, and answer customer questions automatically, without routing to an agent. ### How does an AI help center differ from a traditional help center? Traditional help centers let customers search and read articles manually. AI help centers use conversational AI to understand the customer's question, search the knowledge base, and serve the right answer directly, so customers get help faster without reading multiple articles. ### What are the benefits of an AI help center? AI help centers answer questions instantly, resolve routine inquiries automatically, reduce support ticket volume, and scale without adding headcount. They keep customers from waiting in queue and free your team to focus on complex requests. ### Can an AI help center replace my support team? No. An AI help center handles repetitive inquiries so your team focuses on work that needs judgment, empathy, or problem-solving. It is the next hire you will not need to make, not a replacement for your current team. ## See how the done-for-you model works An AI help center only works if the knowledge base behind it stays current. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Customer Service Software: The Complete Guide Source: https://en.helpfeel.com/customer-service-software # Customer Service Software: The Complete Guide Customer service software is a platform that helps teams manage customer interactions across email, chat, phone, and social media. It tracks inquiries, routes them to the right person or AI agent, and stores answers in a knowledge base so customers can resolve their own questions. Strong service software reduces repetitive volume, keeps response times low, and scales as your customer base grows. This guide covers what customer service software is, the difference between customer service and customer support, what features matter most, and how to choose the right platform for your team. Each section links to a full guide if you want to go deeper. ## What is customer service software? Customer service software is a platform that organizes, tracks, and resolves customer inquiries across multiple channels. It includes ticketing systems that route questions to the right agent, knowledge bases that customers can search before opening a ticket, and AI agents that answer repetitive questions automatically. The software serves two audiences: your customers, who use it to find answers and ask questions, and your team, who use it to manage volume and keep response times low. The best platforms reduce the number of inquiries that need a human while keeping satisfaction high. According to the Global Market Report, the customer service software market is expected to grow from $10.95 billion in 2025 to $13.06 billion in 2026, driven by rising demand for automation and self-service tools. ## What is the difference between customer service and customer support? Customer service and customer support are related but not the same thing. Customer service covers the full customer relationship. It includes billing questions, returns, refunds, order tracking, policy clarifications, and satisfaction work. If a customer is asking about when their order will arrive or how to return a product, that's a service question. Customer support is narrower. It focuses on technical help, product troubleshooting, and getting someone unstuck with a feature. If a customer can't log in or needs help using a feature, that's a support question. In practice, service is transactional and support is technical. Some companies call everything customer service. Others separate the two functions and route inquiries accordingly. Either approach works as long as the team knows which channel handles what. For a deeper look at the technical side, see the [customer support guide](/customer-support). ## What are the most important features in customer service software? The features that matter most are the ones that help you serve more customers without adding headcount. ### Multi-channel support Your customers ask questions across email, chat, phone, social media, and contact forms. Your service software should pull all of those channels into one place so your team can respond from a single interface. If an agent has to switch between five tools to see all the inquiries, response times will climb. ### Ticket management and routing Every inquiry should turn into a ticket that gets tracked, assigned, and resolved. The system should route tickets based on question type, priority, or availability. A billing question goes to billing. An urgent escalation goes to a senior agent. A password reset goes to an AI agent that can resolve it instantly. ### Knowledge base and self-service tools Most inquiries are repetitive. A searchable knowledge base gives customers a way to resolve their own questions before they contact you. The knowledge base should surface in your help center, in chat, and in search results. If customers can't find the answer, they'll open a ticket, which costs you time and costs them speed. ### AI and automation AI agents can resolve routine inquiries without routing to a person. They search your knowledge base, return the answer in conversational language, and route complex questions to a human. Automation handles the repetitive volume so your team has time for work that needs judgment. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. ### Analytics and reporting You need to know whether the system is working. Track self-service rate, first response time, resolution time, ticket volume, and customer satisfaction. If self-service is climbing but satisfaction is falling, customers are getting stuck, not helped. The data tells you where to improve. ## Types of customer service software Customer service software comes in different configurations depending on what you're trying to solve. | Type | Best for | Read | | ----------------------------- | ------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------- | | Internal help desk | IT teams supporting employees | [Internal help desk software](/customer-service-software/internal-help-desk-software) | | Workflow and case management | Teams that handle multi-step, complex requests | [Customer service workflow software](/customer-service-software/customer-service-workflow-software) | | Automation platforms | Teams that want to reduce repetitive volume with AI | [Customer service automation software](/customer-service-software/customer-service-automation-software) | | Cloud-based service platforms | Teams that need scalable, multi-channel support in the cloud | [Cloud-based customer service software](/customer-service-software/cloud-based-customer-service-software) | Each type solves a different problem. If your team spends most of its time answering the same questions, you need automation. If you handle long, complex cases that involve multiple departments, you need workflow software. If you support internal employees instead of external customers, you need an internal help desk. ## How do you choose customer service software? Choose customer service software by mapping your current volume, identifying the question types that show up most often, and picking a platform that can automate those questions while keeping the rest routed to the right person. Start with these questions: 1. **What channels do your customers use to contact you?** If they prefer chat, your platform needs strong chat support. If most inquiries come through email, you need a ticketing system that handles high email volume. 2. **How much of your volume is repetitive?** If your team answers the same five questions all day, prioritize AI and self-service. If every inquiry is unique, prioritize routing and collaboration tools. 3. **How many agents will use the system?** Pricing scales with seats. Make sure the platform fits your budget at your current team size and at the size you expect to reach in the next year. 4. **Do you need to integrate with other tools?** Your service software should connect to your CRM, e-commerce platform, and analytics tools. If it doesn't, your team will spend time copying data between systems. For teams that want to reduce repetitive volume without building and maintaining a knowledge base themselves, Helpfeel handles the content work, the AI layer, and the reporting in one platform. You get the system without the setup cost. ## Customer service software and AI AI is changing how service teams operate. Instead of routing every inquiry to an agent, AI agents resolve the routine questions and route complex questions to a human. This keeps response times low even as volume grows, because the repetitive work happens instantly. The shift is toward self-service as the front line and human agents as the second line. Customers get faster answers for simple questions. Agents spend their time on work that requires empathy, judgment, or problem-solving. For a detailed look at how AI fits into service workflows, see the [AI help center guide](/ai-help-center) and the [customer support automation guide](/customer-support/how-to-automate-customer-support). ## How customer service software reduces costs Customer service software reduces costs by raising self-service rate. Every inquiry that resolves in the knowledge base or through an AI agent is one fewer ticket your team handles. As self-service rate climbs, you can serve more customers without adding headcount. The cost per contact drops because the software handles the repetitive volume automatically. Your team's capacity stays focused on the inquiries that need a human. Over time, this structure keeps operating costs stable even as total volume grows. Helpfeel customers see up to 70% ticket reduction, which translates to fewer agents needed to serve the same volume or the same team serving far more customers. That's the economic case for automation: it lets you scale service without scaling costs linearly. ## Frequently asked questions ### What is customer service software? Customer service software is a platform that helps teams manage customer interactions across channels like email, chat, phone, and social media. It includes ticketing systems, knowledge bases, AI agents, and automation tools that help you track inquiries, deliver answers, and resolve issues. ### What is the difference between customer service and customer support? Customer service covers the full customer relationship, including billing, returns, refunds, order tracking, and general satisfaction. Customer support is narrower and focuses on technical help, product questions, and troubleshooting. Service is transactional, support is technical. ### What are the most important features in customer service software? Multi-channel support (email, chat, phone, social), ticket management and routing, a searchable knowledge base, AI or automation to handle repetitive inquiries, reporting and analytics, and integrations with your CRM or e-commerce platform. ### How much does customer service software cost? Pricing varies by feature set and team size. Entry-level tools start around $15 to $30 per agent per month. Enterprise platforms with AI, workforce management, and advanced automation typically range from $80 to $150 or more per agent per month. ## Go deeper Customer service software is a system you build once and keep improving. Helpfeel runs that system for you: a managed, AI-ready knowledge base plus an AI agent that resolves inquiries before they reach your team. [See how the done-for-you model works](/product), or read the guides below to understand the strategy before you build it. **Explore related topics:** → Read the full guide: [Internal help desk software](/customer-service-software/internal-help-desk-software) → Read the full guide: [Customer service workflow software](/customer-service-software/customer-service-workflow-software) → Read the full guide: [Customer service automation software](/customer-service-software/customer-service-automation-software) → Read the full guide: [Cloud-based customer service software](/customer-service-software/cloud-based-customer-service-software) → See also: [Customer support guide](/customer-support) → See also: [AI help center guide](/ai-help-center) --- ## Cloud-Based Customer Service Software Source: https://en.helpfeel.com/customer-service-software/cloud-based-customer-service-software # Cloud-Based Customer Service Software Cloud-based customer service software runs entirely online, with no servers to buy, no installations to manage, and no IT team required to keep it running. Your team accesses the platform through a browser or app, and the vendor handles hosting, updates, security, and scaling. This guide covers what cloud-based customer service software is, how it differs from on-premise systems, and when it makes sense to choose cloud. For a complete view of the software category and related options, see the [customer service software guide](/customer-service-software). To understand how cloud platforms fit into automation strategy, read [customer service automation software](/customer-service-software/customer-service-automation-software). ## What is cloud-based customer service software? Cloud-based customer service software is a platform that runs on the vendor's servers and delivers service to your team over the internet. You log in through a browser or mobile app. The vendor manages the infrastructure, applies updates automatically, and scales capacity as your team or inquiry volume grows. Also called SaaS (Software as a Service), cloud platforms handle customer inquiries across channels like email, chat, phone, and self-service help centers. They include ticketing systems, knowledge bases, AI agents, routing tools, and reporting, all accessible without installing anything locally. The defining feature is that you don't run the software yourself. The vendor hosts it, secures it, backs it up, and keeps it online. You access it and use it. ## What is the difference between cloud-based and on-premise customer service software? The difference is where the software runs and who manages it. Cloud-based software runs on the vendor's servers. You access it over the internet. The vendor handles infrastructure, updates, security, and scaling. You pay a monthly subscription per user or per feature. On-premise software runs on your own servers inside your building or data center. Your IT team installs it, secures it, backs it up, and applies updates. You buy licenses upfront and own the software, but you also own the responsibility for keeping it running. Cloud platforms cost less to start, deploy faster, and scale automatically. On-premise systems give you full control over data and infrastructure but require capital spending and ongoing IT work. Most organizations choose cloud for customer service because the upfront cost is lower, deployment is faster, and the vendor keeps the platform current without pulling your team's time. On-premise makes sense if you have strict data residency rules, legacy integrations that require local hosting, or an IT team that prefers to manage infrastructure directly. ## Why choose cloud-based customer service software? Choose cloud-based customer service software when you need to deploy quickly, scale without capital spending, and avoid the overhead of managing infrastructure. ### Lower upfront cost Cloud platforms charge a monthly subscription instead of a large license purchase. You don't buy servers, storage, or data center space. You don't hire IT staff to install and maintain the system. You start with what you need and add capacity as you grow. ### Faster deployment Cloud platforms deploy in days or weeks instead of months. You sign up, configure the system, import your data, and start serving customers. There's no hardware procurement, no installation, and no infrastructure setup. Your team can be up and running before an on-premise system would finish procurement. ### Automatic scaling Cloud platforms scale automatically as inquiry volume or team size grows. The vendor adds capacity on the backend. You don't provision servers, expand storage, or worry about peak load. The platform handles it. ### Updates without downtime Cloud vendors push updates and new features automatically. You wake up to improvements without scheduling maintenance windows or running upgrade scripts. Security patches, bug fixes, and new capabilities arrive continuously. ### Access from anywhere Cloud platforms work from any device with internet access. Your team can serve customers from the office, from home, or from anywhere else. This flexibility matters when you're hiring remote agents or covering multiple time zones. ## What are the trade-offs of cloud-based customer service software? Cloud platforms come with trade-offs. You give up some control in exchange for convenience and speed. You depend on the vendor's uptime. If their service goes down, your team can't access the platform. Most cloud vendors maintain 99.9% uptime or higher, but the risk exists. You store customer data on the vendor's servers. Most cloud platforms encrypt data and maintain compliance certifications, but some industries or regions require data to stay inside specific geographic boundaries. Check your vendor's data residency and compliance posture before signing. You're locked into the vendor's update schedule. When they push changes, you use them. You can't delay an update or roll back a feature. This is usually fine, but it means you don't control the pace of change. For most teams, these trade-offs are worth it. The cost savings, deployment speed, and reduced operational burden outweigh the loss of direct control. ## How do you choose the right cloud-based customer service software? Choose cloud-based customer service software by matching the platform's capabilities to your inquiry volume, channel mix, and automation goals. Start with these questions: 1. **What channels do you serve customers on?** Make sure the platform supports email, chat, phone, and self-service. If you serve customers on social media or messaging apps, check that the platform integrates those channels too. 2. **How much automation do you need?** If you handle repetitive volume, choose a platform with a strong knowledge base and an AI agent that can resolve inquiries before they reach your team. Read [customer service automation software](/customer-service-software/customer-service-automation-software) for a breakdown of what to look for. 3. **How fast do you need to deploy?** Cloud platforms deploy faster than on-premise systems, but some are easier to configure than others. If you need to go live in weeks, prioritize platforms with guided setup and managed onboarding. 4. **What reporting do you need?** Make sure the platform tracks self-service rate, resolution time, and customer satisfaction. You can't improve what you don't measure. 5. **What does the vendor manage for you?** Some cloud platforms are self-service. Others offer managed services where the vendor builds and maintains your knowledge base. If you don't have time to write and update help content, choose a platform that does it for you. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. We handle the content work, the AI layer, and the deployment. You get a cloud platform that's ready to serve customers without pulling time from your team. ## How do cloud-based customer service platforms handle AI and self-service? Cloud platforms are built to support AI agents and self-service tools because they can update the underlying models and features continuously without requiring you to install anything. The platform's AI agent searches your knowledge base, retrieves answers, and serves them to customers in real time. When the vendor improves the AI model or adds a new capability, you get the update automatically. You don't upgrade the software or redeploy anything. This is why most AI-powered customer service tools are cloud-based. The models improve weekly, and cloud platforms can push those improvements to every customer at once. On-premise systems can't keep pace without constant manual updates. For more on how AI agents fit into customer service, see [AI help center](/ai-help-center). ## Frequently asked questions ### What is cloud-based customer service software? Cloud-based customer service software runs entirely online. Your team accesses it through a browser or app, with no servers to install or maintain. The vendor hosts everything, handles updates, and scales capacity as your team grows. ### What is the difference between cloud-based and on-premise customer service software? Cloud software runs on the vendor's servers and you access it online. On-premise software runs on your own servers inside your building. Cloud systems cost less upfront, deploy faster, and scale automatically. On-premise systems give you full control but require IT staff and capital spending. ### What are the benefits of cloud-based customer service software? Cloud platforms cost less to start, deploy in days instead of months, scale automatically as volume grows, update without downtime, and let your team work from anywhere. You pay a predictable monthly fee instead of capital expense. ### Is cloud-based customer service software secure? Yes. Most cloud platforms are more secure than on-premise systems because the vendor invests in security full-time, encrypts data, monitors threats continuously, and maintains compliance certifications. You get enterprise-grade security without building it yourself. ## See how the done-for-you model works Cloud-based customer service software removes the infrastructure burden, but you still need to build and maintain the knowledge base, configure the AI agent, and measure what's working. Helpfeel handles all of it. We run the platform, write the content, deploy the AI layer, and track the metrics, so your team can focus on serving customers. [See how the done-for-you model works](/product). --- ## Customer Service Automation Software (2026 Guide) Source: https://en.helpfeel.com/customer-service-software/customer-service-automation-software # Customer Service Automation Software (2026 Guide) Customer service automation software uses AI agents, workflow rules, and self-service tools to handle routine inquiries, so your team focuses on the conversations that need a human. The software answers repetitive questions, routes requests, and manages volume without adding headcount. This guide explains what customer service automation software is, what features matter, and how to pick the right tool for your team. For a broader view of the systems that support service operations, start with the [customer service software guide](/customer-service-software), which covers help desks, workflow tools, and automation platforms together. ## What is customer service automation software? Customer service automation software is a platform that handles routine customer inquiries using AI agents, workflow automation, and self-service tools. It answers questions, routes complex requests to the right team, creates and updates tickets, and tracks resolution without requiring an agent to touch every inquiry. The software works across channels. A customer asking about a return policy in chat gets an instant answer from an AI agent. A customer submitting a billing question through email gets routed to the billing team automatically. A customer searching your help center finds the article they need without contacting you at all. All three are examples of automation working. According to multiple industry reports, only 25% of call centers have successfully integrated AI automation into daily operations as of 2026, even as AI is projected to handle up to 95% of customer interactions by the end of the year. The gap is real. Most teams own automation tools but don't use them effectively. The difference is whether the software fits your actual inquiry volume and whether you've built the knowledge base and routing logic to support it. ## What customer service automation software does Customer service automation software handles four core tasks: 1. **Answers repetitive questions.** The AI agent searches your knowledge base and returns answers to common inquiries: order status, policy lookups, account access, hours, return steps, feature explanations. These questions take agent time but require no judgment. 2. **Routes complex requests.** When a question's too complex, outside the knowledge base, or requires a person, the software routes it to the right team or agent. A billing question goes to billing. A cancellation request goes to retention. An escalation goes to a manager. 3. **Manages ticket volume and workflow.** The software creates tickets, assigns them based on rules, tracks status, and closes resolved inquiries. It tags tickets by question type, priority, or customer segment so your team knows what needs attention. 4. **Tracks performance.** The software measures self-service rate, containment rate, ticket volume, response time, and resolution time. You can see which questions get resolved by automation and which ones need a person. ## Features that matter in customer service automation software Not all automation software is built the same. Here's what to look for. | Feature | What it does | Why it matters | | ------------------------------------------ | ------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | | AI agent with knowledge base search | Reads customer questions, searches your content, and returns answers in conversational language. | This is how you resolve repetitive inquiries without an agent. If the AI can't search your knowledge base accurately, it won't resolve anything. | | Workflow automation and routing | Applies rules to route inquiries, assign tickets, update status, and escalate when needed. | Automation only works if complex requests reach the right person. Workflow rules make sure nothing gets stuck or lost. | | Multi-channel support | Works in chat, email, contact forms, help center search, and phone (IVR). | Your customers ask questions everywhere. The automation needs to work everywhere they are. | | Self-service tools | Help center, searchable knowledge base, FAQ pages, and chat widgets customers can use without contacting you. | Self-service resolves inquiries before they become tickets. The more customers can help themselves, the less volume your team carries. | | Integration with CRM and ticketing systems | Connects to Salesforce, Zendesk, HubSpot, or your existing support platform. | Automation software that can't talk to your CRM or ticketing system creates more work, not less. Integration is required, not optional. | | Handoff to human agents | Knows when to stop and route the inquiry to a person. | Customers get frustrated when automation tries to handle questions it can't resolve. Good software knows its limits and hands off gracefully. | Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The platform includes the content work, the AI layer, and the measurement tools in one package. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. ## How to choose customer service automation software Start with the questions your team answers repeatedly. Pull three months of inquiry data and group it by question type. The question types that show up in volume are your automation targets. Then evaluate software against these criteria: **1. Can it handle your high-volume question types?** If your team spends most of their time answering order status and return policy questions, the software needs to excel at lookups and policy-based answers. If your volume is mostly technical troubleshooting, the software needs to handle conditional logic and multi-step guides. **2. Does it integrate with your existing stack?** If your team lives in Zendesk or Salesforce, the automation software needs to connect to it. If you use a custom CRM, check whether the software offers API access or webhooks. **3. Does it give you clear metrics?** You need to track self-service rate, containment rate, and ticket volume reduction. If the software doesn't measure these, you can't tell whether automation is working. **4. Can your team manage it?** Some automation platforms require engineering resources to configure and maintain. Others are built for support teams to manage directly. Pick software that matches your team's technical capacity. **5. Does it include the content work?** An AI agent is only as good as the knowledge base behind it. If your content is incomplete, outdated, or unclear, automation won't work. Managed platforms like Helpfeel handle the content work for you, so the knowledge base stays current without pulling time from your team. Don't buy for features you won't use in the first six months. Start with the automation that resolves your highest-volume question types, then expand once that's working. For teams that need workflow tools to manage internal processes alongside customer-facing automation, read the [customer service workflow software guide](/customer-service-software/customer-service-workflow-software). ## What to automate first with customer service automation software Automate the questions that meet three criteria: high volume, low complexity, and clear answers. These are the inquiries your team answers the same way every time. **Start with these:** - Order status, tracking, and delivery timelines - Password resets and account access - Return, exchange, and refund policies - Hours, locations, and contact information - Billing questions that need a simple lookup - Product or service feature explanations **Don't automate these yet:** - Complaints or escalations - Requests that require judgment or exception handling - Conversations tied to emotion (cancellations, dissatisfaction, loss) - Anything your team answers differently based on context You want the automation software to carry the repetitive load so your team has time for the conversations that matter. If a question type makes your agents think before answering, it's not ready to automate. For a step-by-step guide to building an automation process, read [how to automate customer support](/customer-support/how-to-automate-customer-support). ## Challenges with customer service automation software Automation software solves repetitive volume, but it creates new challenges if you don't set it up carefully. **Lack of human interaction.** Customers want fast answers, but they also want to feel heard. If automation handles every inquiry and never offers a way to reach a person, satisfaction drops. Good automation software knows when to hand off. **Difficulty handling complex issues.** Automation struggles with questions that require judgment, problem-solving, or empathy. If your software tries to automate these, customers get frustrated and your team gets escalations. **Integration requirements.** If the automation software doesn't connect to your CRM, ticketing system, or knowledge base, it creates duplicate work. Your team ends up managing two systems instead of one. **Content maintenance.** An AI agent is only as good as the knowledge base behind it. If your content is outdated or incomplete, the agent will give outdated or incomplete answers. Automation requires someone to own the content and keep it current. Managed platforms solve the content challenge by handling the knowledge base work for you. Helpfeel runs this work so your team doesn't have to. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. ## How to know if customer service automation software is working Automation software that's working reduces repetitive volume, keeps satisfaction steady or higher, and gives your team time to focus on complex requests. You should see inquiry volume drop, self-service rate climb, and your team spending more time on work that requires a human. Automation software that's failing reduces volume by making customers give up. You see fewer tickets, but you also see satisfaction scores fall, repeat inquiries rise, and customers switching to phone or social media to get around the AI agent. The difference shows up in the data. If your self-service rate and containment rate are both climbing and your CSAT is stable, automation is working. If containment is low or CSAT is falling, the AI agent isn't resolving inquiries. Go back and fix the content or the routing logic. For a detailed breakdown of how to track and interpret these metrics, read the [self-service rate guide](/customer-support-metrics-and-kpis) and the [containment rate guide](/customer-support-metrics-and-kpis/containment-rate-complete-guide). ## Frequently asked questions ### What is customer service automation software? Customer service automation software uses AI agents, workflow rules, and self-service tools to handle routine customer inquiries without an agent. It answers repetitive questions, routes complex requests, and manages ticket volume so your team focuses on work that requires a human. ### What features should customer service automation software include? Look for AI agents that search your knowledge base, workflow automation to route and tag inquiries, self-service tools customers can use directly, and integration with your existing CRM or ticketing system. The software should also know when to hand off to a person. ### How do you choose customer service automation software? Start with the questions your team answers repeatedly. Pick software that can handle those inquiry types, integrates with your current stack, and gives you clear metrics on self-service rate and containment. Don't buy for features you won't use in the first six months. ### Will customer service automation software replace my team? No. Automation software handles repetitive volume so your team can focus on complex requests that need judgment, empathy, or problem-solving. Think of it as the next hire you will not need to make, not a replacement for the people you already have. ## See how the managed model works Customer service automation only works if the knowledge base behind it stays current. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Customer Service Workflow Software: Complete Guide Source: https://en.helpfeel.com/customer-service-software/customer-service-workflow-software # Customer Service Workflow Software: Complete Guide Customer service workflow software automates the routing, assignment, and escalation of customer inquiries, so every request follows a consistent process from start to resolution. It eliminates manual triage, reduces errors, and ensures customers get answers faster. This guide explains what workflow software does, what to automate, and how to measure whether it's working. For a broader look at the customer service software landscape, see the [customer service software guide](/customer-service-software). For the technical side of workflow automation, read [how to automate customer support](/customer-support/how-to-automate-customer-support). ## What is customer service workflow software? Customer service workflow software is a platform that automates how inquiries move through your support operation. It routes incoming requests to the right team or agent, assigns priorities, triggers follow-up tasks, tracks SLAs, and escalates unresolved issues. The goal is to replace manual triage and handoffs with rules that execute instantly. Workflow software works in the background. A customer sends an email asking for a refund. The workflow software reads the content, tags it as a billing issue, routes it to the finance team, sets a 24-hour SLA, and sends an automatic acknowledgment. If the issue isn't resolved within the SLA window, the software escalates it to a manager. No one touched the ticket manually until it landed in the right queue. According to Salesforce, workflow automation reduces manual tasks and errors by handling repetitive actions like data entry, case updates, and follow-up emails, freeing customer service reps to focus on more complex problems. ## What does customer service workflow software do? Workflow software automates the tasks that normally require someone to decide where an inquiry should go and what should happen next. Every platform handles these core functions: **Automatic ticket routing:** Incoming inquiries are assigned to the right agent or team based on priority, question type, skill set, or availability. High-priority issues go to senior agents. Billing questions go to billing. Technical questions go to support engineers. **SLA tracking and alerts:** The software monitors how long each inquiry has been open and alerts the team when an SLA is at risk of being breached. This keeps tickets from falling through the cracks. **Escalation rules:** If an inquiry isn't resolved within a set timeframe or meets certain criteria, the software escalates it to a manager or specialist. Complex issues get attention without manual intervention. **Follow-up automation:** After an inquiry is resolved, the software can trigger follow-up emails, satisfaction surveys, or requests for additional information. These touchpoints happen on schedule without pulling time from the team. **Knowledge article surfacing:** When an inquiry comes in, workflow software can search the knowledge base and surface relevant articles for the agent or customer. This speeds up resolution and increases self-service. The difference between workflow software and a standard ticketing system is what happens after the ticket is created. Ticketing systems store inquiries. Workflow systems move them through the process automatically. ## Why customer service teams use workflow software Teams adopt workflow software to eliminate the manual work that slows down every inquiry. Without automation, every incoming request requires someone to read it, decide where it goes, assign it, and remember to check back if it's not resolved. That handoff work adds time, introduces errors, and makes it hard to keep response times consistent. Workflow software removes those bottlenecks. Inquiries route instantly, SLAs are enforced automatically, and agents spend their time resolving issues instead of triaging them. According to Salesforce, workflows improve single-touch resolutions by automating case classification, routing, and knowledge surfacing, which helps reps resolve issues more quickly and boost first-contact resolution rates. Here's what changes when workflow automation is in place: | Without workflow automation | With workflow automation | | -------------------------------------------- | -------------------------------------------- | | Agents manually triage every inquiry | Inquiries route to the right team instantly | | SLA breaches go unnoticed until too late | Alerts and escalations keep SLAs on track | | Follow-ups depend on someone remembering | Follow-ups trigger automatically on schedule | | Inconsistent service quality across channels | Standard processes apply to every inquiry | Workflow software doesn't replace the team. It removes the repetitive decision-making so agents can focus on solving the actual problem. ## What to automate first in customer service workflows Not every task should be automated right away. Start with the high-volume, low-complexity work that follows clear rules. **Automate these first:** 1. **Ticket routing by question type or channel:** Route billing questions to billing, technical questions to support, and urgent issues to senior agents. 2. **Acknowledgment messages:** Send an instant confirmation when an inquiry comes in, so customers know their request was received. 3. **SLA tracking and alerts:** Monitor open tickets and alert the team when an SLA is at risk. 4. **Escalation rules:** Automatically escalate unresolved issues after a set timeframe or when specific keywords appear. 5. **Post-resolution follow-ups:** Trigger satisfaction surveys or check-in emails after an inquiry is closed. **Don't automate these yet:** - Conversations that require empathy or judgment - Escalations tied to emotion or dissatisfaction - Requests that need context from multiple systems or people - Anything your team handles differently depending on the customer If a task requires a human to think before acting, it's not ready to automate. Workflow software works best when the rules are clear and the action is predictable. ## How to set up customer service workflow automation Follow this sequence to build workflows that reduce manual work without creating new bottlenecks. ### 1. Map the current process for each inquiry type Pull a month of inquiry data and group it by question type: billing, technical support, account access, returns, etc. For each category, write out the steps an inquiry goes through today, from arrival to resolution. Identify the handoffs, delays, and manual decisions. You'll see patterns. Most billing questions follow the same path. Most technical questions need similar information. Those patterns become your automation rules. ### 2. Define routing rules based on priority and skill Decide where each inquiry type should go and who should handle it. Create rules that assign tickets automatically based on content, priority, or channel. For example: inquiries with "refund" in the subject line route to billing. Inquiries tagged "urgent" go to the senior agent queue. Inquiries from VIP accounts route to a dedicated team. Test these rules with historical data before turning them on. You want to catch edge cases before they create customer-facing problems. ### 3. Set SLA targets and escalation triggers For each inquiry type, define how long resolution should take and what happens if that timeframe passes. A billing question might have a 24-hour SLA with escalation to a manager if unresolved after 48 hours. A technical issue might have a 4-hour SLA with immediate escalation if the customer replies saying they're still stuck. Escalation rules keep tickets from sitting unresolved. Make sure every rule has a clear owner. ### 4. Automate acknowledgment and follow-up messages Set up automatic responses for common touchpoints: an acknowledgment when the inquiry arrives, a status update if resolution takes longer than expected, a satisfaction survey after the issue is closed. These messages keep customers informed without requiring manual work. Use clear, conversational language. The goal is to reassure the customer, not to sound like a robot. ### 5. Measure routing accuracy, resolution time, and satisfaction together Track three things: how often workflows route inquiries correctly, how long it takes to resolve them, and whether customers are satisfied with the outcome. Workflow automation should improve all three. If routing accuracy drops, your rules are too broad or too narrow. If resolution time climbs, workflows are creating extra steps instead of removing them. If satisfaction falls, automation is getting in the way of good service. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The platform includes workflow automation, content management, and AI-powered resolution in one system. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. ## How workflow software fits with other customer service tools Workflow software doesn't replace your existing tools. It connects them and automates the handoffs between them. Most teams use workflow software alongside a helpdesk platform, a knowledge base, a CRM, and a chat or email system. The workflow software sits in the middle, reading incoming inquiries, pulling context from the CRM, surfacing articles from the knowledge base, and routing tickets in the helpdesk. For example, a customer sends a chat message asking about an order. The workflow software reads the message, checks the CRM for the customer's history, pulls the order status from the ecommerce platform, and surfaces a knowledge article explaining delivery timelines. If the article resolves the inquiry, the conversation ends. If not, the workflow routes the inquiry to a support agent with all the context already attached. This is why workflow software is often built into larger customer service platforms rather than sold standalone. The value comes from integration, not from the routing logic alone. For teams building their own workflow layer, see the guide on [customer service automation software](/customer-service-software/customer-service-automation-software) for a broader look at how automation tools fit together. ## How to know if workflow automation is working Workflow automation that's working reduces manual triage, speeds up resolution, and keeps service quality consistent. You should see fewer handoffs, faster response times, and higher first-contact resolution rates. Workflow automation that's failing creates new bottlenecks. Tickets route to the wrong team. SLA alerts go ignored. Customers get stuck in loops where the software keeps surfacing the same unhelpful article. You'll see routing errors climb, resolution times increase, and satisfaction scores drop. According to research cited by Apizee, companies that invest in customer experience can see an 80% increase in revenue, with workflow consistency playing a key role in delivering reliable service across channels. The difference shows up in the data. If your first-contact resolution rate climbs and your handling time drops, workflows are helping. If resolution time stays flat or climbs, the automation is adding steps instead of removing them. Go back and simplify the rules. ## Frequently asked questions ### What is customer service workflow software? Customer service workflow software automates the routing, assignment, and escalation of customer inquiries. It directs incoming requests to the right team or agent, triggers follow-ups, and enforces SLAs, so every inquiry follows a consistent process from start to resolution. ### How does workflow automation improve customer service? Workflow automation reduces manual tasks like data entry and ticket assignment, speeds up response times by routing inquiries instantly, and ensures consistent service quality by enforcing standard processes across channels and team members. ### What tasks can customer service workflow software automate? Workflow software automates ticket routing based on priority or skill, SLA tracking and alerts, follow-up emails, case updates, escalations for complex issues, and knowledge article surfacing. Anything repetitive and rules-based can be automated. ### What is the difference between workflow software and helpdesk software? Helpdesk software manages tickets and conversations. Workflow software automates what happens to those tickets, routing them to the right person, triggering actions, and enforcing rules. Most modern platforms combine both capabilities in one tool. ## See how the managed model works Workflow automation works best when it's connected to a strong knowledge base and an AI agent that resolves inquiries before they reach your team. Helpfeel handles the workflow layer, the content work, and the AI resolution in one platform, so your team focuses on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Internal Help Desk Software: Complete Guide Source: https://en.helpfeel.com/customer-service-software/internal-help-desk-software # Internal Help Desk Software: Complete Guide Internal help desk software manages employee requests for IT support, HR assistance, facilities issues, and operational help. It tracks incoming tickets, routes them to the right department, and provides self-service tools so employees can resolve common questions without contacting a person. Strong internal help desk systems reduce resolution time, cut repetitive volume, and let your support teams focus on work that needs judgment. This guide covers what internal help desk software does, how it differs from external customer support tools, what features matter most, and how to measure performance. For broader context on how service software fits into the customer experience, start with the [customer service software guide](/customer-service-software). ## What is internal help desk software? Internal help desk software is a platform that manages and resolves service requests from employees across departments like IT, HR, facilities, finance, and operations. It handles password resets, laptop issues, software access requests, new hire onboarding, office maintenance, policy questions, and benefits inquiries. The software works by centralizing all employee requests in one ticketing system, routing each request to the appropriate team or individual, and tracking the request through resolution. Most platforms include a self-service knowledge base where employees can find answers before opening a ticket. Internal help desk software sits inside your organization. It's the single point of contact for everything an employee might need help with, from a broken monitor to a payroll question to a request for building access. ## What is the difference between internal and external help desk software? Internal help desk software serves employees inside your organization. External help desk software serves customers outside your company. The mechanics overlap, ticket tracking, automation, knowledge bases, and reporting, but the feature priorities differ. **Internal help desk priorities:** - Single sign-on (SSO) and SCIM provisioning to match your employee directory - Integration with Slack, Microsoft Teams, email, and workplace tools - Department-based routing (IT, HR, facilities, finance) - Self-service for common employee requests like password resets or policy lookups - Reporting on internal SLAs and resolution time by department **External help desk priorities:** - Multi-channel support (chat, email, phone, social media) - Customer-facing branding and public knowledge base - CRM integration to pull customer account history - CSAT and NPS tracking to measure customer satisfaction - Escalation paths for billing disputes or product complaints Internal systems prioritize speed and ease for employees who work inside your tools every day. External systems prioritize consistency and satisfaction for customers who interact with your brand across many channels. ## What features should internal help desk software include? The right internal help desk software handles ticket management, automation, self-service, and reporting without requiring employees to learn a new system. Match the platform to how your employees already ask for help, whether that's Slack, email, Teams, or a web portal. ### Ticket management and routing Every request should turn into a ticket, get assigned to the right team, and track through resolution. The platform should route tickets automatically based on the request type: IT issues go to IT, HR questions go to HR, facilities requests go to operations. Manual routing slows everything down. Automation keeps tickets moving. ### Self-service knowledge base Most employee requests are repetitive. Password resets, software access, policy lookups, and how-to questions show up in volume. A self-service knowledge base answers these questions before an employee opens a ticket. The knowledge base should surface answers when employees search, and it should integrate with your help desk so employees can access it from the same place they submit tickets. If self-service is a separate system, employees won't use it. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The same structure that reduces external customer tickets works for internal employee support. ### Automation for repetitive requests Automate the requests that follow a pattern. Password reset workflows, access provisioning, software license requests, and simple policy lookups can resolve without a person. Use automation to close the ticket or route it to the right system. The time your IT and HR teams save on repetitive work is time they can spend on complex requests that need judgment. ### Reporting and analytics Track the metrics that show whether your help desk is working: first response time, resolution time, ticket volume by department, self-service rate, and employee satisfaction. These metrics tell you where the bottlenecks are and which teams need more capacity or better documentation. Good reporting also shows which questions show up most often, so you can write knowledge base articles or automate workflows for those requests. ### Integration with workplace tools Your internal help desk should integrate with the tools your employees use every day: Slack, Microsoft Teams, email, SSO providers, and HR systems. Employees should be able to submit tickets without leaving their workflow. If submitting a ticket requires logging into a separate platform, employees will avoid it or route around it by messaging someone directly. The help desk should meet employees where they are. ## How to choose internal help desk software Choose a platform based on where your employees already ask for help, what volume you handle, and which departments the help desk will serve. | If your organization | Prioritize | | -------------------------------------------------------- | -------------------------------------------------------------------- | | Works primarily in Slack or Teams | Native integration with those platforms | | Has distributed IT, HR, and facilities teams | Strong routing rules and SLA tracking by department | | Handles high repetitive volume (password resets, access) | Automation workflows and a self-service knowledge base | | Needs compliance tracking or audit logs | Reporting tools that export ticket history and resolution timestamps | Start with the features that match how your teams work today, not how you wish they worked. A platform that requires employees to change their behavior won't get adopted. ## How do you measure internal help desk performance? Track these five metrics together to understand whether your help desk resolves requests quickly and sustainably: 1. **First response time:** How long it takes for someone to acknowledge the request. Employees tolerate delays better when they know the ticket's been seen. 2. **Resolution time:** How long it takes to close the ticket. This tells you whether your team has the capacity and tools to resolve issues. 3. **Self-service rate:** The percentage of requests resolved without opening a ticket. Higher self-service means less repetitive work for your team. 4. **Employee satisfaction:** Whether employees feel their issues are resolved quickly and completely. Track this with a short survey after ticket resolution. 5. **Ticket volume by department:** Which departments generate the most requests. This tells you where to invest in better documentation, automation, or staffing. If your resolution time is climbing or your employee satisfaction is falling, dig into the ticket data. You'll find that a small number of question types account for most of the volume. Those are your targets for automation or better self-service content. For more on how to reduce repetitive support volume, see the [customer service automation guide](/customer-service-software/customer-service-automation-software). ## The role of AI in internal help desk software AI helps internal help desks in two ways: resolving repetitive requests automatically and surfacing the right knowledge base article before an employee opens a ticket. An AI agent can handle password reset requests, software access lookups, policy questions, and how-to inquiries without routing to a person. The agent reads the employee's question, searches the knowledge base, and returns an answer in conversational language. If the question's too complex or the knowledge base has no answer, the agent routes the request to the right team. This structure keeps your IT and HR teams focused on the work that requires judgment, not the questions they've answered a hundred times. For a broader look at how AI fits into service operations, read the [AI help desk software guide](/ai-help-center/ai-help-desk-software). ## Frequently asked questions ### What is internal help desk software? Internal help desk software manages and resolves employee requests for IT, HR, facilities, and operations support. It tracks tickets, routes them to the right department, and provides self-service tools so employees can resolve common issues without opening a ticket. ### What is the difference between internal and external help desk software? Internal help desk software serves employees within your organization. External help desk software serves customers outside your company. The mechanics overlap, but internal systems prioritize SSO, SCIM provisioning, and integration with tools like Slack or Teams. ### What features should internal help desk software have? Core features include ticket management, self-service knowledge base, automation for repetitive requests like password resets, reporting on resolution time and request volume, integration with SSO and workplace tools, and routing rules that send requests to the right team. ### How do you measure internal help desk performance? Track first response time, resolution time, self-service rate, employee satisfaction, and ticket volume by department. Together, these show whether your help desk resolves issues quickly and whether employees can resolve common problems on their own. ## See how the managed model works Internal help desks work best when the knowledge base behind them stays current and the AI agent can actually resolve requests. Helpfeel runs that system for you: a managed, AI-ready knowledge base plus an AI agent that resolves inquiries before they reach your team. [See how the done-for-you model works](/product). --- ## Customer Support: The Complete Guide Source: https://en.helpfeel.com/customer-support # Customer Support: The Complete Guide Customer support is the help you give customers when they're stuck with your product or service. It covers technical troubleshooting, how-to questions, account access, and product questions. Strong support teams resolve inquiries quickly and turn repetitive volume into self-service systems so your agents can spend their time on work that actually needs a human. This guide covers what customer support is, how it differs from customer service, how to automate repetitive volume, how to scale as inquiry volume grows, and how to reduce ticket volume before it starts. Each section stands on its own and links to a full guide if you want to go deeper. ## What is customer support? Customer support is the function that helps customers when something's broken, confusing, or unclear about your product. It includes answering technical questions, walking someone through how to use a feature, troubleshooting errors, and resetting passwords. Teams deliver support through email, chat, phone, help centers, and AI agents that resolve inquiries without routing to a person. The work splits into two categories: repetitive questions that can be automated, and complex issues that need a human. The goal is to automate the first category completely so your team has capacity for the second. ## What is the difference between customer support and customer service? Customer support and customer service overlap, but they're not the same thing. Customer support is technical help. It focuses on troubleshooting, product questions, account access, and getting someone unstuck with a feature. Customer service is broader. It includes billing questions, refunds, returns, order tracking, policy clarifications, and general satisfaction work. In practice, support is product help, and service covers the full customer relationship. Some companies call everything customer service. Others separate support (technical, product-focused) from service (transactional, satisfaction-focused). Either structure works as long as the team knows which questions route where. If you sell software or a technical product, most of your volume is support. If you sell physical goods or run a contact center, you likely handle both. ## How do you automate customer support? Automate customer support by identifying the repetitive questions your team answers every day, building a knowledge base that answers those questions in self-service, and deploying an AI agent that retrieves the right answer and serves it before the customer opens a ticket. Automation doesn't replace your team. It removes the repetitive volume so they have time for the work that needs judgment. The playbook has three steps: capture the questions customers ask most often, write clear answers in a searchable knowledge base, and layer an AI agent on top to route inquiries to the right answer or the right agent. Every question the AI resolves is one fewer ticket your team handles. **→ Read the full guide: [How to automate customer support](/customer-support/how-to-automate-customer-support)** ## How do you scale customer support? Scale customer support by raising self-service rate, not by hiring linearly with volume. Every percentage point increase in self-service means fewer tickets per customer, which means you can serve more customers with the same team. The work is building a knowledge base that answers the majority of questions, measuring which gaps remain, and closing those gaps before they turn into tickets. > "Almost 100% could be self-serve. Today it is like 18%." Scaling support is a systems problem. You can't hire your way out of repetitive volume because the volume grows faster than the team. The teams that scale well turn common questions into self-service answers and keep the hiring pace below the growth pace. **→ Read the full guide: [How to scale customer support](/customer-support/how-to-scale-customer-support)** ## How do you reduce support tickets? Reduce support tickets by solving the root causes that create them. The two biggest drivers are missing or hard-to-find help content and product friction that confuses customers before they can resolve their own question. Pull the search queries that returned no useful answer, write articles for those gaps, and make sure the articles surface when customers search. Give customers a faster path to the answer than opening a ticket. When self-service actually works, customers choose it. **→ Read the full guide: [How to reduce support tickets](/customer-support/how-to-reduce-support-tickets)** ## What are the most important customer support metrics? The metrics that matter most are the ones that show whether you're serving customers quickly and sustainably. Self-service rate tells you what share of inquiries resolve without an agent. First response time and resolution time show how fast you answer and close tickets. Cost per contact tells you whether the operation is sustainable as volume grows. Customer satisfaction score shows whether speed and resolution are actually helping. Track these five together. A low cost per contact means nothing if satisfaction is tanking. A high satisfaction score means nothing if it takes three days to respond. The goal is to serve customers well at a cost that scales. For a deeper look at self-service metrics, see the [self-service rate guide](/customer-support-metrics-and-kpis). ## The three parts of a strong customer support strategy | If you want to | Read | | ------------------------------------------------ | -------------------------------------------------------------------------------------- | | Turn repetitive questions into automated answers | [How to automate customer support](/customer-support/how-to-automate-customer-support) | | Serve more customers without hiring linearly | [How to scale customer support](/customer-support/how-to-scale-customer-support) | | Reduce ticket volume before it starts | [How to reduce support tickets](/customer-support/how-to-reduce-support-tickets) | ## How do you build a sustainable customer support operation? Build a sustainable support operation by treating self-service as the front line and human agents as the second line. Most inquiries should resolve in the help center or through an AI agent. The ones that reach a person should be the ones that need judgment, context, or a decision only a human can make. This structure keeps cost per contact stable even as total volume grows, because the repetitive volume goes straight to self-service. It also keeps the team focused on work that matters instead of answering the same five questions all day. The effort was never the problem. The problem was spending that effort on repetitive work instead of building a system that handles it automatically. A managed knowledge base plus an AI agent gives you that system. ## Frequently asked questions ### What is customer support? Customer support is the help you give customers when they are stuck with your product or service. It covers technical troubleshooting, how-to questions, account access, and product questions, delivered through channels like email, chat, phone, and self-service help centers. ### What is the difference between customer support and customer service? Customer support focuses on technical help, product questions, and getting customers unstuck. Customer service is broader and includes billing, returns, refunds, order tracking, and general satisfaction. Support is product help, service covers the whole customer relationship. ### What are the most important customer support metrics? Self-service rate (the share of inquiries resolved without an agent), first response time, resolution time, cost per contact, and customer satisfaction score. Together, they show whether you are serving customers quickly and sustainably. ### How do you reduce customer support volume? Reduce support volume by building a strong self-service knowledge base that answers common questions before customers contact you, using AI to route and resolve repetitive inquiries, and closing gaps in product documentation that create avoidable tickets. ## Go deeper Customer support is a system you build once and keep improving. Helpfeel runs that system for you: a managed, AI-ready knowledge base plus an AI agent that resolves inquiries before they reach your team. See [how the done-for-you model works](/product), or read the guides above to understand the strategy before you build it. --- ## Why Support Cost Scales With Volume Source: https://en.helpfeel.com/customer-support/why-support-cost-scales-with-volume # Why Support Cost Scales With Volume Support cost tends to rise in near lockstep with contact volume because the unit of capacity is a person. Each additional contact needs a proportional slice of agent time, and every repeat of the same question is paid for again from zero. That produces a linear cost curve, which is a structural problem for any business that expects margins to improve as it grows. This is one part of the broader [customer support](/customer-support) challenge. This piece situates why the cost curve looks the way it does. It does not prescribe a fix. ## What does linear support cost actually look like? Picture a mid-market company handling tens of thousands of contacts a month with a lean team, where self-service resolves less than one in five. The rest reach a person. Grow the business by a third and volume climbs with it. Self-service, untouched, resolves the same small fraction, so the absolute number of human-handled contacts rises almost in proportion. The only lever the team has is more people. Cost tracks volume because nothing in the system turns a question answered today into a question that never has to reach a human tomorrow. ## Where does the cost curve come from? The condition is structural, not a staffing mistake. It appears when a company adds customers, products, or geographies faster than it adds the ability to answer them without a human. Three drivers shape it: | Driver | Effect on cost | | -------------------------------- | ------------------------------------------------------------------------ | | People are the unit of capacity | Capacity grows one hire at a time, never continuously | | Repeat questions are re-answered | The same answer is paid for on every occurrence | | Hiring lags demand | A req plus a ramp separate the decision to add capacity from its arrival | ## Why does the curve steepen as you grow? Volume growth compounds and headcount does not. Hiring lags demand by the length of a requisition plus a ramp, so the team is structurally behind during the fastest growth. New products and policy changes generate fresh question types faster than the team absorbs the old ones. And the cost is quiet, because it hides inside a rising payroll line that reads like normal growth rather than inside a metric anyone reviews. ## What does this force a leader to confront? At some threshold the deferred question becomes unavoidable: does support scale through headcount, or does it scale some other way. Every answer inside the linear model is unsatisfying. Hire ahead of demand and you carry cost through every trough. Hire behind it and you fail customers at every peak. Hold headcount flat and quality degrades as volume climbs. The model has a ceiling, and the fastest-growing companies reach it first. The effort was never the problem. The cost structure is. ## Frequently asked questions ### Why does support cost rise with volume? Because the unit of support capacity is a person. Each additional contact needs a proportional amount of agent time, so cost tracks volume almost linearly unless something converts repeat questions into answers that never reach a human. ### What is operating leverage in customer support? Operating leverage is the ability to grow volume faster than cost. Support has low operating leverage when headcount rises at the same rate as contacts, which keeps margin flat exactly when scale should improve it. ### Why is linear support cost a problem? A business expects margins to improve as it grows. When support cost tracks volume one to one, the function has no leverage, so growth stops translating into efficiency and the cost line quietly outpaces revenue. Helpfeel exists because of this exact problem. [See how we think about it](/platform). --- ## The 5 Levels of AI Maturity in Customer Service Source: https://en.helpfeel.com/customer-support/ai-maturity-customer-service # The 5 Levels of AI Maturity in Customer Service AI maturity in customer service has five levels: Ticket Taker, Ticket Deflector, Capable Consultant, Knowledge Curator, and Proactive Guide. Each level marks a shift in what the team is actually optimizing for, from clearing a queue to guiding every customer toward their ideal outcome. Most teams sit lower than they think. This model comes from CX leader Nate Brown. It is useful because it does not measure how much AI you have bolted on. It measures what that AI is for. You can run the same tools at Level Two and Level Five. What changes is the goal. ## Level One: Ticket Taker This is survival mode. The team does the bare minimum to keep the lights on. Requests come in, requests go out, and success is defined as not falling further behind. At this level you are managing queues, not customers. The person on the other end is a ticket number in a backlog. AI, if it exists at all, sorts and routes. It does not resolve. The whole operation is reactive by default, and the customer feels it. ## Level Two: Ticket Deflector Self-service shows up here, but for the wrong reason. The goal is to keep the customer away from a human, not to actually solve their problem. The organization has decided its own time is more valuable than the customer's time, and the experience reflects that. Resolutions are very shallow. A search box returns a wall of articles. A bot answers the easy question and abandons the hard one. The customer technically got "served," but they leave with the real issue intact. Forced self-service is still a step up from a pure backlog, yet it quietly trains customers to expect less. ## Level Three: Capable Consultant This is the first level that genuinely helps. The team is still reactive, waiting for the customer to come to them, but when the customer arrives they get a real answer. Something important also starts here: next-issue avoidance. The agent or the AI does not just close the current ticket. It heads off the follow-up question the customer was about to ask. The focus is still resolving the issue in front of you, not building a long-term relationship, but the resolutions are real and the customer trusts them. Most "good" support organizations live here and assume it is the top. ## Level Four: Knowledge Curator At Level Four the operation starts getting smarter with every customer interaction. AI and people work together, and each conversation feeds the next one. A question that stumped the team yesterday becomes a clear answer today. The knowledge base is no longer a static archive. It is a living system that learns from real demand and improves on its own. Everyone gets smarter in the process, including the customer, who walks away understanding the product better than when they arrived. This is where AI stops being a cost center and starts compounding. Each interaction makes the whole system more capable. ## Level Five: Proactive Guide Level Five has everything from Level Four, plus one change that resets the entire purpose of support. It is proactive. This is no longer about break-fix at all. The team does not wait for something to go wrong. It guides customers toward their ideal state and helps them see what that ideal state could even be. The customer who came in for a password reset leaves knowing the three features that will make their next quarter easier. Support becomes the function that helps customers win, before they think to ask. At this level you are not measuring tickets. You are measuring outcomes. ## How to move up a level The trap is optimizing the level you are already on. A Level Two team buys a better search box and stays a Level Two team. The jump always requires a new goal, not a new tool. | Level | What it optimizes for | The move to the next level | | ------------------------- | ---------------------------- | ----------------------------------------------- | | One: Ticket Taker | Not falling behind | Decide the customer matters more than the queue | | Two: Ticket Deflector | Saving the team's time | Aim for real resolution, not avoidance | | Three: Capable Consultant | Resolving the issue at hand | Make the system learn from every answer | | Four: Knowledge Curator | Getting smarter together | Anticipate the need before it is raised | | Five: Proactive Guide | The customer's ideal outcome | Keep raising the definition of ideal | Notice that the tools can stay the same across several rungs. The same AI that turns customers away at Level Two can curate knowledge at Level Four. Maturity is a decision about intent, then a system built to match it. ## Where most teams actually sit If you ask leaders, most place themselves at Level Three or Four. If you watch the customer experience, most are at Level Two. The gap is the point of the model. It is easy to install self-service and call it maturity. It is hard to build a system where every interaction makes the next one better. Helpfeel is built for the climb from Level Two to Level Five. The search layer resolves real questions instead of returning a list. Every interaction reveals what customers actually ask, so the knowledge base improves itself with human approval in the loop. That is the Level Four shift, and it is the foundation that makes proactive guidance possible. See how the stages map to a working system on the [Helpfeel product page](/product). ## Frequently asked questions ### What are the 5 levels of AI maturity in customer service? Ticket Taker, Ticket Deflector, Capable Consultant, Knowledge Curator, and Proactive Guide. Each level moves from managing queues toward guiding customers to their ideal outcome. ### What is the highest level of customer service AI maturity? Proactive Guide. The team no longer waits for problems. AI and people anticipate needs, reach out first, and help customers reach an ideal state they may not have known was possible. ### How do you move up a level in AI maturity? Stop optimizing the level you are on. Each jump requires a new goal: from clearing queues, to real resolution, to a shared knowledge loop, to anticipating needs before the customer asks. ### Who created the 5 levels of AI maturity model? The framework comes from CX leader Nate Brown, who uses it to separate teams by what their AI is for, not by how much technology they have deployed. --- ## AI Customer Support Agent: What It Is and How to Choose Source: https://en.helpfeel.com/customer-support/ai-customer-support-agent # AI Customer Support Agent: What It Is and How to Choose An AI customer support agent is software that reads customer questions, searches your knowledge base for the answer, and responds in conversational language. It works across channels like help centers, chat, email, and contact forms to resolve repetitive inquiries before they reach your support team. This guide covers what an AI support agent is, how it works, what to look for when choosing one, and how to deploy it effectively. For a broader look at the systems that surround AI agents, start with the [customer support guide](/customer-support), which covers automation strategy, staffing, and metrics. To see the full automation playbook, read the [automation guide](/customer-support/how-to-automate-customer-support). ## What is an AI customer support agent? An AI customer support agent is software that resolves customer inquiries by reading a question, searching your knowledge base or connected systems for the answer, and returning it in natural language. It works in real time across every channel where customers ask questions: your help center, chat widget, contact forms, and email. AI agents are built on large language models that understand natural language, retrieve information from structured content, and generate conversational responses. The agent reads what the customer types, matches it to the right answer in your knowledge base, and delivers that answer in plain language. If the knowledge base has no good answer or the inquiry is too complex, the agent routes the conversation to a human with full context attached. According to Master of Code, 69% of consumers prefer AI-powered self-service tools for quick issue resolution. According to Zendesk, 51% of consumers prefer interacting with bots over humans when they want immediate service. The shift is happening because AI agents resolve routine inquiries faster than waiting for an agent, as long as the answers are accurate and complete. ## How does an AI customer support agent work? An AI agent follows a sequence every time a customer asks a question. **1. The customer submits a question.** This happens in any channel: the help center search bar, a chat widget, an email to support, or a contact form. **2. The agent reads the question and interprets intent.** The AI model parses the natural language query and identifies what the customer is asking. **3. The agent searches your knowledge base or connected systems.** It retrieves the article, data point, or content snippet that best answers the question. Some AI agents also connect directly to order management systems, CRMs, or ticketing platforms to pull real-time data like order status or account details. **4. The agent generates a response in conversational language.** It takes the information it retrieved and formats it as a natural, readable answer. **5. The agent delivers the response in the same channel.** The customer sees the answer immediately, without waiting for a human. **6. The agent routes to a human if needed.** If the agent can't find a good answer, if the customer is frustrated, or if the inquiry requires judgment, the agent hands off to a support agent with full conversation history. | Step | What happens | | --------------------------- | ----------------------------------------------------------------------------------------------- | | Customer asks a question | Help center, chat, email, or contact form | | Agent interprets intent | AI model parses the natural language query | | Agent retrieves information | Searches knowledge base or connected systems | | Agent generates response | Returns a conversational answer based on the content | | Agent delivers or routes | Sends answer immediately, or routes to human if the inquiry is complex or the answer is missing | This process happens in seconds. The customer gets an answer faster than if they waited for a human, and your team only handles the inquiries that need a person. ## What should you look for in an AI customer support agent? Not all AI agents perform the same. The difference between an agent that works and one that frustrates customers comes down to accuracy, integration, routing logic, and measurement. ### Accuracy on your content The agent should return the right answer, not just a plausible one. Accuracy depends on two things: the quality of your knowledge base and the agent's ability to retrieve the best match. If your knowledge base is incomplete, outdated, or unclear, the agent will return incomplete, outdated, or unclear answers. Test the agent on your own content before you deploy it. Ask the questions your customers actually ask. If the agent hallucinates, retrieves the wrong article, or gives a vague answer, it's not ready. ### Integration with your existing tools An AI agent that lives in a silo can't resolve real inquiries. It needs to connect to your knowledge base, ticketing system, CRM, order management platform, or any other system where customer data lives. According to Decagon, AI agents feature deep, bidirectional integration with systems like Salesforce, Zendesk, and Stripe, allowing them to both read and write data. If a customer asks "where is my order," the agent should pull tracking information from your order system and return it in the response. If it can't access that data, the agent will tell the customer to contact support, which defeats the purpose. ### Channel coverage Your AI agent should work everywhere customers ask questions. That means help center search, chat widgets, email, contact forms, and any other channel you support. If the agent only works in chat, customers who email will still create tickets. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The agent works across all channels, so every inquiry has a path to self-service. ### Routing logic that knows when to stop An AI agent should know when to hand off to a human. If the customer's frustrated, if the question is outside the knowledge base, or if the inquiry requires judgment or empathy, the agent should route the conversation immediately. A good agent includes the full conversation history so the human agent starts with context, not from scratch. Test this before you deploy. Try edge cases, emotional language, and escalation requests. If the agent tries to handle everything, it will frustrate customers who need a person. ### Measurement tools that show self-service rate and containment You can't improve what you don't measure. Your AI agent should track self-service rate (the percentage of inquiries resolved without a human), containment rate (the percentage of customers who find an answer and don't contact you again), and satisfaction. If volume drops but satisfaction also falls, the agent is answering questions but not resolving them. For a detailed breakdown of how to track these metrics, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## The role of the knowledge base behind the AI agent An AI agent is only as good as the content it pulls from. If your knowledge base is incomplete, the agent will tell customers it doesn't know. If the articles are outdated, the agent will give outdated answers. If the articles are unclear, the agent will give unclear answers. This is why a managed knowledge base matters. Someone needs to own the content, review it on a schedule, and close gaps as they appear. The AI agent keeps working because the content keeps improving. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. The knowledge base stays current without pulling time from your team. ## How to deploy an AI customer support agent Follow this sequence to deploy an AI agent that actually works. **1. Audit your knowledge base.** Make sure every article is clear, complete, and answers a real customer question. Retire articles that no one reads or that generate follow-up questions. **2. Map inquiry volume by question type.** Pull three months of inquiry data and identify the repetitive questions that show up in volume. Those are the ones the AI agent should resolve. **3. Deploy the agent in one channel first.** Start with your help center or chat widget. Test it with real customers before you expand to email and contact forms. **4. Set clear routing rules.** Define which inquiries the agent should handle and which should go straight to a human. Complaints, escalations, and emotional conversations should route immediately. **5. Measure self-service rate, containment rate, and satisfaction together.** Track all three. If one metric improves but another drops, something's broken. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. The teams that hit those numbers deploy the agent with clean content, clear routing rules, and measurement in place from day one. For additional context on how AI agents fit into the broader help center strategy, read the [AI help center guide](/ai-help-center). ## AI agents and human agents work together An AI agent doesn't replace your support team. It handles the repetitive volume so your team has capacity for work that needs judgment, empathy, or problem-solving. According to NextPhone, 65% of incoming support queries were resolved without human intervention in 2025, up from 52% in 2023. That growth represents time your team gets back to focus on complex inquiries, not headcount you eliminate. > "Self-service really is service if it is done right." (Gina Williams, Midland Radio, on the CX Heroes podcast) Think of an AI agent as the next hire you won't need to make, not a replacement for the people you already have. Your team still owns the customer relationship. The AI agent just removes the repetitive questions that slow them down. ## Frequently asked questions ### What is an AI customer support agent? An AI customer support agent is software that reads customer questions, searches your knowledge base for the answer, and responds in conversational language. It works across channels like help centers, chat, email, and contact forms to resolve inquiries before they reach a human. ### How does an AI customer support agent work? An AI agent reads the customer's question, searches your knowledge base or connected systems for relevant information, generates a conversational response, and delivers it in the same channel the customer asked. If it can't resolve the inquiry, it routes the conversation to a human with full context. ### What should you look for in an AI customer support agent? Look for accuracy on your content, integration with your existing tools, channel coverage across help center, chat, and email, routing logic that knows when to hand off to a human, and measurement tools that show self-service rate and containment rate. ### Will an AI agent replace human support agents? No. An AI agent handles repetitive inquiries so your team has capacity for work that needs judgment, empathy, or problem-solving. It removes the volume that slows your team down, not the team itself. ## See how the managed model works An AI customer support agent only works if the knowledge base behind it stays current. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## AI Customer Support Software: What It Is & How to Choose Source: https://en.helpfeel.com/customer-support/ai-customer-support-software # AI Customer Support Software: What It Is & How to Choose AI customer support software uses AI agents and managed knowledge bases to resolve customer inquiries automatically. It answers repetitive questions before they reach an agent, serves answers through self-service, and routes complex issues to the right person. This guide covers what AI customer support software is, what to look for when evaluating platforms, and how to choose the right solution for your team. For the broader strategy around automation and scaling, see the [customer support guide](/customer-support), which covers the systems and metrics that surround the software layer. ## What is AI customer support software? AI customer support software is a platform that combines an AI agent, a knowledge base, and measurement tools to resolve customer inquiries without routing them to a human. The software searches your knowledge base for the answer, serves it in conversational language, and tracks whether the customer's question was actually resolved. According to McKinsey, AI deployments reduce total interactions by 40-50%. According to Gartner, 80% of routine interactions will be fully handled by AI by 2026. The platforms that hit these numbers share a common structure: a managed knowledge base that stays current, an AI agent that retrieves and serves answers, and measurement that shows whether inquiries are resolved or just redirected. AI customer support software works in every channel customers use: help centers, chat widgets, contact forms, email, and voice. The software should resolve the inquiry wherever it starts, or route it to a human if the question is outside the knowledge base or requires judgment. ## What does AI customer support software do? AI customer support software handles four jobs: 1. **Answers repetitive questions automatically.** The AI agent reads the customer's question, searches the knowledge base, and returns the answer in natural language. No agent involvement. 2. **Reduces ticket volume before it starts.** Customers find answers in self-service instead of opening tickets. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. 3. **Routes unresolved inquiries to the right person.** If the AI agent can't answer the question or the customer asks to speak to someone, the software routes the inquiry to the agent or team best equipped to resolve it. 4. **Measures resolution, not just containment.** Resolution means the customer's question is answered and they don't contact you again. Containment just means the AI touched the inquiry. The difference matters. You want software that tracks both. ## The difference between AI customer support software and an AI agent An AI agent is the software that reads a question and serves an answer. AI customer support software is the full platform: the AI agent, the knowledge base it pulls from, the integrations that connect it to your support channels, and the measurement tools that show whether it's working. You can deploy an AI agent without a managed knowledge base, but it will only be as good as the content you feed it. If the knowledge base is incomplete, outdated, or unclear, the agent will return incomplete, outdated, or unclear answers. This is why most AI agent deployments stall after launch. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The platform includes the content work, the AI layer, and the measurement tools in one package. For more on how AI agents fit into the automation layer, see [AI customer support agent](/customer-support/ai-customer-support-agent). ## How to evaluate AI customer support software Evaluate AI customer support software against the criteria that predict whether it will resolve inquiries, not just touch them. Here's what to look for: ### 1. Resolution capability, not just containment Most AI support software reports containment rate: the share of inquiries the AI touched. That metric tells you nothing about whether the customer's question was actually answered. You want software that tracks resolution: the percentage of inquiries that were fully resolved without a follow-up ticket, without the customer trying again through another channel, and without routing to a human. Target performance benchmarks of containment rates above 65%, accuracy rates of 85% or higher, and CSAT scores above 87% provide concrete evaluation criteria, according to Clarity. ### 2. Knowledge base management included An AI agent is only as good as the knowledge base behind it. If the content is incomplete or out of date, the AI will fail. Look for platforms that include knowledge base management: someone owns the content, reviews it on a schedule, and closes gaps as they appear. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. The knowledge base stays current without pulling time from your team. ### 3. Integration depth with your existing stack The software needs to work in every place customers ask questions: your help center, chat widget, contact forms, email, and ticketing system. Confirm compatibility with CRM systems such as Salesforce or HubSpot and collaboration tools like Slack or Microsoft Teams, according to HappyFox. You don't want to rebuild your support stack to deploy AI. The platform should plug into what you already use. ### 4. Measurement that shows resolution and satisfaction together Track self-service rate, containment rate, and customer satisfaction together. If volume drops but satisfaction also falls, the AI is turning customers away, not resolving them. If satisfaction stays steady or climbs while volume drops, the AI is working. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. The teams that hit those numbers measure resolution and satisfaction side by side and adjust when one starts to slip. For a deeper look at these metrics, see the [self-service rate guide](/customer-support-metrics-and-kpis). ### 5. Scalability with volume and team growth Choose a platform that scales with ticket volume and team growth. Some platforms charge per ticket or per resolution, which means your costs climb directly with volume. Others, like Helpfeel, are priced by the work required to build and maintain the knowledge base, which scales more predictably. You also want software that gets better as volume grows. The AI should learn which answers resolve inquiries and which ones generate follow-ups, and the platform should surface those gaps so you can close them. ## What to automate first with AI customer support software Start with the inquiries that meet three criteria: high volume, low complexity, and clear answers. These are the questions your team answers the same way every time. **Automate these first:** - Order status, tracking, and delivery timelines - Password resets and account access - Return, exchange, and refund policies - Hours, locations, and availability - Billing questions that need a simple lookup - Product or service feature explanations **Don't automate these yet:** - Complaints or escalations - Requests that require judgment or exception handling - Conversations tied to emotion (cancellations, dissatisfaction, loss) - Anything your team answers differently based on context For a full walkthrough of the automation sequence, see [How to automate customer support](/customer-support/how-to-automate-customer-support). ## How to know if your AI customer support software is working AI support software that's working reduces repetitive volume, keeps satisfaction steady or higher, and gives your team time to do work that requires a human. You should see inquiry volume drop, self-service rate climb, and your team spending more time on complex requests and less time answering the same questions. AI support software that's failing reduces volume by making customers give up. You see fewer tickets, but you also see satisfaction scores fall, repeat inquiries rise, and customers switching to phone or social media to get around the AI. The difference shows up in the data. If your containment rate is high and your CSAT is stable, the software is working. If containment is low or CSAT is falling, the AI isn't resolving inquiries. Go back and fix the content or the routing logic. ## The role of a managed knowledge base in AI customer support An AI agent is only as good as the content it pulls from. If your knowledge base is incomplete, outdated, or unclear, the agent will return incomplete, outdated, or unclear answers. This is why most AI support deployments stall after launch. A managed knowledge base solves this. Someone owns the content, reviews it on a schedule, and closes gaps as they appear. The AI agent keeps working because the content keeps improving. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. The knowledge base stays current without pulling time from your team. For a broader look at how knowledge bases fit into the self-service layer, see the [AI help center guide](/ai-help-center). ## AI customer support software and customer service software Customer support software focuses on technical help, product questions, and getting customers unstuck. Customer service software is broader and includes billing, returns, refunds, order tracking, and general satisfaction work. Some AI platforms handle both. Others specialize in one. If you sell software or a technical product, most of your volume is support. If you sell physical goods or run a contact center, you likely handle both support and service inquiries. Choose software that covers the types of inquiries your team actually receives. For more on the service side, see the [customer service software guide](/customer-service-software). ## Frequently asked questions ### What is AI customer support software? AI customer support software combines AI agents and managed knowledge bases to resolve customer inquiries automatically. It answers repetitive questions before they reach an agent, serves answers through self-service, and routes complex issues to the right person. ### How do you choose AI customer support software? Choose AI customer support software by evaluating resolution capability, integration depth, knowledge base management, measurement tools, and whether it scales with volume. Look for platforms that track containment rate and customer satisfaction together, not just ticket volume. ### How much does AI customer support software cost? AI customer support software pricing varies by deployment model. Some platforms charge per ticket or per resolution. Others, like Helpfeel, are fully managed services priced by the work required to build and maintain the knowledge base that powers the AI. ### Will AI customer support software replace my team? No. AI customer support software handles repetitive inquiries so your team focuses on complex work that needs empathy, judgment, or decision-making authority. Think of it as the next hire you will not need to make, not a replacement for current team members. ## See how the managed model works AI customer support software only works if the knowledge base behind it stays current. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## Customer Support KPIs: The 5 Metrics That Matter Source: https://en.helpfeel.com/customer-support/customer-support-kpis # Customer Support KPIs: The 5 Metrics That Matter Customer support KPIs measure whether you are serving customers quickly and at a cost that scales. The five that matter most are self-service rate, first response time, resolution time, cost per contact, and customer satisfaction score. This guide covers what each metric measures, how to track it, and what good performance looks like. For a broader look at how these metrics fit into a support strategy, start with the [customer support guide](/customer-support). ## What are customer support KPIs? Customer support KPIs are quantitative measurements that show how well your support operation is performing. They track speed, efficiency, quality, and cost. The best KPIs give you early warning when something is breaking and confirm when an improvement is working. According to Zendesk, the most tracked customer support KPIs focus on efficiency, quality, and speed of support interactions. The goal is not to optimize every single metric but to track the handful that show whether you are serving customers well at a sustainable cost. ## What are the most important customer support KPIs? Track these five KPIs together. Each one tells part of the story. | KPI | What it measures | Why it matters | | --------------------- | ------------------------------------------------- | -------------------------------------------------------------- | | Self-service rate | Share of inquiries resolved without an agent | Shows whether customers can find answers on their own | | First response time | How quickly you acknowledge an inquiry | Speed matters, customers expect a fast acknowledgment | | Resolution time | How long it takes to fully close a ticket | Tells you whether your team can resolve issues efficiently | | Cost per contact | Total support cost divided by number of inquiries | Shows whether the operation scales sustainably as volume grows | | Customer satisfaction | CSAT or NPS score after a support interaction | Tells you if customers are happy with the help they received | A low cost per contact means nothing if satisfaction is falling. A high satisfaction score means nothing if it costs you three times the industry average to deliver it. Track all five to see the full picture. ## How do you measure self-service rate? Self-service rate is the percentage of customer inquiries that resolve without an agent. It shows whether your help center and AI agent are actually helping or just adding a step before customers contact support. **Formula:** (Inquiries resolved without an agent ÷ Total inquiries) × 100 A strong self-service rate is 60% or higher. Teams with mature knowledge bases and AI agents reach 80 to 98%. If your self-service rate is below 50%, customers cannot find answers on their own, which means every repetitive question is still landing on your team. Helpfeel customers see up to 98% self-service answer rate because the platform combines a managed knowledge base with an AI agent that surfaces the right answer before customers open a ticket. For a detailed breakdown of how to track and improve self-service, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## What is first response time and why does it matter? First response time measures how quickly your team acknowledges a customer inquiry after it arrives. It tracks the time between when a customer submits a question and when an agent first responds. Speed matters because customers expect acknowledgment. Even if you cannot resolve the issue immediately, a fast first response shows the customer that someone is working on it. According to research from multiple support platforms, customers expect a first response within minutes for chat and within a few hours for email. If your first response time is measured in days, customers assume you are ignoring them, even if you are just buried in volume. ## What is resolution time and how is it different from first response time? Resolution time measures how long it takes to fully resolve a customer inquiry and close the ticket. First response time measures acknowledgment. Resolution time measures whether you actually fixed the problem. **Formula:** (Total time from ticket open to ticket close) ÷ Number of tickets resolved A ticket opened at 9am and closed at 2pm has a resolution time of five hours, even if the agent only worked on it for 20 minutes. The clock runs until the issue is resolved. Resolution time tells you whether your team has the information, tools, and authority to close issues quickly. Long resolution times usually mean one of three things: missing information in your knowledge base, too many handoffs between teams, or agents waiting on decisions from someone else. ## What is cost per contact and how do you reduce it? Cost per contact is the total cost of your support operation divided by the number of customer inquiries you handle. It includes agent salaries, software, training, management overhead, and facilities. **Formula:** Total support costs ÷ Total customer contacts Cost per contact shows whether your support operation is sustainable. If it costs you $15 to handle an inquiry and volume doubles next quarter, you need to either hire more agents or reduce the cost per inquiry. Hiring linearly with volume does not scale. The only way to reduce cost per contact without cutting quality is to raise self-service rate. More inquiries resolved by an AI agent means fewer inquiries that reach a paid agent, which brings the average cost down. For example, if 50% of inquiries go to agents and 50% resolve through self-service, your blended cost per contact might be $10. If you raise self-service to 80%, the same inquiry volume costs you less because most inquiries never reach an agent. ## What is customer satisfaction score and how do you track it? Customer satisfaction score (CSAT) measures whether customers are happy with the support they received. It is typically tracked through a post-interaction survey that asks "How satisfied were you with the help you received?" on a 1 to 5 scale. **Formula:** (Number of satisfied responses ÷ Total survey responses) × 100 A satisfied response is usually a 4 or 5. A CSAT score above 80% is strong. Below 70% means customers are not getting the help they need. CSAT tells you whether speed and efficiency are translating into good outcomes. You can have a fast first response time and a low cost per contact, but if customers are unhappy, something in the system is broken. Either the answers are wrong, the routing is bad, or the tone is off. ## How to track customer support KPIs together Do not track these metrics in isolation. A low cost per contact is meaningless if satisfaction is tanking. A high self-service rate is meaningless if customers are giving up instead of finding answers. Track all five together and watch for conflicts. If self-service rate climbs but CSAT drops, customers are getting stuck in your help center instead of finding answers. If resolution time falls but first response time stays high, agents are working faster but customers are still waiting too long for acknowledgment. The goal is not to hit a perfect number on every metric. The goal is to serve customers well at a cost that scales as volume grows. ## How do you improve customer support KPIs? Improving KPIs starts with identifying which one is the bottleneck. If first response time is slow, you need more agents or better routing. If resolution time is slow, you need clearer documentation or fewer handoffs. If self-service rate is low, your knowledge base has gaps or your AI agent is not surfacing answers. The most effective improvement for almost every support team is raising self-service rate. Every percentage point increase in self-service means fewer tickets, which reduces cost per contact, speeds up first response time (because agents have less queue), and often improves CSAT because customers get answers immediately instead of waiting for a reply. To raise self-service rate, pull the search queries that returned no useful answer, write knowledge base articles for those gaps, and deploy an AI agent that retrieves the right answer before customers open a ticket. Helpfeel handles this work for you in one platform: a managed knowledge base that closes content gaps and an AI agent that resolves inquiries before they reach your team. For more on building a support operation that scales, read the guides on [how to automate customer support](/customer-support/how-to-automate-customer-support) and [how to scale customer support](/customer-support/how-to-scale-customer-support). ## Frequently asked questions ### What are the most important customer support KPIs? The five most important customer support KPIs are self-service rate, first response time, resolution time, cost per contact, and customer satisfaction score. Track them together to see whether you are serving customers quickly and sustainably. ### What is a good self-service rate? A strong self-service rate is 60% or higher, meaning more than half of inquiries resolve without an agent. Teams with mature knowledge bases and AI agents reach 80 to 98%. Low self-service means customers cannot find answers on their own. ### What is the difference between first response time and resolution time? First response time measures how quickly your team acknowledges a customer inquiry. Resolution time measures how long it takes to fully resolve the issue and close the ticket. Both matter, but resolution time shows whether you actually fixed the problem. ### How do you reduce cost per contact without hurting satisfaction? Reduce cost per contact by raising self-service rate. More inquiries resolved without an agent means lower average cost per inquiry. Satisfaction stays steady or improves because customers get answers faster through self-service than waiting for a ticket response. ## See how the done-for-you model works Customer support KPIs only improve when the systems behind them improve. Helpfeel is a done-for-you platform that manages your knowledge base, deploys an AI agent to resolve inquiries before they reach your team, and tracks the metrics that show whether it is working. [See how the done-for-you model works](/product). --- ## How to Automate Customer Support (Step-by-Step Guide) Source: https://en.helpfeel.com/customer-support/how-to-automate-customer-support # How to Automate Customer Support (Step-by-Step Guide) Customer support automation uses AI agents and knowledge bases to answer repetitive questions, so your team handles the work that needs a human. When it's done well, automation resolves routine inquiries faster than an agent can, keeps satisfaction steady or higher, and frees your team to focus on complex requests. This guide walks through what to automate, how to set it up, and how to measure success. For a broader view of the systems that surround automation, start with the [customer support guide](/customer-support), which covers strategy, staffing, and quality alongside the automation layer. ## What is customer support automation? Customer support automation is the use of AI agents, managed knowledge bases, and self-service tools to resolve customer inquiries without an agent. It answers the questions that show up in volume and follow a pattern: order status, password resets, return steps, policy lookups, hours, feature explanations. Automation works best when it handles the repetitive volume. It struggles when it tries to replace judgment. A customer asking "where is my order" can get an instant, accurate answer from an AI agent that pulls the tracking number. A customer who needs to cancel a service due to financial hardship should talk to a person. The difference is whether the conversation requires empathy, problem-solving, or authority to make exceptions. When automation is set up right, it resolves the first type and routes the second type to the right person. When it's set up wrong, it tries to handle both and leaves customers frustrated. ## What should you automate first? Automate the questions that meet three criteria: high volume, low complexity, and clear answers. These are the inquiries your team answers the same way every time. **Start with these:** - Order status, tracking, and delivery timelines - Password resets and account access - Return, exchange, and refund policies - Hours, locations, and availability - Billing questions that need a simple lookup - Product or service feature explanations **Don't automate these yet:** - Complaints or escalations - Requests that require judgment or exception handling - Conversations tied to emotion (cancellations, dissatisfaction, loss) - Anything your team answers differently based on context You want the AI agent to carry the repetitive load so your team has time for the conversations that matter. If a question type makes your agents think before answering, it's not ready to automate. One customer success director at a national office-equipment company said their team spent half their time answering the same five questions, and the other half solving real problems. Automation gave them that time back. ## How to automate customer support in five steps Follow this sequence. Each step builds on the one before it. ### 1. Map your inquiry volume by question type Pull three months of inquiry data from your ticketing system or contact center platform. Group inquiries by question type, not by department or tag. You want to see the actual questions customers ask. You'll find that a small number of question types account for most of your volume. Those are your automation targets. ### 2. Build or update your knowledge base with clear, complete answers For each high-volume question, write one article that answers it completely. Use the question itself as the heading. Answer it in the first sentence, then add detail below. Each article should stand on its own. No "see above" references, no jargon, no assumptions about what the reader already knows. The AI agent and the customer both need to extract a full answer from one piece of content. If your knowledge base is already built, audit it. Articles that sit unread or generate follow-up questions aren't helping. Update them or retire them. ### 3. Deploy an AI agent that can search your knowledge base and resolve inquiries An AI agent is software that reads a customer question, searches your knowledge base for the answer, and returns it in conversational language. The agent should work in every place customers ask questions: your help center, chat widget, contact forms, and email. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. The platform includes the content work, the AI layer, and the measurement tools in one package. Your AI agent should also know when to stop. If the customer's frustrated, if the question's complex, or if the knowledge base has no good answer, the agent should route the inquiry to a person immediately. ### 4. Route unresolved inquiries to the right team member Not every question gets resolved by the AI agent. Some questions are outside the knowledge base. Some are too complex. Some customers prefer to talk to a person. Your automation setup needs a clear handoff rule. Tag unresolved inquiries with the question type and route them to the agent or team best equipped to answer. A billing question goes to billing, a product bug goes to support engineering, a cancellation request goes to a senior agent who can negotiate. Make sure every inquiry gets resolved, whether by the AI agent or by a person. ### 5. Measure self-service rate, containment rate, and satisfaction together Track three metrics: - **Self-service rate:** the percentage of inquiries resolved without an agent. This tells you how much volume the AI agent is carrying. - **Containment rate:** the percentage of customers who find an answer and don't contact you again. This tells you whether the answers are actually working. - **Customer satisfaction or CSAT:** whether customers are happy with the resolution. This tells you if the automation is helping or frustrating people. Helpfeel customers see up to 70% ticket reduction and a 98% self-service answer rate. The teams that hit those numbers measure all three metrics and adjust when one starts to slip. If your self-service rate climbs but your satisfaction drops, the AI agent is answering questions but not resolving them. Go back to step two and improve the content. For a detailed breakdown of how to track and interpret these metrics, read the [self-service rate guide](/customer-support-metrics-and-kpis). ## The role of a managed knowledge base in automation An AI agent is only as good as the content it pulls from. If your knowledge base is incomplete, outdated, or unclear, the agent will return incomplete, outdated, or unclear answers. This is why most automation projects stall after launch. A managed knowledge base solves this. Someone owns the content, reviews it on a schedule, and closes gaps as they appear. The AI agent keeps working because the content keeps improving. Helpfeel runs this work for you. We write the articles, watch what customers search for, flag content that isn't resolving inquiries, and ship updates on a regular cadence. The knowledge base stays current without pulling time from your team. ## How to know if automation is working Automation that's working reduces repetitive volume, keeps satisfaction steady or higher, and gives your team time to do work that requires a human. You should see inquiry volume drop, self-service rate climb, and your team spending more time on complex requests and less time answering the same questions. Automation that's failing reduces volume by making customers give up. You see fewer tickets, but you also see satisfaction scores fall, repeat inquiries rise, and customers switching to phone or social media to get around the AI agent. > "Self-service really is service if it is done right." (Gina Williams, Midland Radio, on the CX Heroes podcast) The difference shows up in the data. If your containment rate's high and your CSAT is stable, automation is working. If containment is low or CSAT is falling, the AI agent isn't resolving inquiries. Go back and fix the content or the routing logic. ## Frequently asked questions ### What is customer support automation? Customer support automation uses AI agents, knowledge bases, and self-service tools to answer repetitive customer questions without an agent. It handles routine inquiries so your team focuses on complex requests that need a human. ### What should you automate first in customer support? Automate the repetitive questions that show up in volume: order status, password resets, policy lookups, hours, return steps. These take agent time but require no judgment. Leave complex or emotional conversations to humans. ### How do you measure customer support automation success? Track self-service rate, containment rate, and ticket volume reduction. A healthy automation setup resolves inquiries and keeps satisfaction steady. If volume drops but satisfaction also falls, customers are getting stuck, not helped. ### Will automation replace my support team? No. Automation handles the repetitive volume, so your team does work that requires empathy, judgment, and problem-solving. Think of it as the next hire you will not need to make, not a replacement for the people you already have. ## See how the managed model works Customer support automation only works if the knowledge base behind it stays current. Helpfeel handles the content work, the AI layer, and the measurement in one platform, so your team can focus on the conversations that need a human. [See how the done-for-you model works](/product). --- ## How to Reduce Customer Support Tickets Source: https://en.helpfeel.com/customer-support/how-to-reduce-support-tickets # How to Reduce Customer Support Tickets Reduce support tickets by finding the top repeat questions your team answers every week, fixing or writing the knowledge base articles for those questions, then surfacing those answers with an AI agent at the front door. Most teams see measurable ticket reduction within two weeks of fixing the top five repeat questions. This guide walks through why ticket volume piles up, the four-step approach to reducing it, and what to measure as you go. If you want a broader view of the challenge, start with [customer support](/customer-support), the hub this guide belongs to. ## Why does ticket volume pile up? Ticket volume piles up because the same questions get asked over and over, and customers can't find the answer before they contact you. It looks like a staffing problem. It's actually a findability problem. Three things drive repeat volume: 1. **The answer isn't in the knowledge base.** You fixed the issue in a ticket once, but never wrote it down anywhere a customer could read it. 2. **The answer's there, but customers can't find it.** Search returns nothing useful, or the article's buried five clicks deep, or the title doesn't match how people search. 3. **The answer's there and findable, but nobody surfaces it.** Customers start typing an email or chat message without searching first, and nobody intercepts them with the relevant article. All three are fixable. None of them require hiring more agents. ## How do you reduce support tickets? Reduce support tickets by following a four-step loop: find the top repeat questions, fix the knowledge base for those questions, put an AI agent at the front door to surface answers, then close the loop by tracking what still falls through. Each cycle cuts more volume. Here's the step-by-step approach. ### Step 1: Find the top repeat questions Pull every ticket from the past 30 days and tag each one with the question the customer was asking. Group identical questions together, then sort by count. The top five questions usually account for 30 to 50 percent of total volume. If you don't have tags, read a random sample of 100 tickets and note the question in each one. Patterns show up fast. You're not looking for precision. You're looking for the handful of questions you answer ten times a week. ### Step 2: Fix the knowledge base for those questions For each of the top five repeat questions, check whether a knowledge base article exists. If it doesn't, write one. If it does, read it as if you're a customer and ask: did I find my answer in the first ten seconds? Fix these three things in every article: - **Make the title match how people search.** If customers type "how do I reset my password," the article title should be "How to reset your password," not "Account recovery options." - **Put the answer first.** The first paragraph should completely answer the question. Save context and edge cases for later. - **Keep it short.** One article, one question. If the article tries to answer three different things, split it into three articles. Write the articles in plain language. Short sentences, active voice, you. No jargon the customer doesn't already know. ### Step 3: Add an AI agent at the front door An AI agent watches what customers type into search, chat, or a contact form, then surfaces the relevant knowledge base article before the customer sends a ticket. Most teams see a 30 to 70 percent reduction in ticket volume within the first month because the AI catches repeat questions automatically. The AI only works if the knowledge base is fixed first. An AI that points to bad articles makes the problem worse. Fix the content, then add the AI. Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. ### Step 4: Close the loop on gaps Every two weeks, pull the searches that returned no useful answer and the questions that still became tickets. Pick the top three and repeat the cycle: write or fix the article, surface it with the AI, measure again. Ticket reduction isn't a one-time project. It's a loop that keeps finding and fixing the next repeat question. The teams that cut volume by half run this loop every two weeks. The teams that cut it once and stop see volume creep back up within three months. ## What should you measure? Measure three things: ticket volume, self-service rate, and the top unanswered searches. Ticket volume tells you whether the approach is working. Self-service rate tells you whether customers found real answers or gave up trying. Unanswered searches tell you what to fix next. Track these numbers every two weeks, not once a quarter. You want to see the reduction happen in real time so you can adjust the loop as you go. | Metric | What it measures | Why it matters | | ----------------------- | ---------------------------------------------------------------------- | ---------------------------------------------------------------- | | Ticket volume | How many tickets came in this period | The bottom-line number you're trying to reduce | | Self-service rate | How many customers resolved their question without contacting an agent | Proves the reduction came from real resolutions, not frustration | | Top unanswered searches | The searches that returned no useful article | Shows you exactly what to fix in the next cycle | If you want a deeper guide to tracking ticket reduction and proving ROI, read the full measurement guide at [reduce support ticket volume](/customer-support-metrics-and-kpis/reduce-support-ticket-volume). That page covers cost per contact, containment rate, and how to report the dollar value to your CFO. ## How long does it take to reduce ticket volume? Most teams see measurable reduction within two weeks of fixing the top five repeat questions and adding an AI agent. The first cycle cuts 20 to 40 percent of volume. The second cycle cuts another 10 to 20 percent. After three or four cycles, you hit the floor: the remaining tickets are complex cases that genuinely need a human. Speed depends on how fast you can fix the knowledge base. If you already have articles for the top repeat questions and just need to surface them better, you can see results in days. If you need to write everything from scratch, budget two to three weeks for the first cycle. The work doesn't stop after the first reduction. Ticket volume is a living thing. Products change, new questions appear, and old articles go stale. The teams that keep volume low run the four-step loop every two weeks, forever. ## Can you reduce tickets without cutting staff? Yes. You're handling the repetitive questions automatically so your team can focus on the work that needs a human: complex troubleshooting, high-value accounts, and the edge cases that require judgment. Think of it as the next hire you won't need to make. As your company grows, ticket volume grows with it. An AI agent keeps that growth from turning into a staffing crisis. ## Frequently asked questions ### What is the fastest way to reduce support tickets? Find the top five questions driving ticket volume, write or fix the knowledge base articles for those questions, then surface them with an AI agent. Most teams see measurable reduction within two weeks of fixing the top repeat questions. ### Can you reduce tickets without hiring more people? Yes. Most ticket volume is repeat questions that a strong knowledge base plus an AI agent can answer. You are not replacing current staff. You are handling the repetitive volume so your team can focus on the work that needs a human. ### How much can you reduce support ticket volume? Teams typically reduce ticket volume by 30 to 70 percent when they fix the knowledge base and add an AI agent. The exact number depends on how much of your current volume is repeat questions versus complex cases. ### Do you need to track ticket reduction separately from self-service rate? Yes. Ticket reduction measures volume. Self-service rate measures whether customers actually found an answer or gave up trying. Track both so you know the reduction comes from real resolutions, not frustration. ## Go deeper Ticket reduction is a loop, not a project. Helpfeel runs that loop for you: we build the knowledge base, watch what customers search for, close the gaps before they become tickets, and keep the system improving every two weeks. See [how the done-for-you model works](/product). --- ## How to Scale Customer Support Without Adding Headcount Source: https://en.helpfeel.com/customer-support/how-to-scale-customer-support # How to Scale Customer Support Without Adding Headcount Scaling customer support means handling more volume without adding headcount at the same rate. A team that scales well uses self-service, a managed knowledge base, AI agents, and process design to resolve more questions without hiring proportionally. The result: a support function that grows capacity while keeping costs flat. This is one piece of the broader [customer support](/customer-support) challenge. If you're earlier in the process, start there. If you know you need to scale, read on. ## What does it mean to scale customer support? Scaling customer support means your team handles more questions, across more channels, without hiring at the same pace the volume grows. A team handling 1,000 tickets with five agents can handle 2,000 with six instead of ten. The teams that scale well build systems that absorb repetitive volume before it reaches a person. Self-service answers the common questions. AI agents handle the lookups. Human agents focus on the work that needs judgment. You grow the business without expanding the team at the same rate. The alternative is linear scaling, where every new 200 tickets requires another full-time hire. That works for a while, but it breaks when margins tighten or volume spikes unexpectedly. ## What are the levers for scaling support? Four levers let you handle more volume without growing headcount at the same rate. Most teams need all four working together. **Self-service knowledge base.** A current, searchable help center lets customers resolve their own questions. The better the content and the easier it is to find, the fewer questions reach your team. Strong self-service can answer the majority of inquiries before a customer contacts an agent. **AI agent.** An AI assistant that uses your knowledge base to answer questions in real time handles repetitive volume immediately. Helpfeel customers see up to 70% ticket reduction by resolving common questions before they escalate. The AI handles lookups, status checks, and product questions. Your team works on the cases that need a human. **Process and tiering.** Tiering means routing simple questions to a quick-resolution tier and complex ones to senior agents. Most tickets should resolve in tier one, so your most experienced people focus on cases that need depth. This is structure, not automation, but it's one of the highest-leverage changes you can make. **Proactive content.** Close the gaps in your knowledge base before customers ask. Watch what people search for, pull the questions that returned no answer, and write the article. You turn a future ticket into a resolved question. ## The playbook: how to scale customer support in five steps Follow this sequence. Each step builds on the one before it. **1. Measure your baseline.** Count the tickets you handle today, the time each takes, and how many repeat the same question. You'll learn where automation and self-service will reduce the most volume. Track your [self-service rate](/customer-support-metrics-and-kpis) and cost per contact as starting metrics. **2. Build or refresh your knowledge base.** Write clear, scannable articles for the questions that repeat most often. Each article should answer one question, start with the answer, and use headings people actually search for. A well-written knowledge base is the foundation every other scaling lever depends on. **3. Deploy an AI agent.** An AI assistant that pulls from your knowledge base resolves common questions immediately, before they reach your team. This is the fastest way to absorb repetitive volume. Helpfeel runs as a done-for-you managed service, so you get the knowledge base and the AI agent as one system. **4. Tier your team and triage ruthlessly.** Route simple inquiries to a fast-resolution tier and complex ones to senior agents. Most volume should close in tier one. If everything lands with your best people, you're underusing the team you already have. **5. Close the gaps before they become tickets.** Pull unanswered searches and common follow-up questions each week. Write the missing article or update the one that sent customers away confused. This is how you keep scaling instead of plateauing after the first wave. ## Common mistakes when scaling support **Skipping the baseline.** You can't scale what you haven't measured. Teams that skip the measurement step guess at what automation will help, then build the wrong thing. Start with the data. **Launching self-service and calling it done.** A knowledge base built once and left alone goes stale within months. The teams that scale treat the knowledge base as a system they improve every quarter, not a project they finish. **Adding automation but keeping the same process.** If your AI agent or help center resolves 40% of volume but you still route everything through the same queue, you get no time back. Change the triage process so resolved questions stop reaching the team at all. **Building content customers can't find.** A great article buried in a folder no one opens saves zero tickets. Structure your help center so the highest-volume questions appear at the top, and make search work well. Findability matters as much as accuracy. **Ignoring what AI can't do.** AI handles repetitive questions and lookups. It doesn't replace judgment, empathy, or relationship work. Keep your best people focused on the cases that need a human, and let the AI take the repetitive volume. ## How do you know scaling is working? Track these three metrics to see whether your scaling work is paying off. **Tickets per agent.** If each agent resolves more tickets per week without burning out, you're scaling. The work should feel easier, not harder. **Self-service rate.** Measure the percentage of inquiries resolved before they reach an agent. A good target is 90% or higher. If that number climbs while customer satisfaction stays flat or improves, you're doing it right. **Cost per resolution.** Divide your total support cost by the number of inquiries resolved (both self-service and agent-handled). That number should drop over time as self-service absorbs more volume. If all three metrics improve together, you've built a system that scales. If tickets per agent climbs but satisfaction drops, you're cutting corners instead of scaling well. ## The next hire you won't need to make Scaling support well means the next growth spike doesn't force an immediate hiring round. You handle more volume with the team you already have, because the systems you built absorb the repetitive work. That doesn't mean you never hire again. It means you hire for judgment, expertise, and relationship work, not for answering the same question 200 times a week. The AI and the knowledge base handle the repetitive volume, so your team does the work that needs a human. For related strategies, see [how to reduce support tickets](/customer-support/how-to-reduce-support-tickets) and [how to automate customer support](/customer-support/how-to-automate-customer-support). ## Frequently asked questions ### What does it mean to scale customer support? Scaling customer support means handling more volume without adding headcount at the same rate. A scaled team uses self-service, knowledge base content, AI agents, and process design to resolve more questions without growing the team linearly. ### What is the first step to scaling support? The first step is measuring what volume you actually handle today and identifying which questions repeat. That baseline tells you which self-service content or automation will reduce the most volume. ### How do AI agents help scale support? AI agents answer common questions immediately using your knowledge base, so they resolve repetitive volume before it reaches a human agent. That frees your team to focus on complex issues that need judgment. ### When should you add headcount instead of scaling with systems? Add headcount when you need more judgment, relationship depth, or expertise in complex cases. Scale with systems for repetitive questions, product documentation lookups, and simple transactional requests. ## See how the done-for-you model works Helpfeel is a done-for-you customer support platform: a managed, AI-ready knowledge base plus an AI agent that helps customers find answers and resolve their own questions, so support teams handle less repetitive volume. We build the content, deploy the AI, and keep it current, so you get the scaling systems without adding a project to your backlog. [See how it works](/product). --- ## Support Metrics & KPIs: The Complete Guide Source: https://en.helpfeel.com/customer-support-metrics-and-kpis # Support Metrics & KPIs: The Complete Guide This overview is a placeholder while the full guide is in progress. It brings together the metrics that matter most for measuring customer support performance: self-service rate, deflection rate, containment rate, and ticket volume. Each one is covered in its own complete guide below. ## What this guide covers Support metrics fall into two broad groups: metrics that measure how much support your team is handling, and metrics that measure how well customers are helping themselves. The guides in this section walk through both, including formulas, current benchmarks, and the strategies that move each number. ## Guides in this section - [Self-Service Rate: The Complete Guide](/customer-support-metrics-and-kpis/self-service-rate-complete-guide) - [Deflection Rate: The Complete Guide](/customer-support-metrics-and-kpis/deflection-rate-complete-guide) - [Containment Rate: The Complete Guide](/customer-support-metrics-and-kpis/containment-rate-complete-guide) - [How to Reduce Support Ticket Volume](/customer-support-metrics-and-kpis/reduce-support-ticket-volume) - [Self-Service vs. Deflection vs. Containment Rates](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) --- ## Self-Service Rate vs. Deflection Rate vs. Containment Rate: A Strategic Guide Source: https://en.helpfeel.com/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article # Understanding Self-Service Rate vs. Deflection Rate vs. Containment Rate: A Strategic Guide ## Executive Summary In the evolving landscape of customer support, three key metrics often get conflated or misunderstood: self-service rate, deflection rate, and containment rate. While these terms are sometimes used interchangeably, understanding their distinct meanings and strategic implications is crucial for building effective customer experience (CX) strategies. ## Key Definitions ### [Self-Service Rate](/customer-support-metrics-and-kpis) **Definition**: The percentage of customers who successfully resolve their issues using self-service resources (knowledge base, FAQ, AI chat) without requiring human agent assistance. **Formula**: (Successful self-service resolutions / Total customer inquiries) × 100 **What it measures**: Customer independence and knowledge base effectiveness → [Read the complete guide to Self-Service Rate](/customer-support-metrics-and-kpis) ### [Deflection Rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide) **Definition**: The percentage of support tickets or calls prevented from reaching human agents through proactive self-service interventions. **Formula**: (Inquiries resolved before ticket creation / Total potential inquiries) × 100 **What it measures**: Workload reduction and cost savings → [Read the complete guide to Deflection Rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide) ### [Containment Rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide) **Definition**: The percentage of customer interactions that are fully resolved within the automated system without escalation to human agents. **Formula**: (Automated resolutions without escalation / Total automated interactions) × 100 **What it measures**: AI/automation effectiveness and true resolution quality → [Read the complete guide to Containment Rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide) ## The Critical Differences ### 1. Measurement Point - **Self-service rate** measures at the resolution point - **Deflection rate** measures at the prevention point - **Containment rate** measures at the completion point ### 2. Strategic Focus - **Self-service**: Customer empowerment and satisfaction - **Deflection**: Cost reduction and efficiency - **Containment**: Quality and effectiveness of automation ### 3. Quality vs. Quantity As industry data shows, high deflection rates can mask poor customer experiences. A system might deflect 70% of inquiries but only truly resolve 20% of them: the gap represents frustrated customers who gave up rather than succeeded. ## Industry Benchmarks Based on analysis of 36+ case studies across different support maturity levels: ### Category 1: No Self-Service Today - Typical deflection rate: 0% - Post-implementation potential: 30-40% reduction in inquiry rates - Examples: SOLIA (40% inquiry rate drop), SOKUYAKU (84% reduction, costs -50%) ### Category 2: Weak Self-Service - Current deflection: 10-30% - Improvement potential: 20-40% additional gains - Industry average for tech: 23% deflection rate ### Category 3: Strong Self-Service - Best-in-class: 50%+ deflection with high containment - Retail example: ADT achieving 70% containment rates ## The Vanity Metric Trap Industry leaders increasingly recognize that deflection rate alone is a "vanity metric." High deflection without corresponding containment or resolution quality means: - Customers abandoning rather than succeeding - Automated frustration at scale - Erosion of trust and satisfaction **Key insight**: "If your deflection rate is high while your resolution rate is low, you're slowly spending trust down." ## Strategic Implementation Approaches ### 1. The Quality-First Approach - Focus on accurate, findable content - Measure resolution quality, not just deflection - Build trust through reliable self-service ### 2. The Multi-Channel Strategy - Contact form intercepts ("Chasing FAQ") - Email auto-replies with KB suggestions - Portal search upgrades - Gradual channel shift through visibility ### 3. The Measurement Framework Essential metrics to track together: - Search volume and patterns - Reach rate (found relevant content) - Resolution confirmation - Escalation patterns - Customer satisfaction post-self-service ## Common Pitfalls to Avoid 1. **Measuring deflection without quality**: Leads to automated frustration 2. **Ignoring containment gaps**: Miss where customers abandon vs. succeed 3. **One-size-fits-all targets**: Different channels have different optimal rates 4. **Static content**: Knowledge debt accumulates without continuous updates ## Best Practices for Optimization ### Content Strategy - Regular audits of top 20 articles - Voice of Customer (VoC) data integration - Continuous improvement cycles - Clear, customer-language writing ### Technology Integration - Unified search across channels - Analytics dashboards tracking all three metrics - AI/human handoff optimization - Feedback loops for content improvement ### Organizational Alignment - Cross-functional ownership (not just support) - Executive visibility on true resolution rates - Investment in knowledge management - Long-term commitment (99% renewal rates for committed teams) ## The Future: Beyond Deflection Leading organizations are shifting focus from deflection to: - **Revenue impact**: Did self-service accelerate purchases or prevent churn? - **Strategic value**: Knowledge as a competitive advantage - **Holistic success**: Customer lifetime value, not just ticket reduction ## Key Takeaways 1. **Deflection ≠ Success**: High deflection with low containment erodes trust 2. **Measure what matters**: Track resolution quality, not just prevention 3. **Invest in accuracy**: 1% KB improvement multiplies across all interactions 4. **Think strategically**: Self-service is a revenue function, not just cost reduction 5. **Commit long-term**: The most successful implementations see 99% renewal rates ## Conclusion While self-service rate, deflection rate, and containment rate are related metrics, understanding their distinctions enables more strategic CX decisions. The most mature organizations recognize that true success lies not in deflecting customers but in empowering them with accurate, accessible knowledge that drives both satisfaction and business value. The shift from viewing these metrics as cost-reduction tools to strategic business drivers represents the evolution of modern customer support, from reactive ticket management to proactive customer enablement. ## Explore Each Metric in Depth 📚 **Deep Dive into Individual Metrics:** ### [📊 Self-Service Rate: The Complete Guide](/customer-support-metrics-and-kpis) Master the art of customer empowerment through effective self-service strategies, benchmarks, and best practices. ### [🚀 Deflection Rate: The Complete Guide](/customer-support-metrics-and-kpis/deflection-rate-complete-guide) Learn how to prevent tickets effectively while avoiding the vanity metric trap and maintaining quality. ### [🎯 Containment Rate: The Complete Guide](/customer-support-metrics-and-kpis/containment-rate-complete-guide) Discover how to measure true automation success and deliver complete resolutions within automated systems. --- _Source: Analysis based on Helpfeel case studies, industry research, and CX leadership insights from 2026._ --- ## Containment Rate: The Complete Guide Source: https://en.helpfeel.com/customer-support-metrics-and-kpis/containment-rate-complete-guide # Containment Rate: The Complete Guide [← Back to Overview: Self-Service vs. Deflection vs. Containment](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) ## What is Containment Rate? Containment rate measures the percentage of customer interactions that are fully resolved within automated systems without requiring escalation to human agents. It's the most stringent quality metric for AI and automation effectiveness, focusing on complete resolution rather than just deflection. **Formula**: (Automated resolutions without escalation / Total automated interactions) × 100 ## Why Containment Rate is the Ultimate Quality Metric ### Beyond Surface Metrics While deflection rate shows tickets avoided and self-service rate shows independent success, containment rate reveals: - **True automation effectiveness**: Did the AI actually solve the problem? - **Customer trust indicators**: Are customers satisfied with automated help? - **Quality of resolution**: Not just answered, but completely resolved - **ROI validation**: Actual value delivered vs. claimed savings ### The Quality Gateway Containment rate acts as a reality check on other metrics: - High deflection + Low containment = Frustrated customers - High self-service + Low containment = Incomplete solutions - High containment = True automation success ## Current Industry Performance ### By Technology Type **Rule-Based Chatbots** - Typical containment: 15-25% - Best-in-class: 30-35% - Limitation: Rigid decision trees **AI-Powered Assistants** - Typical containment: 35-50% - Best-in-class: 60-70% - Example: Sierra achieving 70% for retail clients **Hybrid Human-AI Systems** - Typical containment: 45-65% - Best-in-class: 75-85% - Key: Seamless handoff when needed ### By Industry Vertical **E-commerce/Retail** - Average: 55-65% - Leaders: 70-80% - Common issues: Order status, returns, shipping **Financial Services** - Average: 40-50% - Leaders: 60-70% - Challenge: Regulatory compliance, security **Technology/SaaS** - Average: 45-60% - Leaders: 70-75% - Advantage: Tech-savvy user base **Healthcare** - Average: 30-40% - Leaders: 50-60% - Challenge: Complexity, privacy requirements ## Measuring Containment Accurately ### Primary Calculation ``` Containment Rate = (Conversations ended in automation / Total automation conversations) × 100 ``` ### Quality-Adjusted Containment ``` True Containment = (Resolved without escalation AND positive CSAT / Total automated interactions) × 100 ``` ### Channel-Specific Formulas **Chatbot Containment**: ``` (Chats closed by bot with resolution / Total bot conversations) × 100 ``` **IVR Containment**: ``` (Calls completed in IVR / Total IVR calls) × 100 ``` **Email Automation**: ``` (Emails resolved by auto-response / Total automated emails) × 100 ``` ## The Containment Quality Spectrum ### Level 1: Basic Information Delivery (20-30%) - Provides relevant information - Customer must interpret and apply - No action taken by system - Example: "Here's how to reset your password" ### Level 2: Guided Problem Solving (40-50%) - Walks through solution steps - Confirms understanding - Provides alternatives if needed - Example: Interactive troubleshooting flow ### Level 3: Direct Action + Resolution (60-70%) - System takes action for customer - Processes requests end-to-end - Confirms successful completion - Example: Automated refund processing ### Level 4: Predictive Resolution (70-85%) - Anticipates follow-up needs - Handles edge cases gracefully - Personalizes based on history - Example: Proactive order issue resolution ## Critical Success Factors ### 1. Knowledge Foundation **Quality over Quantity** - Accurate, current information - Complete coverage of use cases - Regular updates and validation - Clear escalation triggers **Case Study**: Companies with manually curated knowledge bases show 15-27% higher containment than those relying on automated ingestion alone. ### 2. Natural Language Understanding **Beyond Keywords** - Intent recognition accuracy >95% - Context awareness across conversation - Handling of misspellings, slang, emotions - Multi-language support where needed ### 3. Integration Capabilities **End-to-End Resolution** - API access to backend systems - Ability to execute transactions - Real-time data retrieval - Secure authentication handling ### 4. Conversation Design **Human-Centric Flows** - Natural dialogue patterns - Clear confirmation steps - Graceful error handling - Personality without frustration ## Common Containment Killers ### 1. The Hallucination Problem - AI inventing policies or procedures - Confident delivery of wrong information - Lack of uncertainty expression - No verification mechanisms **Solution**: Implement guardrails, knowledge validation, and uncertainty thresholds ### 2. The Loop of Doom - Circular conversation paths - No escape to human help - Repeated failed attempts - Growing customer frustration **Solution**: Max attempt limits, clear escalation paths, frustration detection ### 3. The Partial Resolution Trap - Answering question asked, not problem faced - Missing follow-up issues - Incomplete action execution - False positive resolution **Solution**: Comprehensive issue mapping, post-resolution validation ### 4. The Channel Mismatch - Voice patterns in text chat - Desktop solutions for mobile users - Generic answers for specific contexts - One-size-fits-all approaches **Solution**: Channel-specific optimization and testing ## Advanced Containment Strategies ### 1. Adaptive Learning Systems - Continuous improvement from interactions - Pattern recognition for new issues - Automatic knowledge base updates - Performance optimization loops ### 2. Emotional Intelligence Integration - Sentiment analysis throughout conversation - Tone adjustment based on customer mood - Proactive de-escalation techniques - Empathy expression when appropriate ### 3. Predictive Containment - Identify high-risk interactions early - Route complex issues intelligently - Preemptive information gathering - Optimize for likely success ### 4. Hybrid Containment Models - AI handles routine, humans handle exceptions - Seamless mid-conversation handoffs - Agent-assist for complex containment - Blended resolution paths ## Case Studies in High Containment ### ADT: 70% Containment in Home Security **Challenge**: Complex technical and emergency scenarios **Solution**: - Deep integration with home systems - Extensive troubleshooting trees - Clear emergency escalation protocols **Result**: 70% containment with high satisfaction ### Decagon: 65-80% Across Industries **Approach**: - Industry-specific model training - Deep workflow integration - Continuous learning loops **Result**: Consistent high containment across diverse clients ### Financial Services Leader: 60% Complex Transaction Containment **Innovation**: - Secure authentication integration - Multi-step transaction handling - Regulatory compliance built-in **Result**: 60% containment on previously agent-only tasks ## Measuring Containment Success ### Immediate Metrics - Raw containment percentage - Containment by issue type - Time to resolution - Conversation length ### Quality Indicators - Post-interaction CSAT - Resolution verification rate - Repeat contact rate - Escalation patterns ### Business Impact - Cost per contained interaction - Team time saved - Revenue protected/generated - Customer lifetime value impact ### Long-term Trends - Containment rate trajectory - New issue adaptability - Knowledge base coverage - Technology ROI validation ## Best Practices for Maximizing Containment ### Design Principles ✓ Start with high-frequency, low-complexity issues ✓ Build trust through consistent success ✓ Always provide human escalation option ✓ Celebrate quality over quantity ✓ Design for the exception, not just the rule ### Operational Excellence ✓ Daily containment monitoring ✓ Weekly pattern analysis ✓ Monthly knowledge updates ✓ Quarterly strategic review ✓ Annual technology assessment ### Continuous Improvement ✓ A/B test conversation flows ✓ Analyze every escalation ✓ Interview customers who escalate ✓ Shadow agent interactions ✓ Benchmark against leaders ## The Future of Containment ### Emerging Capabilities - **Multi-modal interactions**: Voice, text, image, video - **Emotional AI**: Deeper empathy and connection - **Predictive personalization**: Individual-level optimization - **Autonomous improvement**: Self-optimizing systems ### Strategic Evolution From cost reduction to value creation: - **Revenue generation**: Upsell/cross-sell in contained interactions - **Relationship building**: Strengthen customer connection - **Insight gathering**: Learn from every interaction - **Innovation catalyst**: Discover new service opportunities ## Containment vs. Related Metrics - **[Self-Service Rate](/customer-support-metrics-and-kpis)**: Broader measure of independent success - **[Deflection Rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide)**: Focus on prevention vs. complete resolution ## Implementation Framework ### Phase 1: Foundation (Months 1-3) - Define containment criteria - Implement measurement systems - Baseline current performance - Identify quick wins ### Phase 2: Optimization (Months 4-6) - Expand automation scope - Enhance NLU capabilities - Integrate backend systems - Refine conversation flows ### Phase 3: Excellence (Months 7-12) - Advanced personalization - Predictive capabilities - Cross-channel orchestration - Continuous learning loops ## The Containment Mindset Success requires shifting from: - **Deflection → Resolution**: Not avoiding, but solving - **Efficiency → Effectiveness**: Quality over speed - **Cost → Value**: Investment in customer success - **Metrics → Meaning**: Real impact on real people ## Conclusion Containment rate is the truth-teller of customer service automation. It reveals whether technology truly serves customers or merely redirects them. High containment rates indicate not just efficient operations but genuine problem resolution. The path to high containment requires commitment to quality, continuous improvement, and customer-centricity. It's not about replacing human agents but augmenting the entire support ecosystem to serve customers better. When containment is done right, everyone wins: customers get fast, accurate resolutions; agents focus on meaningful, complex interactions; and businesses achieve sustainable operational excellence. --- _Build a comprehensive support strategy that balances all three critical metrics. [Return to the overview](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) to see how they work together._ --- ## Deflection Rate: The Complete Guide Source: https://en.helpfeel.com/customer-support-metrics-and-kpis/deflection-rate-complete-guide # Deflection Rate: The Complete Guide [← Back to Overview: Self-Service vs. Deflection vs. Containment](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) ## What is Deflection Rate? Deflection rate measures the percentage of support tickets or calls prevented from reaching human agents through proactive self-service interventions. It focuses on workload reduction and operational efficiency by intercepting issues before they become support cases. **Formula**: (Inquiries resolved before ticket creation / Total potential inquiries) × 100 ## The Strategic Value of Deflection ### Operational Impact - **Cost savings**: $5-15 saved per deflected ticket - **Capacity optimization**: Agents handle 30-40% more complex issues - **Response time improvement**: Shorter queues for remaining tickets - **Scalability**: Handle volume spikes without adding headcount ### Customer Experience Impact - **Instant resolution**: No waiting in queue - **Consistent quality**: Same answer every time - **Channel preference**: Meet customers where they want help - **Reduced effort**: Find answers without creating tickets ## Industry Benchmarks ### Current Performance Levels **Category 1: No Self-Service (40% of companies)** - Current deflection: 0% - Potential: 30-40% reduction in tickets - Example: SOLIA achieved 40% inquiry rate reduction **Category 2: Basic Self-Service (55% of companies)** - Current deflection: 10-30% - Industry average: 23% (tech sector) - Potential: Additional 20-40% improvement **Category 3: Optimized Self-Service (5% of companies)** - Current deflection: 50%+ - Best-in-class: 70%+ (with high quality) - Example: ADT achieving 70% containment ## Measuring Deflection Accurately ### Method 1: Contact Form Intercept ``` Deflection Rate = (Users who found answer in suggested articles / Users who started contact form) × 100 ``` ### Method 2: Search-to-Ticket Tracking ``` Deflection Rate = (Searches with no subsequent ticket / Total searches) × 100 ``` ### Method 3: Channel Shift Analysis ``` Deflection Rate = (Reduction in tickets after self-service launch / Baseline ticket volume) × 100 ``` ## Warning: The Vanity Metric Trap High deflection without quality resolution creates: - **Frustrated customers**: Forced to self-serve without success - **Hidden problems**: Issues unresolved but unmeasured - **Trust erosion**: Customers avoid support entirely - **False savings**: Short-term gains, long-term customer loss **Key Insight**: "Deflection often measures how many customers gave up, not how many succeeded." ## Effective Deflection Strategies ### 1. The Contact Form Intercept ("Chasing FAQ") **How it works**: 1. Customer starts filling support form 2. System analyzes their input in real-time 3. Suggests relevant articles before submission 4. Tracks if customer proceeds or resolves **Results**: 25-35% deflection rate typical ### 2. Email Auto-Reply Strategy **How it works**: 1. Customer sends email to support 2. Auto-reply includes relevant KB links 3. Ask: "Did this solve your issue?" 4. Only create ticket if customer confirms need **Results**: 20-30% deflection for email channel ### 3. Proactive Chat Interventions **How it works**: 1. Monitor user behavior patterns 2. Trigger chat with helpful content 3. Offer self-service before agent option 4. Escalate only if needed **Results**: 40-50% deflection for chat channel ### 4. In-Product Contextual Help **How it works**: 1. Detect error states or confusion signals 2. Surface relevant help content inline 3. Prevent support contact entirely 4. Track resolution without contact **Results**: 60-70% issue prevention ## Critical Success Factors ### Content Strategy - **Anticipatory writing**: Address issues before they're asked - **Multiple formats**: Text, video, interactive guides - **Scannable structure**: Users find answers in under 30 seconds - **Action-oriented**: Clear next steps, not just information ### Technology Requirements - **Real-time analysis**: Understand intent instantly - **Intelligent matching**: Connect questions to answers - **Seamless experience**: No friction between channels - **Comprehensive tracking**: Measure every interaction ### Organizational Alignment - **Cross-functional ownership**: Not just support's job - **Executive buy-in**: Investment in long-term success - **Continuous improvement**: Regular optimization cycles - **Cultural shift**: From reactive to proactive support ## Common Deflection Killers ### 1. Poor Search Experience - Generic results that don't match intent - No results for common questions - Buried search functionality - Slow or broken search ### 2. Low-Quality Content - Outdated information - Technical jargon - Incomplete solutions - No visual aids ### 3. Forced Deflection - Hiding contact options - Too many barriers to human help - Aggressive automation - No escalation path ### 4. Measurement Blindness - Tracking quantity not quality - Ignoring customer feedback - No follow-up validation - Missing channel connections ## Advanced Deflection Techniques ### Dynamic Content Suggestion - Machine learning ranks articles by likelihood to deflect - Personalization based on customer history - A/B testing different suggestion strategies - Continuous algorithm refinement ### Multi-Channel Orchestration - Consistent deflection across all touchpoints - Shared knowledge base powers all channels - Unified analytics across channels - Seamless handoffs when needed ### Predictive Deflection - Identify high-risk customers before they contact - Proactive outreach with helpful resources - Behavioral triggers for intervention - Issue prevention vs. reaction ## Case Studies ### Belluna: 50% Email Deflection - Implemented email auto-reply with KB suggestions - Achieved 12.5% overall deflection rate - Key: Rich email analysis and smart content matching ### RJ Young: Multi-Channel Success - Chat deflection: 44-45% - Web form deflection: 12-15% - Overall improvement: 20% ticket reduction - Key: Channel-specific optimization strategies ### Naviant: ROI-Focused Approach - Conservative scenario: 10% deflection = $116K/year saved - Moderate scenario: 20-25% deflection - Key: Clear financial modeling and tracking ## Measuring True Deflection Success ### Quantity Metrics - Deflection rate by channel - Ticket volume reduction - Cost per deflection - Agent capacity gained ### Quality Metrics - Resolution confirmation rate - Customer effort score - Follow-up contact rate - Satisfaction post-deflection ### Business Metrics - Cost savings achieved - ROI on self-service investment - Impact on customer lifetime value - Agent productivity improvement ## Best Practices for Sustainable Deflection ### Do's ✓ Start with high-volume, simple issues ✓ Test deflection strategies before full rollout ✓ Maintain easy escalation paths ✓ Celebrate quality over quantity ✓ Invest in continuous content improvement ### Don'ts ✗ Hide human support options ✗ Force customers through multiple barriers ✗ Measure deflection without resolution ✗ Neglect content maintenance ✗ Ignore customer feedback ## The Evolution of Deflection ### From Cost Center to Strategic Asset Modern deflection strategies focus on: - **Customer success**: Faster time to value - **Revenue impact**: Reduce churn, increase adoption - **Competitive advantage**: Superior support experience - **Data insights**: Learn from deflection patterns ### Future Trends - **AI-powered intent prediction**: Deflect before customers ask - **Omnichannel orchestration**: Seamless deflection everywhere - **Conversational deflection**: Natural language interactions - **Predictive content creation**: Generate articles for emerging issues ## Deflection Rate vs. Related Metrics - **[Self-Service Rate](/customer-support-metrics-and-kpis)**: Measures successful independent resolution - **[Containment Rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide)**: Focuses on complete automation without escalation ## Implementation Roadmap ### Phase 1: Foundation (Months 1-2) - Audit existing content - Implement basic tracking - Launch contact form intercept - Establish baseline metrics ### Phase 2: Optimization (Months 3-4) - Analyze deflection patterns - Expand to email channel - Refine content based on data - A/B test strategies ### Phase 3: Scale (Months 5-6) - Add chat and in-product help - Implement predictive deflection - Advanced personalization - Cross-channel orchestration ## Conclusion Deflection rate is a powerful metric when balanced with quality and customer satisfaction. The goal isn't to prevent all human contact but to help customers find fast, accurate answers when self-service is the best option. Success requires viewing deflection not as avoiding customers but as serving them more efficiently. When implemented thoughtfully, high deflection rates benefit both customers (faster resolution) and companies (reduced costs, happier agents, better insights). The key is remembering that behind every deflected ticket is a real person with a real problem. Measure success not just by tickets avoided but by problems solved. --- _Ready to implement a comprehensive support strategy? [Explore how all three metrics work together](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) to create exceptional customer experiences._ --- ## How to Reduce Support Ticket Volume with Searchable Self-Service Source: https://en.helpfeel.com/customer-support-metrics-and-kpis/reduce-support-ticket-volume # How to Reduce Support Ticket Volume with Searchable Self-Service Most support teams answer the same questions over and over. When customers can find those answers themselves, ticket volume drops, agents focus on harder problems, and satisfaction goes up. The lever is **findable** self-service: not just having documentation, but making it searchable the way people actually ask. ## Why most knowledge bases don't deflect tickets A knowledge base only deflects tickets when customers can locate the right answer in the words they already use. Traditional keyword search fails the moment a customer's phrasing doesn't match your article titles, so they give up and open a ticket anyway. - The answer exists, but search can't surface it - Customers phrase questions differently than your docs are written - Empty or near-miss search results push people straight to a ticket ## What actually moves the number Ticket deflection improves when search understands intent, not just exact strings. A few durable principles: 1. Index questions, not only article titles: customers search in questions. 2. Surface a confident answer fast, even for partial or fuzzy queries. 3. Close the loop: when search returns nothing useful, that gap is your next article. ## How to measure deflection You can't improve what you don't measure. Start with three signals: search queries with no useful result, the ratio of self-service sessions to tickets opened, and the share of tickets that repeat an already-answered question. Track them over time and let the gaps tell you what to write next. For a deeper look at searchable self-service, see [Helpfeel](https://en.helpfeel.com). --- ## Self-Service Rate: The Complete Guide Source: https://en.helpfeel.com/customer-support-metrics-and-kpis/self-service-rate-complete-guide # Self-Service Rate: The Complete Guide [← Back to Overview: Self-Service vs. Deflection vs. Containment](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) ## What is Self-Service Rate? Self-service rate measures the percentage of customers who successfully resolve their issues independently using available resources without needing to contact human support agents. It's the north star metric for customer empowerment and knowledge base effectiveness. **Formula**: (Successful self-service resolutions / Total customer inquiries) × 100 ## Why Self-Service Rate Matters ### For Customers - **24/7 availability**: Get answers anytime without waiting - **Immediate resolution**: No queue times or business hours restrictions - **Privacy and control**: Handle sensitive issues independently - **Consistency**: Same accurate information every time ### For Businesses - **Cost efficiency**: Each self-service resolution costs 90-95% less than agent-handled tickets - **Scalability**: Handle growth without proportional headcount increases - **Agent focus**: Free agents for complex, high-value interactions - **Customer satisfaction**: 69% of customers prefer self-service for simple issues ## Current Industry Benchmarks Based on analysis of 625+ Helpfeel implementations: ### By Industry - **E-commerce**: 65-80% (mature implementations) - **SaaS/Technology**: 45-60% - **Financial Services**: 40-55% - **Healthcare**: 35-50% - **Manufacturing**: 25-40% ### By Company Size - **Enterprise (1,000+ employees)**: 50-70% - **Mid-market (100-999)**: 40-60% - **SMB (under 100)**: 30-50% ## How to Calculate Self-Service Rate ### Method 1: Direct Measurement ``` Self-Service Rate = (KB views with no ticket creation / Total KB views) × 100 ``` ### Method 2: Inverse Calculation ``` Self-Service Rate = 100% - Contact Rate Where Contact Rate = (Support tickets / Active users) × 100 ``` ### Method 3: Resolution-Based ``` Self-Service Rate = (Self-resolved issues / Total issues) × 100 ``` ## Key Success Factors ### 1. Content Quality - **Accuracy**: Information must be correct and current - **Findability**: Content must be easily discoverable - **Readability**: Written in customer language, not internal jargon - **Completeness**: Cover all steps, edge cases, and outcomes ### 2. Search Effectiveness - **Natural language processing**: Handle how customers actually ask - **Typo tolerance**: Find content despite spelling errors - **Synonym matching**: Connect different terms for same concepts - **Intent recognition**: Understand what customers really need ### 3. User Experience - **Prominent placement**: Self-service options visible at key moments - **Mobile optimization**: 60%+ of searches happen on mobile - **Fast loading**: Every second of delay reduces usage by 7% - **Clear navigation**: Intuitive paths to relevant content ## Common Challenges and Solutions ### Challenge 1: Low Adoption **Symptoms**: Good content exists but customers don't use it **Solutions**: - Place search prominently on homepage and support pages - Add contextual help within product interfaces - Proactively suggest articles during support interactions - Market self-service options in customer communications ### Challenge 2: Poor Search Results **Symptoms**: High search abandonment, repeated searches **Solutions**: - Implement advanced search technology (like Helpfeel's fuzzy matching) - Analyze failed searches to identify content gaps - Add synonyms and alternate phrasings - Create content for common "no results" queries ### Challenge 3: Outdated Content **Symptoms**: High bounce rates, escalations after self-service attempts **Solutions**: - Establish regular content review cycles - Implement version control and change tracking - Create feedback mechanisms on every article - Assign content ownership to subject matter experts ## Advanced Strategies ### 1. The Email Intercept Strategy When customers email support, auto-reply with relevant KB articles based on their message content. Companies report 20-30% deflection from this approach alone. ### 2. Progressive Disclosure Start with simple solutions, then reveal advanced options. Prevents overwhelming new users while serving power users. ### 3. Contextual Pre-emptive Help Surface relevant articles based on user behavior before they seek help: - Error message? Show troubleshooting guide - First login? Display getting started content - Feature usage spike? Offer advanced tips ### 4. Multi-Channel Integration - **In-product**: Contextual help tooltips and sidebars - **Mobile app**: Native help center integration - **Chatbot**: KB-powered conversational interface - **Voice assistant**: Audio-friendly help content ## Measuring Success Beyond the Rate ### Leading Indicators - Search volume trends - Article view patterns - Search refinement rates - Time to resolution ### Quality Metrics - Article helpfulness ratings - Search success rate - Bounce/exit rates - Follow-up contact rate ### Business Impact - Cost per resolution - Customer effort score - First contact resolution - Customer lifetime value ## Case Studies ### MUJI: From 0 to 70% in 6 Months - Started with 270 migrated articles - Achieved 70% reduction in peak inquiries - 150% YoY increase in self-service users - Key: Comprehensive content migration and search optimization ### SOKUYAKU: 84% Inquiry Reduction - Inquiry rate dropped from 40% to 8% - Customer service costs reduced by 50% - Key: Analytics-driven content creation based on ticket patterns ### S.RIDE: 30% Rate Reduction Despite Growth - Maintained low inquiry rate despite rapid user growth - Key: Proactive FAQ updates aligned with new features ## Best Practices Checklist ✓ **Content Creation** - [ ] Write in customer language, not company jargon - [ ] Include visual aids (screenshots, videos, diagrams) - [ ] Cover common variations of each issue - [ ] Test content with real customers - [ ] Update within 24 hours of product changes ✓ **Search Optimization** - [ ] Implement fuzzy matching for typos - [ ] Add industry-specific synonyms - [ ] Track and optimize failed searches - [ ] A/B test search result rankings - [ ] Monitor search-to-resolution paths ✓ **Continuous Improvement** - [ ] Weekly review of top searches - [ ] Monthly content accuracy audits - [ ] Quarterly customer feedback surveys - [ ] Annual comprehensive content overhaul - [ ] Ongoing competitive benchmarking ## The Future of Self-Service ### Emerging Trends - **AI-powered content generation**: Automatically create articles from resolved tickets - **Predictive help**: Surface content before customers realize they need it - **Multimedia search**: Find help using images or voice descriptions - **Personalized experiences**: Tailored content based on user profile and history ### Strategic Evolution Self-service is evolving from cost reduction to revenue enablement: - **Onboarding acceleration**: Reduce time-to-value for new customers - **Feature adoption**: Drive usage of premium capabilities - **Upsell opportunities**: Educate about advanced features - **Retention improvement**: Prevent churn through proactive education ## Related Metrics - **[Deflection Rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide)**: Focuses on tickets prevented vs. issues resolved - **[Containment Rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide)**: Measures complete resolution within automated systems ## Conclusion Self-service rate is more than a support metric: it's a measure of customer empowerment and operational excellence. The most successful organizations view it not as a way to avoid customers but as a way to serve them better, faster, and more consistently. The key is balancing automation with authenticity, ensuring that self-service enhances rather than replaces the human elements of customer support. When done right, high self-service rates correlate with both reduced costs and increased customer satisfaction. --- _Want to improve your self-service rate? [Learn how the three metrics work together](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article) to create a comprehensive support strategy._ --- ## CX Glossary: Customer Experience Terms Defined Source: https://en.helpfeel.com/glossary # CX Glossary Clear, plain-language definitions of the customer experience, support, and self-service terms that come up when you run a modern help center. Each entry explains what the term means, how it is measured or used, and how it connects to the metrics that matter. Browse every term A to Z below. - [AI Agent](/glossary/ai-agent) - [Average Handle Time](/glossary/average-handle-time) - [Average Resolution Time](/glossary/average-resolution-time) - [Canned Response](/glossary/canned-response) - [Chatbot](/glossary/chatbot) - [Contact Rate](/glossary/contact-rate) - [Containment Rate](/glossary/containment-rate) - [Conversational AI](/glossary/conversational-ai) - [Cost Per Contact](/glossary/cost-per-contact) - [CSAT](/glossary/csat) - [Customer Effort Score](/glossary/customer-effort-score) - [Customer Lifetime Value](/glossary/customer-lifetime-value) - [Customer Self-Service](/glossary/customer-self-service) - [Customer Support vs Customer Service](/glossary/customer-support-vs-customer-service) - [Deflection Rate](/glossary/deflection-rate) - [Escalation Rate](/glossary/escalation-rate) - [First Contact Resolution](/glossary/first-contact-resolution) - [First Response Time](/glossary/first-response-time) - [Generative AI in Customer Service](/glossary/generative-ai-customer-service) - [Help Center](/glossary/help-center) - [Help Desk](/glossary/help-desk) - [Knowledge Base](/glossary/knowledge-base) - [Knowledge Management](/glossary/knowledge-management) - [NPS](/glossary/nps) - [Omnichannel Support](/glossary/omnichannel-support) - [Self-Service Rate](/glossary/self-service-rate) - [SLA](/glossary/sla) - [Support Ticket](/glossary/support-ticket) - [Ticket Volume](/glossary/ticket-volume) - [Tiered Support](/glossary/tiered-support) ## Frequently asked questions ### What is a CX glossary? A CX glossary is a reference list of customer experience terms with clear definitions. This one covers support metrics, self-service concepts, and the tools teams use, each with a plain-language explanation and, where it applies, the standard formula. ### Which customer support metrics matter most? The most-watched support metrics are self-service rate, first contact resolution, CSAT, and average resolution time. Together they show whether customers get fast answers and whether your knowledge base and team are working well. ### How is customer support different from customer service? Customer support solves specific technical or product issues. Customer service covers the whole customer relationship, from onboarding to billing. Support is a specialized part of the broader service experience. ## See it work Helpfeel turns these terms into results. We set up and tune a knowledge base so your customers find answers on their own, and your team handles only what truly needs a human. [See how the done-for-you model works](/platform). --- ## AI Agent Source: https://en.helpfeel.com/glossary/ai-agent # AI Agent An AI agent is software that autonomously handles customer inquiries from start to finish. Unlike a chatbot that follows scripted paths, an AI agent reasons through problems, searches your knowledge base, pulls data from your systems, and takes actions like updating an order or issuing a refund, all without human intervention. It is customer support that runs itself, resolving issues instantly at any hour. **Key point:** An AI agent is autonomous. It does not hand off to a human for every edge case. ## How does an AI agent work? AI agents combine three layers to handle support inquiries end-to-end: | Layer | What it does | | ------------- | ----------------------------------------------------------------------------------------------------------------------------------------- | | Understanding | Reads the customer's question in natural language and determines intent, context, and required actions. | | Reasoning | Decides the best path to a resolution, whether that means searching documentation, pulling account data, or running a workflow. | | Action | Executes tasks across your systems through APIs, such as updating a ticket, processing a return, or escalating only when truly necessary. | The difference from a chatbot: the agent decides what to do next based on the situation, not a decision tree you coded in advance. ## Why does an AI agent matter? AI agents resolve the majority of routine inquiries instantly, so customers never wait and your team never sees those tickets. They work 24/7, scale without adding headcount, and improve as they learn from real conversations. For most support teams, an AI agent is the path to handling growth without hiring linearly. For the full picture on when to deploy an AI agent and how it fits with your existing tools, read the [guide to AI customer support agents](/customer-support/ai-customer-support-agent). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Chatbot](/glossary/chatbot): rule-based conversational software that responds to keywords. - [Conversational AI](/glossary/conversational-ai): natural language systems that understand context and intent. ## Frequently asked questions ### What is the difference between an AI agent and a chatbot? A chatbot follows scripted paths and hands off complex issues. An AI agent reasons through problems, accesses real data, and takes actions like updating orders or searching knowledge bases, all autonomously without escalating to a human. ### Can an AI agent work with my existing tools? Yes. AI agents integrate with support platforms, knowledge bases, CRMs, and order systems through APIs. They pull data and take actions across your existing stack, so customers get complete answers in one conversation. ### Do AI agents replace human support teams? No. AI agents handle routine inquiries so your human team can focus on complex, high-touch cases. They are the next hire you will not need to make, augmenting your team rather than replacing it. ## See it work Helpfeel is built to deploy AI agents that answer real questions and resolve real issues. We set up and tune everything so it works on day one. [See how the done-for-you model works](/platform). --- ## Average Handle Time Source: https://en.helpfeel.com/glossary/average-handle-time # Average Handle Time Average Handle Time (AHT) is the total time an agent spends on a customer ticket, from the moment they pick it up to the moment they mark it resolved. It includes talk time, hold time, and any after-call work like notes or follow-up tasks. It is the most common metric for measuring agent efficiency, and it tells you how long each customer interaction takes from start to finish. **Formula:** (Total talk time + Total hold time + Total after-call work) / Total number of tickets ## How is average handle time calculated? Most support platforms calculate it for you, but the formula breaks into three parts: | Component | What it includes | | --------------- | ------------------------------------------------------------------------------------------------------------------------- | | Talk time | The minutes or seconds the agent actively speaks with or messages the customer | | Hold time | Time the customer waits while the agent researches, consults a colleague, or looks up an answer | | After-call work | Time spent writing notes, tagging the ticket, updating records, or sending follow-up messages after the conversation ends | Add those three for every ticket, then divide by the number of tickets. That gives you the average. ## Why does average handle time matter? AHT is a window into how efficiently your team operates. A lower AHT means agents can help more customers in a shift, which reduces cost per ticket and shortens the queue. But pushing AHT too low can backfire: if agents rush to hit a number, quality drops and customers leave unhappy. The healthiest path is to reduce AHT by removing friction, not by cutting corners. That means better knowledge bases, clearer internal processes, and self-service that keeps simple questions out of the queue. AHT is one piece of good [customer support metrics](/customer-support-metrics-and-kpis). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [First Response Time](/glossary/first-response-time): how long customers wait for the first reply. - [Average Resolution Time](/glossary/average-resolution-time): total time from ticket open to close. ## Frequently asked questions ### What is a good average handle time? It depends on your industry and ticket complexity. Simple ecommerce inquiries might average two to five minutes, while technical SaaS support can run fifteen to thirty minutes. The best benchmark is your own trend: if AHT rises while quality holds, that may signal process friction or tool gaps. ### How is average handle time calculated? Add total talk time, hold time, and after-call work, then divide by the number of tickets. The formula is (Talk time + Hold time + After-call work) / Total tickets. Most support platforms calculate it automatically. ### Should you try to reduce average handle time? Only if you can do it without sacrificing quality. Lowering AHT by rushing agents harms customer satisfaction. The smarter path is to reduce it by giving agents better tools, clearer processes, and self-service options that answer simple questions before they become tickets. ## See it work Helpfeel reduces average handle time naturally. When customers find answers on their own, simple tickets never reach your team. When agents do need to step in, a tuned knowledge base puts the right answer in their hands faster. [See how the done-for-you model works](/platform). --- ## Average Resolution Time Source: https://en.helpfeel.com/glossary/average-resolution-time # Average Resolution Time Average resolution time is the mean time it takes to fully resolve a customer issue, measured from the moment the customer first makes contact to the moment the issue is closed. It is one of the primary metrics for evaluating the speed and efficiency of a support operation. A lower average resolution time means customers get back to their day faster, and the support team moves through the queue more quickly. **Formula:** Total resolution time for all tickets / Number of resolved tickets ## How is average resolution time calculated? Most teams calculate it by channel, because resolution speed varies widely by medium: | Channel | Typical calculation | Example | | ------------- | ---------------------------------------------------------- | ------------------------------------------------------- | | Email | Total hours from first email to closure / Tickets resolved | 240 hours across 20 tickets = 12 hours average | | Chat or phone | Total minutes from start to resolution / Tickets resolved | 600 minutes across 40 tickets = 15 minutes average | | Self-service | Seconds from search to answer found / Successful sessions | 12,000 seconds across 200 sessions = 60 seconds average | Most support platforms calculate this automatically by tracking the timestamp of the first customer message and the final closure timestamp. ## Why does average resolution time matter? A short resolution time signals that your support operation gives customers their answer quickly, without long back-and-forth loops or days of waiting. It reflects both agent skill and the quality of the systems behind them: good routing, accessible knowledge, and clear escalation paths all shorten resolution time. It is part of good [customer support metrics](/customer-support-metrics-and-kpis). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Average handle time](/glossary/average-handle-time): active agent work time per ticket. - [First contact resolution](/glossary/first-contact-resolution): issues resolved in one interaction. ## Frequently asked questions ### What is a good average resolution time? It varies by channel and complexity. Email support averages 12 to 24 hours, chat and phone typically resolve in 10 to 30 minutes, and self-service aims for under 2 minutes. The best benchmark is your own baseline: improvement over time matters more than hitting an industry average. ### How is average resolution time different from average handle time? Average handle time measures the active work time an agent spends on a ticket. Average resolution time includes everything: agent work, customer wait time, and any delays between back-and-forth exchanges. Resolution time is always longer. ### How do you reduce average resolution time? Faster answers come from better knowledge bases, fewer escalations, and routing tickets to the right agent the first time. Self-service and AI tools handle simple issues instantly, freeing agents to focus on complex cases that need human judgment. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your customers find answers on their own. [See how the done-for-you model works](/platform). --- ## Canned Response Source: https://en.helpfeel.com/glossary/canned-response # Canned Response A canned response is a pre-written answer saved in your help desk or customer service software so support agents can reuse it across multiple tickets. The same reply, sent to many customers, without retyping it each time. Teams build libraries of canned responses for the questions they answer over and over: password resets, shipping timelines, return policies, account setup steps. **Purpose:** Save time on repetitive answers so agents can focus on the tickets that need a human touch. ## What are examples of canned responses? The most common templates handle predictable questions or workflows: | Use case | Example canned response | | ------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Password reset | "Hi [Name], I have sent a password reset link to your email. It will arrive in the next few minutes. If you do not see it, check your spam folder or let me know and I can resend it." | | Refund confirmation | "Your refund of [Amount] has been processed. It should appear in your account within 3 to 5 business days, depending on your bank. Let me know if you have questions." | | Feature request acknowledgment | "Thanks for the suggestion. I have added your request to our product roadmap tracker. Our team reviews these monthly, and I will follow up if we move forward with it." | | Escalation handoff | "I am looping in our [Team] team. They will reply to you within [Timeframe] with next steps. Thanks for your patience." | Good canned responses leave room for the agent to drop in the customer's name, ticket details, or a personal line before hitting send. ## Why do canned responses matter? They cut down the time agents spend retyping the same answers, which means faster replies for customers and more bandwidth for the team to solve the complex issues. A library of strong templates also ensures every customer gets a consistent, accurate answer, even when the team is onboarding new agents or handling a surge in tickets. Canned responses are part of good [customer service software](/customer-service-software). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Help desk](/glossary/help-desk): the system where canned responses live. - [Support ticket](/glossary/support-ticket): the conversation where a canned response gets sent. ## Frequently asked questions ### When should you use a canned response? Use them for repetitive questions where the answer stays the same: password resets, return policies, account activation steps, or shipping timelines. The best time to turn an answer into a template is the third time you type it. ### Do canned responses make support feel robotic? Only if you use them word-for-word every time. Good teams treat them as a starting draft, then personalize the greeting, reference the customer's specific issue, and adjust tone to match the conversation. ### How do you organize canned responses for a team? Tag them by category, topic, or workflow stage. Use short searchable names. Keep a shared library so new agents can find them fast. Archive outdated templates so the library stays clean and agents trust what they see. ## See it work Helpfeel answers the repetitive questions before they become tickets. Your team handles the work that needs a human, and customers get instant answers to everything else. [See how the done-for-you model works](/platform). --- ## Chatbot Source: https://en.helpfeel.com/glossary/chatbot # Chatbot A chatbot is an automated software program that simulates conversation with users through text or voice interfaces. It responds to questions, executes tasks, and guides users through processes without requiring a human agent. Chatbots range from simple rule-based systems that follow decision trees to sophisticated AI-powered assistants that understand natural language and learn from interactions. **Key point:** The effectiveness of a chatbot depends on its underlying intelligence, the quality of the knowledge base it draws from, and how gracefully it escalates to a human when needed. ## How does a chatbot work? The mechanism depends on the type: | Type | How it works | Best for | | ------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ | | Rule-based | Follows predefined scripts and decision trees. User selects from menus or keywords trigger canned responses. | Simple, high-volume FAQs with predictable paths. | | AI-powered (conversational AI) | Uses natural language processing and machine learning to understand intent, search a knowledge base, and generate contextual responses. | Complex questions, varied phrasing, and interactions that need to improve over time. | | Hybrid | Starts with AI to interpret intent, escalates to a human agent when confidence is low or the request is complex. | Customer service where speed and accuracy both matter. | Most modern customer service chatbots are hybrid: they use AI to handle common questions and seamlessly hand off to a human when the conversation exceeds their capability. ## Why does a chatbot matter? A well-built chatbot gives customers instant answers at any hour, without a wait in the queue. For businesses, it frees support teams from repetitive questions so they can focus on issues that actually need human judgment. Done well, it is a natural extension of the self-service experience, part of a good [AI help center](/ai-help-center) that blends search, chat, and human escalation into one seamless flow. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Conversational AI](/glossary/conversational-ai): AI systems that understand and generate natural conversation. - [Generative AI customer service](/glossary/generative-ai-customer-service): AI that creates original responses in real time. ## Frequently asked questions ### What is the difference between a chatbot and conversational AI? A chatbot follows scripted rules or decision trees. Conversational AI uses machine learning to understand intent, learn from interactions, and generate natural responses. Modern conversational AI can handle ambiguous questions and improve over time, while traditional chatbots are limited to the paths their rules define. ### What are common uses for chatbots in customer service? Chatbots handle FAQs, order status lookups, account changes, booking appointments, and routing complex questions to human agents. They work best for high-volume, predictable questions where the correct answer is clear and the customer needs a quick reply. ### Do chatbots replace human customer service agents? No. Chatbots handle repetitive, predictable questions so human agents can focus on complex issues that need judgment, empathy, or creative problem solving. The best implementations use chatbots as the first line that escalates when needed, not as a replacement. ## See it work Helpfeel is built to give your customers instant, accurate answers, whether they search or chat. We set up and tune the system so it works from day one. [See how the done-for-you model works](/platform). --- ## Contact Rate Source: https://en.helpfeel.com/glossary/contact-rate # Contact Rate Contact rate is the percentage of customers who reach out to support for help instead of resolving their issue through self-service. It measures the proportion of your customer base that opens a ticket, starts a chat, or calls in. A lower contact rate means more customers are finding answers on their own, which reduces the load on your support team and speeds up resolutions for everyone. **Formula:** (Number of support tickets / Total active users or sessions) × 100 ## How is contact rate calculated? The standard method depends on your business model and data: | Method | Calculation | | ------------- | ------------------------------------------------------------------------------------------- | | User-based | (Tickets opened / Active users in period) × 100 | | Session-based | (Tickets opened / Total sessions in period) × 100 | | Inverse | 100% − self-service rate, where self-service rate = (Self-resolved / Total inquiries) × 100 | User-based is the most common because it ties the metric directly to customer behavior across a week or month. ## Why does contact rate matter? Contact rate is the inverse view of self-service rate. When it climbs, your support queue grows. When it falls, more customers are getting instant answers without waiting in line. A declining contact rate means your knowledge base and search are doing their job, freeing your team to focus on the complex or sensitive conversations that genuinely need a human. It is part of good [customer support metrics](/customer-support-metrics-and-kpis) to track how many customers you are serving without human intervention. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Self-service rate](/glossary/self-service-rate): percentage who resolve issues on their own. - [Ticket volume](/glossary/ticket-volume): total number of support requests. ## Frequently asked questions ### What is a good contact rate? It varies by industry and business model. SaaS and ecommerce companies typically target 20 to 40 percent. The lower the better, but rapid drops can signal confused customers skipping support entirely. A healthy contact rate is one that declines steadily as your self-service improves without abandoning genuine help seekers. ### How is contact rate different from self-service rate? Contact rate measures the percentage of customers who reach out for help. Self-service rate measures those who resolve issues on their own. They are inverse metrics: when self-service rate climbs, contact rate falls. Together they show whether your help center is answering real questions. ### How do you calculate contact rate? Divide the number of support tickets by total active users or sessions in the same period, then multiply by 100. The result is the percentage of customers who needed help from a human agent. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your customers find answers on their own, lowering contact rate while keeping genuine questions answered. [See how the done-for-you model works](/platform). --- ## Containment Rate Source: https://en.helpfeel.com/glossary/containment-rate # Containment Rate Containment rate is the percentage of customer conversations that are fully resolved inside an automated system, such as a chatbot or interactive help center, without escalating to a human agent. It measures how well your automated channels answer and close customer issues on their own. A high containment rate means your automation is handling the routine work so your team can focus on the complex cases. **Formula:** (Conversations resolved by automation / Total conversations in the automated channel) × 100 ## How is containment rate calculated? Most teams measure it by tracking conversations that enter an automated channel and comparing how many end without escalation: | Metric | Definition | | ---------------------- | ----------------------------------------------------------------------------------------------------------------------- | | Total conversations | All customer interactions that start in the automated channel (chatbot, virtual assistant, or search-first help center) | | Resolved conversations | Interactions that end without the customer requesting a human agent or opening a support ticket | | Containment rate | (Resolved / Total) × 100 | A contained conversation is one where the customer gets the answer they need and leaves satisfied. The system tracks whether the session ended naturally or whether the customer clicked "contact support" or asked to speak to someone. ## Why does containment rate matter? A high containment rate means your automated systems are doing the work they were built for: answering routine questions instantly so your team has time for the conversations that need judgment, empathy, or deep product knowledge. It tells you whether your knowledge base and automation are good enough to close the loop without a human. For benchmarks, strategies to improve this metric, and how containment fits into the full support efficiency picture, read the [complete guide to containment rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Deflection rate](/glossary/deflection-rate): tickets prevented before they reach an agent. - [Self-service rate](/glossary/self-service-rate): customers who resolve issues on their own. ## Frequently asked questions ### What is a good containment rate? It depends on the complexity of your product and the strength of your automated systems. Many teams reach 50 to 70 percent containment when their knowledge base and chatbot are well tuned. The goal is not to contain every conversation, but to resolve the routine ones so your team can focus on the complex cases. ### How is containment rate different from deflection rate? Containment rate measures full resolution inside an automated system. Deflection rate measures tickets prevented before they reach a human. Deflection is broader: a customer who finds an answer and leaves may never have intended to contact you. Containment assumes the customer entered your system and the automation kept them there. ### How do you calculate containment rate? Divide the number of conversations fully resolved by automation by the total number of conversations that entered the automated channel, then multiply by 100. A contained conversation is one that ends without escalation to a human agent. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your automated channels resolve more on their own. [See how the done-for-you model works](/platform). --- ## Conversational AI Source: https://en.helpfeel.com/glossary/conversational-ai # Conversational AI Conversational AI is technology that uses natural language processing and machine learning to understand, interpret, and respond to customer questions in real time. Instead of relying on fixed scripts or keyword matching, it generates answers based on context, intent, and the content available to it. In customer service, conversational AI powers chatbots, help center search, voice assistants, and email support tools. **Key point:** It is not one product. It is the language-understanding layer that makes self-service tools feel like a conversation instead of a keyword hunt. ## How does conversational AI work? Conversational AI combines three capabilities to turn a customer question into an answer: | Step | What happens | | ------------- | ------------------------------------------------------------------------------------------------------------------------------------ | | Understanding | Natural language processing interprets the intent behind the question, even if it is phrased casually, has typos, or lacks keywords. | | Reasoning | The system searches your knowledge base, product data, or help center content to find the relevant answer. | | Responding | It generates a clear, context-aware reply, either surfacing an article, drafting a message, or routing to the right team. | Most implementations today use large language models as the reasoning and generation engine, which is why conversational AI has become synonymous with AI chat and AI search. ## Why does conversational AI matter? Customers type questions the way they think, not the way your documentation is titled. Conversational AI closes that gap. It handles routine questions instantly, 24 hours a day, and gives your team the space to focus on the complex conversations that actually require judgment or empathy. For the customer, it means an instant answer instead of a wait in the queue. For a full breakdown of how conversational AI fits into a modern support stack, the tradeoffs between chat and search interfaces, and the setup that actually works, read the [guide to conversational AI for customer service](/ai-help-center/conversational-ai-for-customer-service). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Chatbot](/glossary/chatbot): automated tool that simulates conversation with customers. - [AI agent](/glossary/ai-agent): system that takes actions on behalf of a user. ## Frequently asked questions ### What is the difference between conversational AI and a chatbot? A chatbot follows scripted paths and keyword matching. Conversational AI uses language models to understand intent, handle nuance, and generate responses. Most modern chatbots now incorporate conversational AI, which is why the line between the two has blurred. ### What are examples of conversational AI in customer service? AI chat assistants that answer product questions, help center search that understands natural language, voice assistants that route calls, and email assistants that suggest replies. All of them process a customer question and respond conversationally rather than matching keywords. ### Is conversational AI replacing human support agents? No. It handles routine questions so agents can focus on complex issues that require judgment, empathy, or problem solving. The goal is not replacement, it is giving your team time back for the conversations that actually need a human. ## See it work Helpfeel is built to answer customer questions the moment they are asked. We set up the conversational AI layer and tune it against your content so your customers get instant answers. [See how the done-for-you model works](/platform). --- ## Cost Per Contact Source: https://en.helpfeel.com/glossary/cost-per-contact # Cost Per Contact Cost per contact is the average expense your support team incurs to handle one customer inquiry, whether by email, chat, phone, or ticket. It includes agent labor, software, infrastructure, and overhead, divided by the total number of contacts in a period. Teams track this metric to understand the economics of their support operations and to identify where automation or better self-service can lower costs without compromising quality. **Formula:** Total support costs / Total number of contacts ## How is cost per contact calculated? The calculation is straightforward, but the inputs vary by how you count costs and contacts: | Component | What to include | | ------------------- | -------------------------------------------------------------------------------------------------- | | Total support costs | Agent salaries and benefits, software subscriptions, infrastructure, training, management overhead | | Total contacts | All tickets, calls, chats, and emails handled in the same period | | Calculation | Divide total support costs by total contacts to get the average cost per contact | Most teams calculate this monthly or quarterly. The more accurate your cost accounting, the more useful the metric becomes for decisions about staffing, tooling, and automation. ## Why does cost per contact matter? A lower cost per contact means your support operation is more efficient. For businesses with high ticket volume, even a small drop in cost per contact can save thousands of dollars each month. The goal is not just to cut costs, but to shift expensive human time away from repetitive, low-value questions and toward the complex conversations where an agent genuinely makes a difference. Self-service, automation, and a strong knowledge base all help move this number in the right direction. This metric is one part of good [customer support metrics](/customer-support-metrics-and-kpis). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Contact rate](/glossary/contact-rate): percentage of customers who reach out to support. - [Customer lifetime value](/glossary/customer-lifetime-value): total revenue from one customer over time. ## Frequently asked questions ### What is a good cost per contact? It varies by channel and industry. Email support typically costs five to ten dollars per contact, chat costs three to eight dollars, and phone support can run ten to twenty dollars or more. Reducing cost per contact matters, but never at the expense of the customer experience. ### How is cost per contact different from cost per resolution? Cost per contact measures the expense of handling one inquiry, regardless of outcome. Cost per resolution measures the expense of fully solving one issue, which may span multiple contacts. Resolution-based metrics give a fuller picture of efficiency. ### How do you calculate cost per contact? Divide your total support costs by the total number of customer contacts in the same period. Include agent salaries, software, infrastructure, and overhead, then divide by tickets, calls, chats, or emails handled. ## See it work Helpfeel is built to lower your cost per contact by giving customers instant answers before they reach your team. [See how the done-for-you model works](/platform). --- ## CSAT Source: https://en.helpfeel.com/glossary/csat # CSAT CSAT, or Customer Satisfaction Score, is a metric that measures how satisfied customers are with a specific product, service, or interaction. It is collected through a simple survey question, usually asking customers to rate their satisfaction on a scale from 1 to 5. CSAT is the most direct way to capture how a customer feels at a specific moment. **Formula:** (Number of satisfied customers / Total responses) × 100 ## How is CSAT calculated? The most common method is to ask customers to rate their satisfaction on a 5-point scale, then calculate the percentage who chose the top two scores (4 or 5): | Step | Action | | --------------- | ----------------------------------------------------------------------------------- | | Survey | Ask: "How satisfied are you with [your purchase / this interaction / our service]?" | | Scale | 1 = Very Unsatisfied, 5 = Very Satisfied | | Count satisfied | Count responses rated 4 or 5 | | Calculate | (Satisfied responses / Total responses) × 100 | For example, if 80 out of 100 customers rate 4 or 5, your CSAT is 80 percent. ## Why does CSAT matter? CSAT gives you a fast, clear signal of whether customers are happy at a specific touchpoint. It is the only metric that directly asks the customer how they feel, which makes it valuable for pinpointing problems in a purchase flow, a support interaction, or a product feature. It is part of good [customer support metrics](/customer-support-metrics-and-kpis) that tell you where experience breaks and where it works. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [NPS](/glossary/nps): measures loyalty and likelihood to recommend. - [Customer Effort Score](/glossary/customer-effort-score): measures how easy an interaction was. ## Frequently asked questions ### What is a good CSAT score? It varies by industry. Retail and ecommerce typically see scores of 75 to 85 percent, while software and financial services land in the 70 to 80 percent range. Track the trend over time to see whether satisfaction is rising or falling. ### How is CSAT different from NPS? CSAT measures satisfaction with a specific interaction or transaction. NPS measures overall loyalty and the likelihood a customer will recommend your brand. CSAT is transactional, NPS is relational. ### When should you send a CSAT survey? Send it immediately after the interaction you want to measure: post-purchase, after a support ticket closes, or after a feature interaction. The closer to the moment, the more accurate the response. ## See it work Helpfeel is built to improve the scores that matter. We set up and tune your help center so customers get answers instantly, which makes every interaction smoother. [See how the done-for-you model works](/platform). --- ## Customer Effort Score Source: https://en.helpfeel.com/glossary/customer-effort-score # Customer Effort Score Customer Effort Score (CES) is a metric that measures how easy customers find it to interact with your company. It asks customers to rate the effort required to resolve an issue, complete a transaction, or get a question answered. The idea is simple: low-effort experiences drive loyalty and repeat business, while high-effort interactions push customers away. Companies track CES to identify and remove friction from the customer journey. **Formula:** Ask "How easy was it to [solve your issue / complete your task]?" on a scale from 1 (very difficult) to 7 (very easy), or as agree/disagree with "The company made it easy for me to handle my issue." ## How is Customer Effort Score calculated? CES is measured with a single-question survey, sent immediately after a customer interaction. The two most common formats are: | Format | Method | Calculation | | -------------- | ---------------------------------------------------------------------------------------------------- | ----------------------------------------------- | | 7-point scale | "On a scale of 1 to 7, how easy was it to [task]?" | Average score, or % scoring 5 to 7 (low effort) | | Agree/disagree | "The company made it easy for me to handle my issue." (5-point: strongly disagree to strongly agree) | % who agree or strongly agree | Some teams track CES as an average (mean score). Others report it as the percentage of customers who found the experience easy (scoring in the top 2 or 3 points of the scale). The second method is clearer for benchmarking. ## Why does Customer Effort Score matter? CES predicts customer loyalty more reliably than satisfaction alone. Research shows that customers who have low-effort experiences are more likely to renew, buy again, and speak positively about your company. When customers have to repeat themselves, wait on hold, or dig through a confusing help center, they remember the friction. Reducing that effort is one of the most direct levers you have to improve retention and word of mouth. CES is part of good [customer support metrics](/customer-support-metrics-and-kpis) and helps you identify where the experience breaks down so you can fix it. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [CSAT](/glossary/csat): satisfaction with a specific interaction. - [NPS](/glossary/nps): loyalty and likelihood to recommend. ## Frequently asked questions ### What is a good Customer Effort Score? Most companies aim for a CES above 5 on a 7-point scale, or above 70 percent agreement on the statement-based format. Higher scores mean customers find interactions easier. What matters most is the trend: steady improvement means you are removing friction from the experience. ### How is Customer Effort Score different from CSAT or NPS? CSAT measures satisfaction with a specific interaction. NPS measures loyalty and willingness to recommend. CES measures effort: how hard customers had to work to get what they needed. Low effort predicts retention better than high satisfaction alone. ### When should you measure Customer Effort Score? Measure CES immediately after a service interaction, like closing a support ticket, completing a return, or finishing onboarding. The question is most accurate when customers can still recall how easy or hard the interaction felt. ## See it work Helpfeel is built to reduce customer effort. We set up and tune your knowledge base so customers find answers instantly, without hunting or waiting. [See how the done-for-you model works](/platform). --- ## Customer Lifetime Value Source: https://en.helpfeel.com/glossary/customer-lifetime-value # Customer Lifetime Value Customer lifetime value is the total revenue a single customer generates over the entire span of their relationship with your company. It is a forward-looking measure of customer worth, combining purchase behavior, retention, and profitability into one number. A higher CLV means each customer relationship is more valuable, which justifies spending more to acquire and keep them. **Formula:** Average purchase value × Purchase frequency × Average customer lifespan ## How is customer lifetime value calculated? The standard formula has three inputs: | Input | Definition | | ------------------------- | --------------------------------------------------------------------- | | Average purchase value | Total revenue / Number of purchases over a period | | Purchase frequency | Number of purchases / Number of unique customers over the same period | | Average customer lifespan | Average number of years a customer stays active | Multiply all three together to get CLV. For example, if a customer spends 50 dollars per order, orders twice a year, and stays for five years, the CLV is 50 × 2 × 5 = 500 dollars. ## Why does customer lifetime value matter? CLV tells you how much a customer relationship is actually worth, which informs how much you can afford to spend acquiring and supporting them. It shifts the focus from short-term conversions to long-term retention and upsell. For support teams, it makes the case for investing in better experiences: every improvement that keeps a customer around longer, or that encourages repeat purchases, raises CLV. It is part of good [customer support metrics](/customer-support-metrics-and-kpis) that tie the help desk to revenue, not just ticket counts. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [CSAT](/glossary/csat): customer satisfaction score, a leading indicator of retention. - [Cost per contact](/glossary/cost-per-contact): what each support interaction costs to deliver. ## Frequently asked questions ### What is a good customer lifetime value? It depends on your industry and average order value. What matters is the ratio: CLV should be at least three times higher than customer acquisition cost. A rising CLV over time signals stronger retention, more repeat purchases, or successful upsells. ### How is customer lifetime value calculated? Multiply average purchase value by purchase frequency, then multiply by average customer lifespan. For example, if a customer spends 50 dollars per order, orders twice a year, and stays for five years, CLV is 500 dollars. ### Why does customer lifetime value matter for support teams? Support quality directly affects retention and repeat purchases. A single bad experience can cut CLV short. High-CLV customers justify more investment in personalized support, while efficient self-service scales help for the broader base. ## See it work Helpfeel is built to keep CLV growing. We set up self-service and agent tools that answer customers fast, so they stay longer and buy more. [See how the done-for-you model works](/platform). --- ## Customer Self-Service Source: https://en.helpfeel.com/glossary/customer-self-service # Customer Self-Service Customer self-service is any system, tool, or resource that lets people find answers, solve issues, and complete tasks on their own, without contacting a human agent. The most common forms are searchable knowledge bases, help centers, FAQs, and chatbots. A customer who finds a shipping policy in your help center, watches a how-to video, or uses a search bar to troubleshoot a problem is using self-service. **Key point:** Self-service is not about avoiding customers. It is about giving them the fastest path to an answer, on their terms. ## What are examples of customer self-service? Customer self-service shows up across every channel and industry. Here are the most common forms: | Type | Example | | --------------- | ----------------------------------------------------------------------------- | | Knowledge base | Searchable help center with articles, guides, and troubleshooting steps | | FAQ page | Structured list of common questions and answers, often linked from the footer | | Chatbot | AI or rule-based bot that answers questions in real time | | Community forum | Customer-to-customer discussion space, often moderated by staff | | Video tutorials | Step-by-step walkthroughs, how-to videos, or product demos | | Order tracking | Self-service portal where customers check order status without calling in | The best self-service systems combine search, structure, and context. A customer should be able to type a question in natural language and land on the right answer in seconds. ## Why does customer self-service matter? Customers prefer instant answers over waiting in a queue, especially for simple, repeatable questions. A strong self-service system gives them that speed while freeing your team to focus on the conversations that actually need a human touch. It also scales without adding headcount, which matters when your customer base grows faster than your support budget. For the full playbook on building customer self-service that works, read the [guide to customer self-service](/ai-help-center/customer-self-service). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Self-service rate](/glossary/self-service-rate): measures how many customers resolve issues on their own. - [Help center](/glossary/help-center): the central hub for self-service content. ## Frequently asked questions ### What are examples of customer self-service? Common self-service tools include searchable knowledge bases, FAQs, video tutorials, chatbots, and community forums. Ecommerce order tracking pages and SaaS in-app help widgets are also self-service. The defining feature is that customers find the answer or complete the task themselves. ### Why is customer self-service important? Self-service gives customers instant answers at any hour, without waiting for an agent. For businesses, it reduces support costs, frees agents for complex issues, and scales support without adding headcount. The best self-service systems also improve over time as they learn what customers ask. ### How do you measure customer self-service success? Common metrics include self-service rate, which shows the percentage of customers who resolve their own issues, and time to answer. Tracking knowledge base search queries and article views also reveals what customers need most. ## See it work Helpfeel is built to deliver fast, accurate self-service answers. We set up and tune the knowledge base so your customers find what they need on their own. [See how the done-for-you model works](/platform). --- ## Customer Support vs Customer Service Source: https://en.helpfeel.com/glossary/customer-support-vs-customer-service # Customer Support vs Customer Service Customer support is the function that helps customers resolve specific technical or product issues. Customer service is the broader experience covering every touchpoint a customer has with your company, from the first sale through onboarding, renewals, and complaints. Support is a specialized part of service, focused on problem-solving. Service includes everything from answering general questions to managing relationships. **Key point:** Support is reactive and technical. Service is proactive and relational. ## How do customer support and customer service differ? The two functions overlap, but they have different goals and tools: | Dimension | Customer Support | Customer Service | | ------------ | ----------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | | **Focus** | Fix a specific issue or answer a product question | Manage the full customer relationship | | **Timing** | Reactive, after a problem surfaces | Proactive and reactive throughout the journey | | **Examples** | Troubleshooting login errors, resetting passwords, explaining a feature | Onboarding new users, processing returns, handling billing questions, gathering feedback | | **Channels** | Help center, chat, support tickets, knowledge base | Email, phone, in-person, social media, account management | In practice, many teams blend both. A support agent answering a ticket might also ask how onboarding went. A service rep might troubleshoot a billing issue. The distinction matters most in large organizations where specialization makes sense. ## Why does the distinction matter? The words you choose shape how customers find help and how your team is structured. Using the right term improves clarity and sets expectations. If your site says "customer service" but your team only handles technical issues, customers will be confused. If you label everything "support" when you also manage relationships, you undersell what you do. Understanding the difference also helps you decide what to automate and what stays human. Technical support questions lend themselves to self-service and AI resolution. Relational service moments, like a frustrated customer calling about a delayed order, need empathy and judgment. Both are part of good [customer support](/customer-support). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Support ticket](/glossary/support-ticket): the primary unit of work in customer support. - [Help desk](/glossary/help-desk): the system and team that manage customer support requests. ## Frequently asked questions ### What is the main difference between customer support and customer service? Customer support focuses on solving specific technical or product issues. Customer service is broader, covering every interaction a customer has with your company, from onboarding to returns to renewals. Support is a function within service. ### Does my team need both customer support and customer service? Most companies do not separate them formally. Small and mid-size teams call the function customer support or customer service and train for both. Larger organizations may have specialized support teams for technical issues and service teams for general inquiries. ### Which term should I use on my website or help center? Customer support is more common in tech and SaaS, especially when you are helping users troubleshoot. Customer service fits better for ecommerce, hospitality, and consumer brands. Use the one your customers already search for. ## See it work Helpfeel handles both support and service questions with a single search experience. Customers get instant answers whether they need technical help or general guidance. [See how the done-for-you model works](/platform). --- ## Deflection Rate Source: https://en.helpfeel.com/glossary/deflection-rate # Deflection Rate Deflection rate is the percentage of potential support tickets that are prevented before they reach a human agent. It measures how many customers find the answer they need through a help center, knowledge base, or automated system, without creating a ticket. A higher deflection rate means fewer tickets in the queue and more customers getting instant answers. **Formula:** (Tickets prevented / Total potential tickets) × 100 ## How is deflection rate calculated? Most teams measure it one of three ways, depending on the data available: | Method | Calculation | | ------------ | ------------------------------------------------------------------------------------------------------ | | Direct | (Help center visits with no ticket created / Total visits) × 100 | | Inverse | 100% − ticket creation rate, where ticket creation rate = (tickets created / help center visits) × 100 | | Ticket-based | (Baseline tickets − current tickets / Baseline tickets) × 100 | The direct method is the most common because it ties the metric straight to help center behavior. ## Why does deflection rate matter? A high deflection rate means customers get answers instantly, at any hour, without waiting for an agent. For the business, each prevented ticket frees your team for the complex issues that genuinely need a human. It is one of the clearest signals that your self-service resources are working. For benchmarks, calculation nuances, and strategies to move the number, read the [complete guide to deflection rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Self-service rate](/glossary/self-service-rate): customers who resolve issues on their own. - [Containment rate](/glossary/containment-rate): issues fully resolved inside an automated system. ## Frequently asked questions ### What is a good deflection rate? It varies by industry and maturity. Mature self-service programs reach 50 to 70 percent, meaning half or more of potential tickets are resolved before they reach an agent. The trend matters more than the absolute number: a rising rate means your knowledge base is improving. ### How is deflection rate different from self-service rate? Deflection rate measures tickets prevented. Self-service rate measures issues customers resolve on their own. They overlap, but deflection focuses on what never arrives in the queue, while self-service focuses on the customer succeeding independently. ### How do you calculate deflection rate? Divide tickets prevented by total potential tickets, then multiply by 100. In practice, teams measure it as knowledge base visits that did not create a ticket, or as 100 percent minus the ticket creation rate. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your customers find answers before they need to contact support. [See how the done-for-you model works](/platform). --- ## Escalation Rate Source: https://en.helpfeel.com/glossary/escalation-rate # Escalation Rate Escalation rate is the percentage of support tickets that must be passed from the initial responding agent or tier to a higher tier, specialist team, or manager for resolution. It measures how often your frontline cannot close an issue on their own. A lower rate usually means your first responders have the right knowledge, tools, and authority to solve most customer problems without handing them off. **Formula:** (Number of escalated tickets / Total tickets resolved) × 100 ## How is escalation rate calculated? Most teams measure escalation rate by tracking ticket handoffs in their support platform: | Method | Calculation | | ----------- | ----------------------------------------------------------------- | | Standard | (Escalated tickets / Total tickets resolved) × 100 | | By tier | (Tickets escalated from tier 1 / Tier 1 tickets handled) × 100 | | By category | (Escalations in a topic area / Total tickets in that topic) × 100 | The standard method gives you the overall health number. Breaking it down by tier or category helps you spot which knowledge gaps or routing rules need attention. ## Why does escalation rate matter? A stable escalation rate means your frontline team has the knowledge and authority to handle the majority of issues. A rising rate signals training gaps, unclear escalation rules, or product complexity outpacing your team's expertise. Tracking it over time helps you decide where to invest in documentation, training, or process changes. For a deeper look at escalation alongside the full set of [customer support metrics](/customer-support-metrics-and-kpis), connect it to resolution time and first-contact resolution. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [First-contact resolution](/glossary/first-contact-resolution): issues solved in the first reply. - [Tiered support](/glossary/tiered-support): organizing teams by skill and complexity. ## Frequently asked questions ### What is a good escalation rate? It varies by team structure and product complexity. Many organizations target 5 to 15 percent, but what matters most is stability: sudden spikes signal a knowledge gap, new product issue, or misrouted tickets that need attention. ### How is escalation rate calculated? Divide the number of escalated tickets by total tickets resolved, then multiply by 100. Track it by time period (weekly, monthly) and by escalation path to spot patterns and training opportunities. ### Why does escalation rate matter? A stable escalation rate means your frontline team has the knowledge and authority to resolve most issues. A rising rate signals training gaps, routing problems, or product complexity creeping beyond what tier one can handle. ## See it work Helpfeel reduces unnecessary escalations by giving every agent and every customer the same instant, accurate answers. [See how the done-for-you model works](/platform). --- ## First Contact Resolution Source: https://en.helpfeel.com/glossary/first-contact-resolution # First Contact Resolution First Contact Resolution (FCR) is the percentage of customer issues that are resolved on the first interaction with support, without requiring a follow-up contact. It is the core measure of how well your support team answers questions right away. A higher FCR means customers get their problem solved faster, and your team spends less time on repeat contacts for the same issue. **Formula:** (Issues resolved on first contact / Total issues) × 100 ## How is first contact resolution calculated? Most teams measure FCR by tracking whether a customer contacts support again about the same issue within a set timeframe, usually 7 days: | Method | Calculation | | -------------------- | ------------------------------------------------------------------------- | | Direct | (Tickets closed with no follow-up / Total tickets) × 100 | | Survey-based | Percentage of customers who say their issue was resolved on first contact | | Repeat contact check | 100% minus the percentage of customers who recontacted within 7 days | The direct method is the most common because it ties the metric to actual ticket data. ## Why does first contact resolution matter? A high FCR means customers get their issue solved immediately, with no back-and-forth or repeat contacts. That improves satisfaction and reduces frustration. For the team, each first-contact resolution is one less follow-up ticket, which frees agents to handle new issues instead of revisiting old ones. It is one of the strongest predictors of customer satisfaction and a sign that your knowledge base, agent training, and escalation paths are working together. For more context on how FCR fits with other metrics like CSAT and average resolution time, see the guide to [customer support metrics](/customer-support-metrics-and-kpis). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [CSAT](/glossary/csat): customer satisfaction score after a support interaction. - [Average resolution time](/glossary/average-resolution-time): how long it takes to close a ticket. ## Frequently asked questions ### What is a good first contact resolution rate? Industry benchmarks range from 65 to 75 percent. What matters most is the trend: a rising FCR means your knowledge base and agent training are getting better at solving issues right away, which improves customer satisfaction and reduces repeat contacts. ### How is FCR different from average resolution time? FCR measures whether an issue is solved in one contact, regardless of how long that contact takes. Average resolution time measures how long it takes to close a ticket. A high FCR and a low resolution time together signal efficient, effective support. ### How do you improve first contact resolution? Give agents better knowledge bases, smarter search, and clear escalation paths. Use self-service to catch simple questions before they reach an agent. Review repeat-contact tickets to find gaps in training or documentation, then close those gaps. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your agents and customers find answers right away. [See how the done-for-you model works](/platform). --- ## First Response Time Source: https://en.helpfeel.com/glossary/first-response-time # First Response Time First response time (FRT) is the average time it takes for a customer support agent or system to send the first reply after a customer opens a ticket, starts a chat, or submits a contact form. It measures the speed of acknowledgment, not the speed of resolution. A short first response time tells customers their issue has been seen, even if the full answer takes longer. **Formula:** Sum of all time-to-first-reply values / Total number of tickets ## How is first response time calculated? The metric is typically measured from the moment a ticket enters the queue to the timestamp of the first outbound message from support. Most helpdesk platforms calculate it automatically. | Component | Definition | | ----------- | ----------------------------------------------------- | | Start | Customer submits ticket, opens chat, or sends message | | End | First reply from support agent or automated system | | Measurement | Elapsed time (usually in hours or minutes) | | Metric | Average across all tickets in a period | For example, if three tickets take 10 minutes, 30 minutes, and 20 minutes to receive the first reply, the first response time is (10 + 30 + 20) / 3 = 20 minutes. ## Why does first response time matter? First response time is one of the strongest predictors of customer satisfaction. Research consistently shows that a fast first reply, even if it is only an acknowledgment, reduces customer anxiety and sets a positive tone for the rest of the interaction. For teams, it is a simple proxy for queue health: a climbing first response time usually means understaffing, rising volume, or routing problems. Improving this metric is part of good [customer support metrics](/customer-support-metrics-and-kpis) discipline. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Average Handle Time](/glossary/average-handle-time): total time spent on a ticket. - [Average Resolution Time](/glossary/average-resolution-time): time to fully resolve an issue. ## Frequently asked questions ### What is a good first response time? It depends on the channel. Email support typically targets under 12 to 24 hours, live chat aims for under 1 minute, and social media usually targets under 1 hour. The fastest teams use AI-powered responses to hit sub-second first replies on chat and contact forms. ### How is first response time different from resolution time? First response time measures only the speed to the first reply. Resolution time measures the full time from ticket open to final resolution. A team can have fast first response but slow overall resolution if the problem requires multiple back-and-forth exchanges. ### How do you improve first response time? The fastest path is automating the first reply with AI chat or contact form acknowledgments that include relevant help articles. Inside human-staffed queues, prioritize by urgency, staff peak hours properly, and use templates for common opening responses. ## See it work Helpfeel delivers instant first responses with AI-powered chat and search that answer customer questions the moment they arrive. [See how the done-for-you model works](/platform). --- ## Generative AI in Customer Service Source: https://en.helpfeel.com/glossary/generative-ai-customer-service # Generative AI in Customer Service Generative AI in customer service uses large language models to create original, contextual responses to customer questions in real time. Instead of retrieving a pre-written article or following a decision tree, the system reads your knowledge base, understands the customer's question, and generates a natural-language answer on the spot. It handles follow-up questions, adapts tone, and works across channels without requiring teams to script every scenario. **Key point:** The AI generates the answer, not retrieves it. Each response is new. ## How does generative AI in customer service work? The system runs in three steps every time a customer asks a question: | Step | What happens | | ------------ | ---------------------------------------------------------------------------------------------------------------------------------- | | 1. Search | The AI searches your knowledge base for the most relevant information related to the customer's question. | | 2. Synthesis | It reads the retrieved sources and generates a natural-language answer, combining facts and adapting the tone to fit the question. | | 3. Delivery | The system presents the answer to the customer in chat, search, or email, with source links for verification. | Some systems add a fourth step: escalation to a human agent when confidence is low or the question falls outside the knowledge base. ## Why does generative AI in customer service matter? Generative AI lets your team answer an unlimited variety of customer questions without scripting responses in advance. Customers get answers instantly, in their own words, at any hour. For support teams, it means fewer repetitive tickets, faster resolution, and the freedom to focus on complex issues that need human judgment. The system learns from your existing knowledge, so it works with what you already have. For the complete breakdown of how to deploy it safely, what to measure, and how to choose between search and chat interfaces, read the [guide to generative AI in customer service](/ai-help-center/generative-ai-in-customer-service). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Chatbot](/glossary/chatbot): automated conversation system for customer interactions. - [Conversational AI](/glossary/conversational-ai): AI that understands and responds to natural language. ## Frequently asked questions ### How is generative AI different from a traditional chatbot? Traditional chatbots follow decision trees or retrieve pre-written answers. Generative AI creates new responses for each question, adapting language and detail to the customer's exact words. It can handle nuance, follow-up questions, and novel phrasing without a script. ### Is generative AI safe for customer-facing support? When built correctly, yes. The system must be grounded in your knowledge base to prevent hallucination, include guardrails to stay on topic, and offer escalation when confidence is low. Done-for-you systems handle these safety layers for you. ### What does generative AI in customer service actually do? It reads a customer's question, searches your knowledge base for relevant information, synthesizes an answer in natural language, and delivers it in seconds. Unlike lookup systems, it writes the response fresh each time, tailored to the question. ## See it work Helpfeel is built on generative AI that stays grounded in your knowledge. We set up the system, tune it to your content, and handle the safety layers so customers get accurate answers every time. [See how the done-for-you model works](/platform). --- ## Help Center Source: https://en.helpfeel.com/glossary/help-center # Help Center A help center is a self-service website where customers find answers to product or service questions without contacting support. It typically includes a knowledge base of articles, a search bar, category navigation, and sometimes video guides or community forums. The goal is to deliver instant, accurate answers 24/7, reducing the load on your support team while giving customers faster resolutions. ## What are examples of help centers? Every help center has the same core: search, organized content, and instant answers. Here are the most common setups: | Type | What it includes | | -------------------- | ------------------------------------------------------------------- | | Basic help center | Article library, category pages, search bar | | Advanced help center | AI-powered search, video guides, interactive tutorials, chat widget | | Community-powered | Discussion forums, user-contributed answers, upvoting | | Hybrid | Knowledge base, chatbot, live chat, and community all in one place | Most modern help centers combine a knowledge base with AI search, chat, and contextual suggestions that surface the right article at the right moment. ## Why does a help center matter? A help center is the first place customers look when they have a question. If they find the answer, they are back to work in seconds. If they do not, they wait in a queue for an agent. A fast, searchable help center improves the customer experience and reduces ticket volume at the same time, freeing your team to handle the conversations that actually need a human. A help center is the foundation of any [AI help center](/ai-help-center), which adds natural language understanding, agentic search, and continuous improvement. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Knowledge base](/glossary/knowledge-base): the article library inside a help center. - [Customer self-service](/glossary/customer-self-service): any system that lets customers solve problems without contacting support. ## Frequently asked questions ### What is the difference between a help center and a knowledge base? A help center is the customer-facing website where people find answers. A knowledge base is the library of articles inside it. The help center is the storefront; the knowledge base is the inventory. ### What should a help center include? At minimum, a search bar, organized article categories, and answers to common questions. Advanced help centers add chat, video guides, community forums, and AI-powered search that understands natural language questions. ### How does a help center reduce support tickets? Customers who find an answer in a help center do not need to submit a ticket. A well-organized, searchable help center resolves the majority of common questions instantly, freeing agents for complex issues. ## See it work Helpfeel is built to turn your help center into an answer engine. We set up and tune the knowledge base so your customers find what they need instantly. [See how the done-for-you model works](/platform). --- ## Help Desk Source: https://en.helpfeel.com/glossary/help-desk # Help Desk A help desk is a centralized service that receives, tracks, and resolves customer or employee support requests. It organizes incoming questions into tickets, routes them to the right person or team, and keeps a record of every interaction until the issue is closed. The goal is to make sure no request falls through the cracks and every user gets a timely answer. **Key point:** Help desks can be internal (IT support for employees) or external (customer support for buyers or users). ## How does a help desk work? Most help desks follow a standard flow: | Step | What happens | | ---------------------- | --------------------------------------------------------------------------------------------------- | | **1. Request arrives** | A customer or employee submits a question via email, chat, phone, or a web form. | | **2. Ticket created** | The system logs the request as a ticket with a unique ID and timestamp. | | **3. Routing** | The ticket is assigned to the right agent, team, or queue based on category, priority, or workload. | | **4. Resolution** | The agent investigates, answers, or escalates. All updates are logged in the ticket thread. | | **5. Closure** | Once resolved, the ticket is marked closed and archived for reporting and audit. | Modern help desks also automate parts of this flow, like categorizing tickets, suggesting answers from a knowledge base, or escalating urgent cases. ## Why does a help desk matter? Without a help desk, support requests scatter across email inboxes, chat threads, and sticky notes. Tickets get lost, duplicate work happens, and no one knows who is handling what. A help desk brings structure: every request is visible, trackable, and assigned. That makes support faster, more consistent, and easier to measure. For teams looking to reduce ticket volume and give customers instant answers before they ever open a ticket, help desks are part of good [customer service software](/customer-service-software). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Support ticket](/glossary/support-ticket): a record of one customer request. - [SLA](/glossary/sla): the promised response and resolution times. ## Frequently asked questions ### What is the difference between a help desk and a service desk? A help desk handles specific requests and incidents, typically IT or customer support. A service desk is broader: it manages the full service lifecycle, including changes, assets, and projects. Help desks are reactive; service desks are proactive. ### What are examples of help desk software? Zendesk, Freshdesk, Jira Service Management, HubSpot Service Hub, and Intercom are common examples. Each centralizes tickets, automates routing, and tracks resolution status. Some specialize in IT, others in customer support. ### Is a help desk the same as customer service software? They overlap, but help desk software is one type of customer service software. Help desks focus on ticketing, routing, and tracking. Customer service software can also include chat, phone, self-service, and knowledge management in one platform. ## See it work Helpfeel is built to answer customer questions before they become tickets. We set up and tune the knowledge base so your customers find what they need on their own, and your help desk stays focused on the hard problems. [See how the done-for-you model works](/platform). --- ## Knowledge Base Source: https://en.helpfeel.com/glossary/knowledge-base # Knowledge Base A knowledge base is a central library of articles, guides, and answers that customers or employees search to solve problems on their own. It is the content core of any self-service system. When someone searches your help center or asks a chatbot a question, the knowledge base is where the answer lives. A good knowledge base turns a wait-for-an-agent moment into an instant answer. **Key point:** It is not just a collection of articles. It is a searchable system organized so people can find the right answer at the moment they need it. ## What are examples of knowledge bases? | Type | Example | What it contains | | ----------------------- | -------------------------------- | ------------------------------------------------- | | Customer knowledge base | Stripe Docs, Shopify Help Center | Product guides, troubleshooting, billing FAQs | | Internal knowledge base | Company wiki, onboarding portal | HR policies, process docs, team playbooks | | Technical documentation | GitHub Docs, API reference sites | Developer guides, code samples, integration steps | All share the same goal: get the person to the answer without needing to ask someone. ## Why does a knowledge base matter? A knowledge base is the foundation of self-service. It gives customers answers instantly, at any hour, without waiting for an agent. For the business, it frees your team from answering the same question 50 times and lets them focus on the complex issues that actually need human judgment. Every article that works is time saved on both sides of the conversation. A knowledge base on its own is just a library. To make it useful, you need search, organization, and the right infrastructure around it. That is what [knowledge base management](/knowledge-base-management) solves. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Help center](/glossary/help-center): the customer-facing hub for support resources. - [Knowledge management](/glossary/knowledge-management): the discipline of organizing and maintaining knowledge. ## Frequently asked questions ### What are examples of knowledge bases? Public customer help centers (like Stripe or Zendesk's own docs), internal employee wikis, technical documentation sites, and FAQ libraries. Any searchable collection of answers organized to help people solve problems on their own is a knowledge base. ### What is the difference between a knowledge base and a help center? A help center is the customer-facing site where support resources live. A knowledge base is the article library inside it. Help centers often include other tools like contact forms or chat, but the knowledge base is the searchable content core. ### How do you build a good knowledge base? Start with the questions customers actually ask. Write clear, scannable answers. Organize by topic, make search easy, and review what people search for versus what they find. A good knowledge base is never finished; it evolves with the questions. ## See it work Helpfeel builds and tunes the knowledge base for you. We start with your existing content, shape it around real customer questions, and keep it sharp as those questions change. [See how the done-for-you model works](/platform). --- ## Knowledge Management Source: https://en.helpfeel.com/glossary/knowledge-management # Knowledge Management Knowledge management is the process of creating, organizing, sharing, and using information within an organization so people can make better decisions and work more effectively. It captures what your team knows, from product documentation and support best practices to customer insights and lessons learned, and makes that knowledge accessible when it is needed. Done well, it means new hires ramp faster, customers find answers on their own, and institutional knowledge survives when people leave. **Key point:** Knowledge management is both a strategy (deciding what knowledge matters and how to preserve it) and a set of tools and practices (wikis, documentation, search, governance) that turn individual expertise into shared organizational assets. ## How does knowledge management work? Effective knowledge management follows a cycle of creating, organizing, sharing, and refining information: | Stage | What happens | | -------------- | ------------------------------------------------------------------------------------------------------------------------- | | Create | Teams document what they know: best practices, process guides, product specs, troubleshooting steps. | | Organize | Information is categorized, tagged, and made searchable so people can find it later. | | Share | Knowledge is distributed through help centers, internal wikis, onboarding materials, or team channels. | | Use and refine | People find and apply the knowledge. Feedback loops surface what is missing or outdated so the system improves over time. | The hardest part is not the first pass. It is keeping the system alive so the knowledge stays accurate and trusted. ## Why does knowledge management matter? Without it, every question gets answered from scratch, every new hire rebuilds context, and every customer waits for an agent who already answered the same thing yesterday. Good knowledge management turns repeatable questions into instant answers, frees experts to focus on new problems, and scales what your team knows faster than headcount. For customer-facing teams, the operational side of knowledge management is [knowledge base management](/knowledge-base-management): building and maintaining the help center that delivers answers to customers. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Knowledge base](/glossary/knowledge-base): a centralized repository of articles that answer customer questions. - [Customer self-service](/glossary/customer-self-service): when customers resolve their own issues without contacting support. ## Frequently asked questions ### What is the difference between knowledge management and knowledge base management? Knowledge management is the broad organizational practice of capturing and sharing all institutional knowledge. Knowledge base management is narrower: it focuses on building and maintaining the customer-facing help center or internal documentation system. One is culture and strategy, the other is execution. ### What are examples of knowledge management? Common examples include centralized wikis for internal teams, onboarding playbooks that capture how work gets done, case study libraries that preserve lessons from past projects, and customer help centers that document product knowledge so customers can self-serve. ### Why does knowledge management fail? Most knowledge management efforts fail because they rely on humans to document and update information manually, which is slow and inconsistent. Without a clear owner, a forcing function to keep content current, and tools that make contribution easy, the system decays into outdated articles no one trusts. ## See it work Helpfeel is built for the customer-facing side of knowledge management. We set up and maintain the knowledge base so your customers find the answers they need. [See how the done-for-you model works](/platform). --- ## NPS Source: https://en.helpfeel.com/glossary/nps # NPS Net Promoter Score (NPS) is a customer loyalty metric that measures how likely customers are to recommend your product or service to others. It is based on a single question: "On a scale of 0 to 10, how likely are you to recommend us to a friend or colleague?" The score ranges from negative 100 to positive 100, and it is one of the most widely adopted benchmarks for tracking customer sentiment and loyalty over time. **Formula:** % Promoters (9-10) − % Detractors (0-6) = NPS ## How is NPS calculated? Customers are grouped into three categories based on their rating: | Rating | Label | Definition | | ------ | ---------- | --------------------------------------------------------------------------- | | 9-10 | Promoters | Loyal customers who will recommend you and drive growth. | | 7-8 | Passives | Satisfied but unenthusiastic. Vulnerable to competitors. | | 0-6 | Detractors | Unhappy customers who may damage your brand through negative word of mouth. | To calculate NPS, subtract the percentage of Detractors from the percentage of Promoters. Passives count toward the total but do not affect the score directly. **Example:** If 60% of respondents are Promoters, 30% are Passives, and 10% are Detractors, your NPS is 60 − 10 = 50. ## Why does NPS matter? NPS is a simple proxy for customer loyalty and retention. A high score suggests customers trust you enough to stake their reputation on a recommendation, which is a strong signal of product market fit and long term growth potential. A low or negative score is an early warning that churn risk is high and customer experience needs attention. NPS is part of good [customer support metrics](/customer-support-metrics-and-kpis), alongside operational measures like response time and resolution rate. Together, they give a full picture of how well you serve customers. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [CSAT](/glossary/csat): satisfaction score for specific interactions. - [Customer Effort Score](/glossary/customer-effort-score): measures how easy it was to resolve an issue. ## Frequently asked questions ### What is a good NPS score? Any score above zero is positive. 0 to 30 is good, 30 to 70 is great, and above 70 is world class. Context matters: B2B SaaS typically sees 30 to 40, while consumer brands average 20 to 50. Compare your score to your own past performance and industry benchmarks, not just the absolute number. ### How is NPS different from CSAT? NPS measures loyalty and likelihood to recommend. CSAT measures satisfaction with a specific interaction or product. NPS is a broader gauge of the relationship over time, while CSAT is transactional and moment specific. Both are useful, but they measure different things. ### How often should you measure NPS? Most companies run NPS surveys quarterly or after major milestones. Survey fatigue is real, so avoid over surveying the same customers. A single annual or biannual NPS pulse, plus targeted relationship surveys for key accounts, is a common cadence in B2B. ## See it work Helpfeel is built to improve every metric that matters, including NPS. We set up and tune the knowledge base so your customers get instant, accurate answers, which drives satisfaction and loyalty. [See how the done-for-you model works](/platform). --- ## Omnichannel Support Source: https://en.helpfeel.com/glossary/omnichannel-support # Omnichannel Support Omnichannel support is a customer service approach that connects all communication channels, like email, chat, phone, social media, and self-service, into one unified system. When a customer switches from one channel to another, their history, context, and conversation follow them. The goal is to deliver a seamless experience no matter how or where the customer reaches out. **Key point:** The difference between multichannel and omnichannel is integration. Multichannel means you offer many channels. Omnichannel means those channels talk to each other. ## How does omnichannel support work? Omnichannel support depends on a shared data layer that tracks every interaction across channels. Here is how the core components fit together: | Component | What it does | | --------------------- | ------------------------------------------------------------------------------------------------------- | | Unified inbox | All messages from every channel land in one queue, so agents see the full picture. | | Customer profile | Every interaction adds to a single profile with history, preferences, and past issues. | | Context handoff | When a conversation moves from chat to email to phone, the context moves with it. | | Cross-channel routing | The system routes conversations to the right agent based on full history, not just the current message. | The system treats each customer as one person across all touchpoints, not as a new case every time they switch channels. ## Why does omnichannel support matter? Customers expect to pick up where they left off. If they start a question in chat and follow up by email the next day, they do not want to repeat the entire story. Omnichannel support removes that friction. It speeds up resolution, reduces customer effort, and builds trust because the experience feels intentional and connected. For support teams, it means agents spend less time hunting for context and more time solving problems. Omnichannel support is part of good [customer service software](/customer-service-software) that scales without losing the thread. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Help desk](/glossary/help-desk): the system that manages and routes support tickets. - [Customer self-service](/glossary/customer-self-service): tools that let customers find answers on their own. ## Frequently asked questions ### What is the difference between multichannel and omnichannel support? Multichannel support offers customers several ways to reach you, like email, chat, and phone, but each channel operates independently. Omnichannel support connects those channels so a conversation that starts on chat can continue over email without the customer repeating themselves. ### What are examples of omnichannel support? A customer starts a question in live chat, receives a follow-up email with their chat transcript attached, then picks up the conversation by phone with an agent who sees the full history. The channel changes, but the context stays intact. ### Why is omnichannel support important? It removes friction. Customers do not have to repeat themselves when they switch channels, agents see the full history before they respond, and the experience feels seamless. That speeds up resolution and builds trust. ## See it work Helpfeel brings omnichannel thinking to self-service. Customers search once and get answers across all your content, no matter where it lives. [See how the done-for-you model works](/platform). --- ## Self-Service Rate Source: https://en.helpfeel.com/glossary/self-service-rate # Self-Service Rate Self-service rate is the percentage of customers who resolve their own issues using a help center, knowledge base, or search, without contacting a human agent. It is the core measure of how well your self-service resources answer real customer questions. A higher rate means more customers get an instant answer and fewer wait in a queue. **Formula:** (Successful self-service resolutions / Total customer inquiries) × 100 ## How is self-service rate calculated? Most teams measure it one of three ways, depending on the data they have: | Method | Calculation | | ---------------- | ------------------------------------------------------------------------ | | Direct | (Knowledge base views with no ticket created / Total views) × 100 | | Inverse | 100% − contact rate, where contact rate = (tickets / active users) × 100 | | Resolution-based | (Self-resolved issues / Total issues) × 100 | The direct method is the most common because it ties the metric straight to help center behavior. ## Why does self-service rate matter? A high self-service rate means customers get answers instantly, at any hour, without waiting for an agent. For the business, each resolved self-service session is one less ticket in the queue, which frees your team for the complex conversations that actually need a human. It is one of the clearest signals that your knowledge base is doing its job. For benchmarks by industry and company size, the full playbook for moving this number, and the strategies behind it, read the [complete guide to self-service rate](/customer-support-metrics-and-kpis/self-service-rate-complete-guide). ## Related terms - [Deflection rate](/customer-support-metrics-and-kpis/deflection-rate-complete-guide): tickets prevented before they reach an agent. - [Containment rate](/customer-support-metrics-and-kpis/containment-rate-complete-guide): issues fully resolved inside an automated system. - [Self-service, deflection, and containment compared](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article): how the three metrics fit together. ## Frequently asked questions ### What is a good self-service rate? It varies by industry. Mature ecommerce help centers reach 65 to 80 percent, while SaaS and financial services typically land in the 40 to 60 percent range. What matters more than the absolute number is the trend: a rate climbing over time means your knowledge base is getting better at answering real questions. ### How is self-service rate different from deflection rate? Self-service rate measures issues customers resolve on their own. Deflection rate measures tickets prevented before they reach an agent. They overlap, but self-service rate is framed around the customer succeeding, while deflection is framed around the ticket never arriving. ### How do you calculate self-service rate? Divide successful self-service resolutions by total customer inquiries, then multiply by 100. In practice, teams measure it as knowledge base views that did not lead to a ticket, or as 100 percent minus the contact rate. ## See it work Helpfeel is built to move this number. We set up and tune the knowledge base so your customers find answers on their own. [See how the done-for-you model works](/platform). --- ## SLA Source: https://en.helpfeel.com/glossary/sla # SLA A Service Level Agreement (SLA) is a contract between a service provider and a customer that defines the standards the provider commits to meet. In customer service, SLAs typically set response times, resolution times, and availability guarantees, so both sides have a shared understanding of what counts as good service. When someone asks what your SLA is, they want to know how fast you will reply and how reliably you will be there. **Key commitments:** Response time, resolution time, availability, and sometimes quality targets like customer satisfaction score. ## What goes in a customer service SLA? The core elements depend on what matters most to your customers and what you can reliably deliver: | Element | What it defines | | --------------- | ---------------------------------------------------------------------------------- | | Response time | How quickly you send the first reply (e.g., within 1 hour for priority tickets) | | Resolution time | How quickly you close the ticket (e.g., 24 hours for standard issues) | | Availability | Hours the service is staffed or accessible (e.g., 24/7, business hours) | | Scope | What is covered (technical support, billing, onboarding) and what is not | | Quality targets | Optional metrics like customer satisfaction score or first-contact resolution rate | | Consequences | What happens if you miss the target (refund, credit, escalation) | The tighter the SLA, the more your operations have to be built to support it. ## Why do SLAs matter? An SLA sets expectations clearly, which reduces frustration on both sides. Customers know when they will hear back. Your team knows what to prioritize. For enterprise sales, an SLA is often a deal requirement, it is proof you take reliability seriously. The challenge is writing one you can actually meet. An SLA you miss regularly is worse than none at all because it becomes a broken promise. SLAs are part of good [customer service software](/customer-service-software) and support operations. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Help desk](/glossary/help-desk): software that tracks and manages support requests. - [First response time](/glossary/first-response-time): how quickly you send the first reply. ## Frequently asked questions ### What goes in a customer service SLA? The most common commitments are response time (how quickly you reply), resolution time (how quickly you close the ticket), and uptime (how many hours the service is available). Some SLAs also set quality targets like customer satisfaction score or first-contact resolution rate. ### What happens when an SLA is broken? The consequence is whatever the contract says. Common remedies include service credits, discounts, or the right to terminate. Internal SLAs typically trigger escalations or post-mortems rather than financial penalties. ### Do all companies need an SLA? Not every company. SLAs are most valuable when customers depend on fast or reliable support, when you have multiple support tiers, or when you sell to enterprise buyers who expect contractual guarantees. ## See it work Helpfeel is built to keep you ahead of your SLA. We answer questions instantly in the knowledge base so tickets never arrive in the first place. [See how the done-for-you model works](/platform). --- ## Support Ticket Source: https://en.helpfeel.com/glossary/support-ticket # Support Ticket A support ticket is a record of a customer request, question, or problem, created when a customer contacts a company through email, chat, web form, or phone. The ticket is logged in a help desk system, assigned a unique ID, and tracked through each stage of resolution until the issue is closed. It serves as the single source of truth for what the customer asked, who is handling it, and what has been done so far. **Key point:** Every support ticket has a status (open, pending, resolved, closed) and moves through a workflow until the issue is solved. ## What are examples of support tickets? Support tickets span a wide range of customer needs. Here are common types: | Type | Example | | ---------------- | --------------------------------------------------------- | | Technical issue | "My account won't sync after the latest update." | | Billing question | "I was charged twice for the same order." | | How-to request | "How do I export my data to a CSV file?" | | Feature request | "Can you add a dark mode to the mobile app?" | | Bug report | "The checkout button doesn't work on Safari." | | Account access | "I forgot my password and the reset email never arrived." | Each ticket gets logged, categorized, and assigned to the right team or agent for resolution. ## Why does support ticket management matter? Good ticket management means customers get answers faster and nothing falls through the cracks. A well-organized ticket system helps teams prioritize urgent issues, track response and resolution times, and identify recurring problems that might need a product fix or a new help article. It also creates a record of every interaction, so agents can pick up where the last person left off without asking the customer to repeat themselves. Tickets are a core part of good [customer support](/customer-support). Reducing ticket volume starts with strong self-service resources that answer common questions before they become tickets. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Help desk](/glossary/help-desk): software that manages and tracks support tickets. - [Ticket volume](/glossary/ticket-volume): total number of support tickets received in a period. ## Frequently asked questions ### What information does a support ticket contain? A support ticket typically contains the customer's name, contact information, request or issue description, timestamp, ticket ID, and status. It may also include attachments, order numbers, priority level, and the agent assigned to resolve it. ### What is the difference between a ticket and a case? In most systems, a ticket and a case are the same thing: a record of a customer issue. Some platforms use case for complex or multi-touch issues and ticket for simpler requests, but the terms are often used interchangeably. ### How do support tickets improve customer service? Support tickets ensure no request is lost or forgotten. They create a single place to track status, assign responsibility, measure response times, and review past interactions. Teams can prioritize urgent issues and spot patterns across many tickets. ## See it work Helpfeel is built to reduce the number of tickets that reach your team by answering customer questions in self-service. [See how the done-for-you model works](/platform). --- ## Ticket Volume Source: https://en.helpfeel.com/glossary/ticket-volume # Ticket Volume Ticket volume is the total number of support tickets received by your team in a given time period, such as a day, week, month, or quarter. It is a core operational metric that shows the raw demand on your support team. When ticket volume is high, your team is stretched thin. When it drops, agents have room to deliver better service or take on proactive work. **Formula:** Count of all tickets received in the period ## How is ticket volume measured? Most teams track it by day, week, and month, then break it down by channel and topic: | Dimension | What it shows | | ---------------------------------------- | --------------------------------------------- | | Total tickets per period | The raw load on the team | | Tickets per channel (email, chat, phone) | Where customers go for help | | Tickets per topic or category | What customers need help with | | Trend over time | Whether volume is rising, falling, or spiking | Breaking volume down by topic is the most actionable view because it tells you which issues drive the most tickets, and where self-service can make the biggest dent. ## Why does ticket volume matter? High ticket volume stretches the team, slows response times, and makes it harder to deliver consistent quality. Reducing it sustainably means customers get faster answers through self-service and agents have more time for the complex questions that really need a human. For the full playbook on bringing volume down, including what to measure, which content to write first, and how to surface it proactively, read the [guide to reducing support tickets](/customer-support/how-to-reduce-support-tickets). ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Support ticket](/glossary/support-ticket): formal customer inquiry to support. - [Contact rate](/glossary/contact-rate): percentage of customers who contact support. ## Frequently asked questions ### What is considered high ticket volume? It varies by company size and industry. A rapid climb, recurring spikes around product updates, or volume that consistently overwhelms the team are all signals that ticket volume is a problem worth addressing. The absolute number matters less than the pattern. ### What causes ticket volume to spike? Product bugs, unclear documentation, missing self-service paths, confusing product changes, and seasonal events are the most common drivers. Each spike leaves a signal in the ticket data, so tracking volume by topic is the fastest path to the root cause. ### How do you reduce ticket volume? Build a strong self-service knowledge base, make it easy to find with semantic search, surface it proactively in-product, and close the loop by turning high-volume ticket patterns into new help content. The cycle improves over time. ## See it work Helpfeel is built to reduce ticket volume. We set up the knowledge base, tune it to answer your top ticket drivers, and make it easy for customers to find answers on their own. [See how the done-for-you model works](/platform). --- ## Tiered Support Source: https://en.helpfeel.com/glossary/tiered-support # Tiered Support Tiered support is a structure that organizes your customer support team by skill level and issue complexity. Tier 1 agents handle simple, high-volume questions like password resets and account setup. Tier 2 agents troubleshoot product issues that require deeper knowledge. Tier 3 specialists resolve technical escalations, software bugs, and problems that need engineering or domain expertise. The goal is to route each issue to the right person, so customers get answers quickly and your most skilled people focus on the hardest problems. **Key point:** Tiering reduces cost and resolution time by matching complexity to capability. ## What are the three tiers of support? Here is how the structure typically breaks down: | Tier | Focus | Examples | | ---------- | -------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- | | **Tier 1** | Common questions, account help, password resets, basic troubleshooting | "How do I log in?" "Where is my order?" "How do I cancel my subscription?" | | **Tier 2** | Product issues, configuration, bug reproduction, non-routine troubleshooting | "Feature X stopped syncing." "I'm getting an error when I export." "How do I configure Y for my use case?" | | **Tier 3** | Engineering escalations, code-level bugs, deep technical problems, custom integrations | "This API call returns the wrong value." "The database migration broke feature Z." "We need a custom SSO config." | Each tier escalates to the next when an issue exceeds their scope or authority. ## Why does tiered support matter? Tiered support lets you scale without hiring specialists for every question. Simple issues get resolved quickly by Tier 1, while complex problems reach the people who can actually fix them. It keeps costs predictable, reduces resolution time, and protects your most expensive team members from repetitive work. As your customer base grows, a tiered structure ensures every issue lands in the right hands, so nothing falls through the cracks and no one is overwhelmed. For more on how support structures fit into a broader [customer support](/customer-support) strategy, including self-service as a Tier 0 layer, see the full guide. ## Related terms - [CX glossary](/glossary): browse all customer experience terms. - [Escalation rate](/glossary/escalation-rate): percentage of issues moved to a higher tier. - [Support ticket](/glossary/support-ticket): customer requests tracked through resolution. ## Frequently asked questions ### What is the difference between Tier 1, 2, and 3 support? Tier 1 handles common questions and password resets. Tier 2 solves product issues that need troubleshooting or configuration. Tier 3 tackles deep technical problems, bugs, and escalations that require engineering or specialist knowledge. ### What are the benefits of tiered support? It routes issues to the right skill level, so simple questions get instant answers and complex problems land with the people who can fix them. It keeps costs down, reduces resolution time, and frees specialists to focus on hard work. ### Do I need a tiered support structure? If your team is small and everyone handles every issue, you probably don't. As volume and complexity grow, tiering makes sense: it protects specialist time and ensures customers get fast answers for common questions. ## See it work Helpfeel reduces the load on every tier by answering common questions before they become tickets. Self-service handles Tier 1 volume automatically, so your team focuses on what actually needs a human. [See how the done-for-you model works](/platform). --- ## Knowledge Base Management: The Complete Guide Source: https://en.helpfeel.com/knowledge-base-management # Knowledge Base Management: The Complete Guide Knowledge base management is the ongoing work of keeping a self-service knowledge base accurate, findable, and current, so customers resolve their own questions without contacting an agent. Most teams treat the knowledge base as a project they finish once. The teams that cut inquiry volume treat it as a system they keep improving. This guide covers what a managed knowledge base is, how to run one well, and how to prove its value. Each section below stands on its own, and links to a full guide if you want to go deeper. If you are earlier in the journey, start with [self-service optimization](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article), the broader strategy this fits inside. ## What is a managed knowledge base? A managed knowledge base is one that a defined process keeps current, not one that goes stale after launch. Someone owns it, a workflow flags gaps and outdated articles, and updates ship on a regular cadence. This is the difference between a help center that quietly decays and one that keeps resolving more questions every quarter. The "managed" part is the whole point. A static set of articles answers fewer questions each month as the world moves on. A managed base closes that gap on purpose. **→ Read the full guide: [What is a managed knowledge base](/knowledge-base-management/what-is-a-managed-knowledge-base)** ## What are knowledge base best practices? Strong knowledge base articles share a few habits: one article answers one question, the answer comes first, headings match how people actually search, and paragraphs stay short and scannable. Each article should also link to related articles so readers and search engines can follow the trail. Follow these habits and your articles get found, read, and reused. Skip them and even good content sits unread. **→ Read the full guide: [Knowledge base best practices](/knowledge-base-management/knowledge-base-best-practices)** ## What is an AI-ready knowledge base? An AI-ready knowledge base is structured cleanly enough that an AI assistant can retrieve and trust a single answer from it. That means clear headings, self-contained sections, and accurate, current content. The same structure that helps a person skim helps an AI cite you. You do not need a separate project to "get AI-ready." Get the content prepared, and AI readiness follows. **→ Read the full guide: [The AI-ready knowledge base](/knowledge-base-management/ai-ready-knowledge-base)** ## How do you measure knowledge base ROI? Measure the inquiries your knowledge base resolves before they reach an agent, then multiply by your cost per contact. That gives you the dollars saved. Track self-service rate and search success rate next to it, so you can see whether the savings come from real resolutions or from customers giving up. A knowledge base that deflects volume but leaves customers stuck is not saving money. It is spending trust. Measure both. **→ Read the full guide: [Knowledge base ROI](/knowledge-base-management/knowledge-base-roi)** ## How do you keep a knowledge base from going stale? Keep a knowledge base current by reviewing it on a schedule, not when someone complains. Pull the searches that returned no useful answer, flag articles no one reads, and refresh anything tied to a product that changed. A short, regular cadence beats a once-a-year overhaul. Staleness is the default state of any help content. The only thing that prevents it is a maintenance habit. **→ Read the full guide: [Knowledge base maintenance](/knowledge-base-management/knowledge-base-maintenance)** ## The five parts of knowledge base management | If you want to | Read | | ------------------------------ | ----------------------------------------------------------------------------------------------- | | Understand the managed model | [What is a managed knowledge base](/knowledge-base-management/what-is-a-managed-knowledge-base) | | Write articles that get found | [Knowledge base best practices](/knowledge-base-management/knowledge-base-best-practices) | | Make your content work with AI | [The AI-ready knowledge base](/knowledge-base-management/ai-ready-knowledge-base) | | Prove the dollar value | [Knowledge base ROI](/knowledge-base-management/knowledge-base-roi) | | Stop content from going stale | [Knowledge base maintenance](/knowledge-base-management/knowledge-base-maintenance) | ## Frequently asked questions ### What is knowledge base management? Knowledge base management is the ongoing practice of keeping a self-service knowledge base accurate, findable, and current. It covers writing, organizing, measuring, and improving help content so customers resolve their own questions without contacting an agent. ### Is a knowledge base a one-time project? No. A knowledge base built once and left alone goes stale within months. Search terms shift, products change, and unanswered questions pile up. A managed knowledge base improves on a set cadence, which is what keeps inquiry volume down over time. ### How do you measure knowledge base ROI? Measure the inquiries your knowledge base resolves before they reach an agent, then multiply by your cost per contact. Track self-service rate and search success rate alongside cost saved, so you can see both the quality and the dollar value. ## Go deeper A knowledge base is only as good as the work that keeps it current. Helpfeel runs that work for you: we build the content, watch what customers search for, and close the gaps before they become tickets. See [how the managed model works](/product), or read the broader strategy in [self-service optimization](/customer-support-metrics-and-kpis/self-service-deflection-containment-rates-knowledge-article). --- ## Why Knowledge Bases Go Out of Date Source: https://en.helpfeel.com/knowledge-base-management/why-knowledge-bases-go-out-of-date # Why Knowledge Bases Go Out of Date A knowledge base is not a static asset. It is a running claim about how a product works, and the product does not hold still. Every release, price change, and policy update widens the gap between what customers need to know and what the help center says. Left alone, the content drifts out of alignment with reality and quietly becomes wrong. This is a core part of [knowledge base management](/knowledge-base-management). This piece situates why decay happens and why it accelerates. It stops short of prescribing a fix. ## Where does knowledge decay begin? It begins with an ownership vacuum. In most teams no single person is accountable for keeping articles true. The people who know what changed are shipping the change, and documentation is not on their critical path. The people who own the help center learn what changed when a customer reports that the instructions no longer match the screen. Maintenance is everyone's job at the margin, which makes it no one's job in practice. ## Why does the decay curve steepen instead of flattening? Three forces bend the curve upward over time: 1. **Change compounds.** Every release and policy shift is a fresh chance for an existing article to become false. Modern release cadence is fast, so new decay is continuous. 2. **Trust decays faster than content.** Each stale article a reader hits lowers trust in the whole corpus, so people stop consulting it and stop reporting its errors, removing the informal mechanism that was catching decay. 3. **Surface area outruns attention.** As the corpus grows, the amount needing maintenance grows with it while the fraction anyone reviews shrinks. Growth itself accelerates decay. ## Why is a stale answer worse than no answer? Picture a customer typing a real question into a help center. Search returns a confident, well-formatted article. It is out of date. They follow it, the steps do not match the product, and they lose time before realizing the answer failed them. Only then do they contact a human, now frustrated. Compare that to a help center with no article at all. The customer contacts support immediately, no time wasted, no trust spent. A stale answer carries negative value. It costs more than silence. ## What does this force into view? Eventually the numbers surface the problem the maintenance never did. Self-service rate quietly falls. Customers route to humans for things the help center supposedly covers. A newly deployed AI layer starts returning bad answers because the source it reads is stale. The uncomfortable clarity is that the original effort was never the problem. The articles were good when written. A knowledge base simply has no built-in mechanism to stay true, and the product it describes never stops moving. ## Frequently asked questions ### Why do knowledge bases become outdated? Because a knowledge base describes a product that keeps changing. Every release, price change, and policy update can make an existing article wrong, and most teams have no one accountable for keeping the content aligned with the current product. ### Is a stale help center article worse than no article? Often yes. A stale article returns a confident answer that costs the reader time before it fails them, then sends them to a human anyway. A missing article sends them to a human immediately, with no time or trust spent. ### Why does knowledge decay get worse over time? Change compounds, trust decays faster than content, and the surface area to maintain outruns the attention available. A larger corpus is reviewed less thoroughly, so growth itself accelerates decay. Helpfeel exists because of this exact problem. [See how we think about it](/platform). --- ## Voice of Customer Analytics: The Guide Source: https://en.helpfeel.com/voc-analytics # Voice of Customer Analytics: The Guide Voice of customer analytics is the practice of turning your support conversations into a ranked map of customer demand. It reads the tickets, chats, searches, and call logs you already collect and answers three questions: what are customers asking about, how do they feel about it, and what are they trying to do? The answers tell you what to fix, in what order. Most support teams sit on years of this data and act on almost none of it. This guide covers the three techniques that turn that volume into decisions. Each section stands on its own and links to a full guide. If you are also working on the content side of self-service, this pairs naturally with [knowledge base management](/knowledge-base-management). ## What is voice of customer analytics? Voice of customer analytics is how you listen to customers at scale. Instead of reading a sample of tickets and guessing at the pattern, you analyze all of them and let the pattern show itself. The output is a prioritized list of what your customers need, backed by the volume and emotion behind each item. The bottleneck it solves is a modern one: teams already have more data than they know what to do with, and what's missing is a system to turn that pile into a short list of what to do. Voice of customer analytics is that system. ## What is topic modeling? Topic modeling groups thousands of unlabeled support messages into the handful of themes customers keep raising. You do not define the categories in advance. The model finds them, counts them, and ranks them, so the biggest drivers of contact rise to the top. It's the same technique statisticians call cluster analysis, in plainer English. This is where most teams start, because it answers the first question every support leader has: what are people actually contacting us about? **→ Read the full guide: [Topic modeling for customer support](/voc-analytics/topic-modeling)** ## What is sentiment analysis? Sentiment analysis reads the emotional tone in each message and sorts it as positive, negative, or neutral. Laid over your topics, it reweights your priorities: you fix the topic that's both common and painful first, instead of chasing whatever pile is biggest. It's also an early warning system, since rising negative sentiment often shows up before churn does. **→ Read the full guide: [Sentiment analysis for customer support](/voc-analytics/sentiment-analysis)** ## What is intent analysis? Intent analysis reads what a customer is actually trying to do. Where topic modeling sorts by subject and sentiment reads emotion, intent reads the goal, so your support system can act on the need and predict the next one. It's the deepest layer of the practice, and the one that moves support from reacting to getting ahead. **→ Read the full guide: [Intent analysis in customer support](/voc-analytics/intent-analysis)** ## How do the three techniques fit together? They stack. Each one adds a dimension to the same set of conversations, and they're strongest combined. | Technique | Question it answers | What you do with it | | ------------------------------------------------------- | -------------------------------- | ------------------------------------- | | [Topic modeling](/voc-analytics/topic-modeling) | What are customers asking about? | Rank what to fix by volume | | [Sentiment analysis](/voc-analytics/sentiment-analysis) | How do they feel about it? | Reprioritize by pain, not just volume | | [Intent analysis](/voc-analytics/intent-analysis) | What are they trying to do? | Act on the need, predict the next one | Read together, they answer what customers ask, how they feel, and what they want, which is everything you need to decide where to spend your team's next hour. ## Frequently asked questions ### What is voice of customer analytics? Voice of customer analytics is the practice of turning support conversations into insight. It reads tickets, chats, searches, and call logs to show what customers ask about, how they feel, and what they are trying to do, so you can act on real demand. ### What are the main types of voice of customer analysis? Three build on each other: topic modeling groups messages by subject, sentiment analysis reads the emotion, and intent analysis reads the goal behind each message. Together they tell you what customers ask, how they feel, and what they want to do next. ### What data do you need for voice of customer analytics? Any text record of customer contact: support tickets, chat transcripts, search queries, contact form messages, or call transcripts. A year of history is a solid starting sample. More data sharpens the picture, but you rarely need every record. ## Go deeper Voice of customer analytics is only worth as much as what you do with it. Helpfeel runs the analysis continuously and turns it into action: it reads every search, click, and contact message, drafts the help content your customers are missing, and queues it for a one-click human review. See [how the platform works](/platform), or start with [topic modeling](/voc-analytics/topic-modeling), the first layer of the practice. --- ## Topic Modeling for Customer Support Data Source: https://en.helpfeel.com/voc-analytics/topic-modeling # Topic Modeling for Customer Support Data Topic modeling groups thousands of unlabeled support messages into the handful of themes your customers keep asking about. Instead of reading tickets one by one, you get a ranked list of topics, each with its own volume, so you can see what to fix first. It's the same idea as cluster analysis, just said in plainer English. Building out a voice-of-customer practice? This page sits alongside [sentiment analysis](/voc-analytics/sentiment-analysis) and [intent analysis](/voc-analytics/intent-analysis) in the [VoC analytics guide](/voc-analytics). Most teams start here, because topic modeling answers the first question every support leader asks: what are people actually contacting us about? ## What is topic modeling? Topic modeling reads a big pile of text and sorts it into groups by similarity. You don't define the categories in advance. Feed it 20,000 tickets and it hands back something like 40 topics: "password reset," "shipping delays," "return policy," "spec change on an existing order." Each one carries a count, so the biggest drivers of contact rise to the top on their own. One quick note on the name. Statisticians call this **cluster analysis**. Same method, different label. Tim Thijsse, a top-50 CX leader, used "topic modeling" on the CX Heroes podcast because it's easier to understand for an English-speaking audience: it names what you walk away with, a set of topics, not the math behind it. We use it here for the same reason. Topic modeling is unsupervised. You don't hand it a list of categories and ask it to file messages into them. It finds the structure that's already sitting in the data, which is exactly what makes it useful for support: your customers rarely ask questions in the neat categories your help center was built around. ## Why does topic modeling matter for support now? Because the bottleneck has moved. For years the problem was too little data. Now it's the opposite. > "Companies solved the 'not enough data' problem. Now they have too much data and no system for turning it into prioritized, actionable insights. That is where most CX programs are stuck right now." > > Tim Thijsse, CX strategist and author of _Maturing in Customer Experience Optimization_ Every support team already sits on a mountain of signal: tickets, chats, searches, call logs. What almost none of them have is a way to turn that volume into a short, ranked list of what to do first. Topic modeling is that system. It turns raw contact volume into a prioritized map of demand, so the next help article you write answers the question you get 800 times a month, not whatever happened to be top of mind. For you, that changes the job: work from evidence instead of guessing which gaps matter. ## How does topic modeling work on support data? The workflow is consistent across tools. Five steps get you there: 1. **Gather the text.** Pull your raw records: tickets, chat transcripts, search queries, contact form messages, or call transcripts. Any text log of what customers asked works. A year of history is a solid first sample. 2. **Clean and prepare it.** Strip personal data, normalize formatting, and, for call audio, transcribe to text first. Quality here sets the ceiling on everything downstream. 3. **Cluster into topics.** The model groups similar messages together by meaning, not just matching keywords, so "can't log in," "forgot my password," and "locked out of my account" land in one topic instead of three. 4. **Rank and label.** Each cluster gets a plain-language label and a size. Sort by volume, or weight by cost-to-serve, and the priorities order themselves. 5. **Decide and act.** Read the top topics and turn them into work: a new help article, a clearer product flow, a fix for the thing generating repeat contacts. One practical caveat: you don't always need every record. A well-designed sample often gives you the same picture as the full set, at a fraction of the processing cost. The real skill is choosing a sample that faithfully represents the whole, and that's where good tooling earns its keep. ## Topic modeling vs sentiment analysis vs intent analysis These three techniques answer different questions about the same conversations. Run them together and you get the full picture. | Technique | Question it answers | What you get | | ---------------------- | -------------------------------- | ------------------------------------------------------------------- | | **Topic modeling** | What are customers asking about? | Ranked themes by volume ("returns," "spec changes") | | **Sentiment analysis** | How do they feel about it? | Positive, negative, or neutral tone per message or topic | | **Intent analysis** | What are they trying to do? | The underlying goal behind the message, and what should happen next | Topic modeling tells you _what_. Sentiment tells you _how they feel_. [Intent analysis](/voc-analytics/intent-analysis) goes further still and reads _what the customer is trying to accomplish_, which lets a support system act on a need instead of just filing it away after the fact. Most VoC practices start with topic modeling, then add intent analysis as the practice matures. ## How Helpfeel applies topic modeling Helpfeel Analytics is topic modeling built into your support operation, and run for you instead of by you. It watches every search, click, and contact message, then clusters and ranks them continuously. What that looks like in practice: - It groups incoming questions into topics and surfaces the ones your customers struggle to get answered today. - It drafts the missing or outdated help content for each high-volume gap, then queues the draft for a one-click human review. - It rolls rising topics into voice-of-customer reports your product and marketing teams can read, so the same analysis that sharpens your help center also feeds your roadmap. The model is simple. A person approves, the AI answers, the cycle repeats. Knowledge work stops being a quarterly cleanup and turns into a flow your team supervises, one that keeps closing the gaps in your help center as they show up. Want to see how continuous topic modeling turns your support volume into a maintained help center? [See how Helpfeel works](/platform). ## Frequently asked questions ### Is topic modeling the same as cluster analysis? Yes. They describe the same technique: grouping unlabeled text into themes by similarity. Cluster analysis is the statistical name; topic modeling is the term that reads more clearly in English and says what you get, a set of topics. ### What data do you need to run topic modeling on support? Any text record of what customers ask: tickets, chat logs, search queries, contact form messages, or call transcripts. A year of history is a good starting sample. More data sharpens the themes but you rarely need every record. ### Do you need a data scientist to do topic modeling? Not anymore. Running the math by hand takes one. Productized tools now do the clustering, ranking, and labeling for you, so a support or CX practitioner can read the output and act on it directly. --- ## Sentiment Analysis for Customer Support Source: https://en.helpfeel.com/voc-analytics/sentiment-analysis # Sentiment Analysis for Customer Support Sentiment analysis reads the emotional tone in customer messages and sorts each one as positive, negative, or neutral. In support, it turns a pile of tickets and chats into a map of where customers are frustrated, so you can prioritize by pain instead of by volume alone. It is one layer of a [voice-of-customer analytics](/voc-analytics) practice. Where [topic modeling](/voc-analytics/topic-modeling) tells you what customers ask about, sentiment analysis tells you how they feel while they ask. Put the two together and the topics worth fixing first rise to the surface. ## What is sentiment analysis? Sentiment analysis is a text-mining technique that classifies the feeling behind a message. At its simplest it labels each message positive, negative, or neutral. More detailed setups break negative down further into shades like mild dissatisfaction, annoyance, and anger, which helps you tell a minor gripe from a customer about to churn. The value is in aggregate. One angry ticket is just one bad day. But when a quarter of the messages about a single topic read as negative, that pattern tells you something the raw volume never would. ## Why does sentiment analysis matter for support? Because volume alone is a blunt instrument. Your highest-volume topic might be a quick, painless question people are happy to self-serve. A lower-volume topic might be quietly enraging every customer who hits it. Sentiment analysis is how you tell those two apart. Layered onto your topics, it reweights your priorities. Instead of "fix the biggest pile first," you get "fix the topic that is both common and painful first." That is a far better use of a support team's limited time. It also gives you an early warning system. A rising share of negative sentiment on a topic often shows up before the churn or the escalation does. ## How does sentiment analysis work on support data? The workflow mirrors any voice-of-customer analysis, with a scoring step in the middle. 1. **Gather the text.** Pull tickets, chat transcripts, survey comments, or contact form messages. Any text where customers express how they feel works. 2. **Clean and prepare it.** Strip personal data and normalize the formatting. For call audio, transcribe to text first. 3. **Score the sentiment.** A model labels each message, and often each sentence, as positive, negative, or neutral, with a confidence score. 4. **Combine with topics.** Overlay sentiment on your [topic model](/voc-analytics/topic-modeling) so each topic carries both a volume and a sentiment reading. 5. **Prioritize and act.** Sort for topics that are high in both volume and negative sentiment. Those are your fastest wins for customer experience. One honest caveat: no sentiment model is perfect. Sarcasm, mixed emotions, and very short messages fool it. Trust the pattern across thousands of messages, and be skeptical of any single score. ## Sentiment vs topic modeling vs intent analysis These three techniques answer different questions about the same conversations, and they work best together. | Technique | Question it answers | What you get | | ------------------------------------------------- | -------------------------------- | ---------------------------------------------- | | [Topic modeling](/voc-analytics/topic-modeling) | What are customers asking about? | Ranked themes by volume | | **Sentiment analysis** | How do they feel about it? | Positive, negative, or neutral tone per topic | | [Intent analysis](/voc-analytics/intent-analysis) | What are they trying to do? | The goal behind the message, and the next step | [Intent analysis](/voc-analytics/intent-analysis) goes one step further and reads what the customer is trying to accomplish. Most teams add sentiment second, right after they can already see their topics. ## How Helpfeel uses customer sentiment Most sentiment reads are a one-time study, a snapshot until the next survey goes out. Helpfeel Analytics reads your support conversations continuously. It groups incoming questions into topics, weighs which ones customers struggle with, and rolls the results into voice-of-customer reports your product and marketing teams can act on. Sentiment is part of how the picture gets prioritized. The topics where customers strain the most are the ones Helpfeel flags first, drafting the missing or outdated help content for each gap and queuing it for a one-click human review. The same reading that tells you where the frustration is also produces the fix. If you want to see how continuous analysis turns customer feeling into a better help center, [see how Helpfeel works](/platform). ## Frequently asked questions ### What is sentiment analysis in customer support? Sentiment analysis reads the emotional tone in customer messages and sorts each one as positive, negative, or neutral. In support, it shows you which topics frustrate customers most, so you can prioritize fixes by pain, not just by volume. ### How accurate is sentiment analysis? Good enough to be useful, not perfect. Sarcasm, mixed messages, and short text trip it up. Treat sentiment scores as a strong signal across thousands of messages, not a verdict on any single one. ### What is the difference between sentiment analysis and topic modeling? Topic modeling tells you what customers ask about. Sentiment analysis tells you how they feel about it. Run them together and you can find the high-volume topics that also carry the most frustration. --- ## Intent Analysis in Customer Support Source: https://en.helpfeel.com/voc-analytics/intent-analysis # Intent Analysis in Customer Support Intent analysis is the practice of reading what a customer is actually trying to do. A search for "return window" and a search for "how long do I have to send this back" carry the same intent. Intent analysis groups messages by that underlying goal, so your support system can act on the need instead of matching keywords. This is the deepest layer of a [voice-of-customer analytics](/voc-analytics) practice. It builds on [topic modeling](/voc-analytics/topic-modeling), which tells you what customers ask about, and [sentiment analysis](/voc-analytics/sentiment-analysis), which tells you how they feel. Intent analysis answers the harder question: what are they trying to accomplish, and what should happen next? ## What is intent analysis? Intent analysis maps a customer's message, or their behavior, to the goal behind it. Where topic modeling sorts by subject, intent analysis sorts by purpose. Two people can raise the same topic with completely different intents: one wants to buy, the other wants to cancel. Same topic, opposite next steps. That distinction matters because intent is what you can act on. Knowing a customer's topic tells you where to file the conversation. Knowing their intent tells you what to do about it: surface an article, route to the right team, or send a proactive message. Good intent analysis reads more than the literal request. It reads the goal, ranks the most likely goals by confidence, and looks one step ahead to what the customer will probably need next. ## Why does intent analysis matter? Because most support stacks are reactive. Intent has to be spelled out, in a search, a chat, or a ticket, before anything happens. By the time a customer types the question, they're already stuck. The frustration has already started. The signal exists earlier. A customer who checks an order three times in ten minutes, or bounces between a spec page and their purchase history, is telling you their intent through behavior before they ever open a chat. Intent analysis is how you read that signal and help sooner. For a practitioner, this is the shift from sorting the past to shaping the next interaction. Topic and sentiment analysis describe what already happened. Intent analysis lets you get ahead of it. ## How does intent analysis work? Think of it as a loop with four stages. You can run the early stages as pure analysis today, and add the later ones as your practice matures. 1. **Detect.** Gather high-confidence signals of intent: search queries, chat messages, and behavioral cues like repeated visits, stalled checkouts, or account status. Start with a few trustworthy signals rather than a firehose of noise. 2. **Predict.** Map those signals to ranked intent hypotheses with a confidence score. For example: high confidence this customer is checking whether a spec change affects an order they already placed. 3. **Analyze.** Watch intent patterns over time. A cluster like "spec-change concerns" spiking after a document update is an insight your product and support teams can act on immediately. 4. **Act.** When confidence passes a threshold, respond from a fixed list of pre-approved actions. Surface an article, route the case, or send a proactive notice. Every action is authored and approved by a human in advance. The discipline in that last stage is what makes intent analysis safe to deploy. The system proposes, a person approves the playbook, and the system only ever executes moves from that approved list. Determinism is the point. ## Intent analysis vs topic modeling vs sentiment analysis The three techniques answer different questions about the same conversations. Intent analysis sits on top of the other two. | Technique | Question it answers | What you do with it | | ------------------------------------------------------- | -------------------------------- | ----------------------------------------- | | [Topic modeling](/voc-analytics/topic-modeling) | What are customers asking about? | Rank what to fix by volume | | [Sentiment analysis](/voc-analytics/sentiment-analysis) | How do they feel about it? | Prioritize the painful topics | | **Intent analysis** | What are they trying to do? | Act on the need, and predict the next one | Topic modeling and sentiment analysis describe your support conversations. Intent analysis acts on them. Most teams build the first two layers first, then reach for intent analysis when they want support to get ahead of the customer instead of reacting after the fact. ## How Helpfeel applies intent analysis Intent prediction is the core of how Helpfeel works. Its search reads the intent behind a query rather than matching keywords, which is why a customer finds the right answer even when they phrase the question in a way your help center never anticipated. That intent-first retrieval is the engine under the whole platform. From there, the same intent signal feeds the rest of the loop. Helpfeel detects intent from search and behavior, predicts the most likely goal, surfaces intent patterns for your team to review, and acts only through playbooks your experts have approved in advance. A person sets the rules, the system executes within them, and every action stays auditable. That's intent analysis as an operating layer rather than a one-off study. If you want to see intent-first support in practice, [see how Helpfeel works](/platform). ## Frequently asked questions ### What is the difference between intent analysis and topic modeling? Topic modeling groups messages by subject, telling you what customers ask about. Intent analysis reads the goal behind the message, what the customer is trying to accomplish, so a support system can act on the need rather than just categorize it. ### Can you detect intent before a customer contacts you? Yes. Behavior carries intent before words do. Repeated visits to a spec page, a stalled checkout, or an order-status check all signal a goal. Predicting intent from these signals lets you help before frustration sets in. ### Does intent analysis require you to act automatically? No, and it should not act on its own. The safe pattern is human-approved actions only: the system detects and predicts intent, then executes from a fixed list of pre-approved responses. Nothing is improvised. --- ## Keyline: An AI Reading Assistant for Your PDFs and Docs Source: https://en.helpfeel.com/docs/keyline # Keyline: An AI Reading Assistant for Your PDFs and Docs Keyline is an AI reading assistant for your PDFs and Office documents. Upload a file and Keyline reads every page as an image, taking in the figures, diagrams, and tables along with the text. Ask a question and the answer comes back cited to the block it came from, with the page number and a cropped image of that block. Keyline is built for the moments when you need an answer from inside a document, or from across a whole library of documents, faster than reading page by page. ## Two ways in One engine reads your documents, and there are two ways to reach it. Every page in this documentation is written for one of them, and says which at the top. - **The Keyline app** is the hosted interface you sign in to. You upload files into a library, read them in the PDF view, and ask questions in a conversation panel beside the page. Keyline renders the answer, the page it came from, and the cropped block behind it. The [product page for teams](/keyline/teams) covers what it is and how teams put it to work. - **The Keyline API** is the same retrieval engine over HTTP, with no interface of its own. Your agent, your application, or your internal tool makes the call and receives text, page numbers, and image crops to render inside the interface you already ship. [Keyline for developers](/keyline) carries that argument. The two share a corpus rather than holding separate ones: an API call is scoped by a project id you copy out of the app's project settings page. Where the app says library, the API says project. ## What Keyline can do - **Read a single document deeply.** Keyline understands page layout, figures, diagrams, and tables, so a question about a chart gets a real answer. - **Answer with citations.** Every claim comes back with its page number and a cropped image of the block behind it, so you can verify it in seconds. - **Search across many files at once.** A multi-file search index lets you ask a question once and get results pulled from every relevant document in your library. - **Stay organized.** Libraries and bundles keep your uploaded files grouped so you can find and reuse the right set later. ## Using the app - [Getting started](/docs/keyline/getting-started): upload your first document and see what happens next. - [Reading and analyzing PDFs](/docs/keyline/reading-pdfs): the PDF view, page navigation, and how Keyline reads figures and tables. - [Conversations and citations](/docs/keyline/conversations-and-citations): asking questions and following an answer back to its source. - [Searching across multiple files](/docs/keyline/multi-file-search): asking one question across your whole library. - [Organizing your library](/docs/keyline/organizing-your-library): libraries and bundles. ## Building on the API - [Keyline API quickstart](/docs/keyline/api-quickstart): your first cited answer over HTTP, and putting your own documents in. - [The agent runbook](/docs/keyline/agent-runbook): the whole API contract on one page, written for a coding agent to read in a single pass. ## Both surfaces - [FAQ](/docs/keyline/faq): file types, size limits, sharing, languages, the API, and troubleshooting. --- ## Getting Started with the Keyline App Source: https://en.helpfeel.com/docs/keyline/getting-started # Getting Started with the Keyline App The fastest way to understand Keyline is to upload a document and ask it a question. This page walks through that first upload in the Keyline app, what happens behind the scenes, and where to start reading once it is done. To do the same thing from your own code instead, start with the [Keyline API quickstart](/docs/keyline/api-quickstart). ## Upload your first document Open your library in the app and choose the upload option, then pick a file from your computer. Keyline accepts PDF, PNG, JPEG, and Office files (DOCX, XLSX, PPTX, DOC, XLS, PPT). Preliminary per-file size and page limits apply. Once the file starts uploading, you will see a progress indicator while Keyline processes it. ## What happens after you upload Uploading a file does two things at once: 1. **Analysis.** Keyline reads every page as an image, so the figures, diagrams, and tables on it are understood alongside the text. 2. **Indexing.** Keyline builds a search index for the file so it can be found later, both on its own and as part of a [multi-file search](/docs/keyline/multi-file-search) across your whole library. Both steps run automatically. You do not need to do anything else once the upload starts; Keyline shows progress and the document becomes readable as soon as processing finishes. ## Where to start reading Once a document finishes processing, open it to reach the PDF view, where you can page through the document and start a conversation about it. See [Reading and analyzing PDFs](/docs/keyline/reading-pdfs) for how the page view works, or jump straight to [Conversations and citations](/docs/keyline/conversations-and-citations) to start asking questions. --- ## Reading and Analyzing PDFs Source: https://en.helpfeel.com/docs/keyline/reading-pdfs # Reading and Analyzing PDFs The PDF view in the Keyline app is where you read a document and start a conversation about it. This page covers the page view itself, how to navigate a long document, and how Keyline reads the figures, diagrams, and tables on a page. The reading described here is what the [Keyline API](/docs/keyline/api-quickstart) returns to your own interface. ## The PDF view Opening a processed document shows the page images alongside a conversation panel. The page view loads progressively as you scroll, so long documents open quickly instead of waiting for every page to render at once. ## Page navigation You can move through a document page by page, or jump directly to a specific page. When an answer cites a page, following that citation takes you straight to it, so you never need to scroll to find the source of an answer manually. ## Figures, diagrams, and tables Keyline reads a page as a whole image and detects the semantic blocks on it: text, figures, tables, and equations. A chart, a diagram, or a table is understood in context, so a question like "what does the trend in this chart show" gets a real answer, and the block itself is what comes back as the citation. This is the same reading that powers [conversations and citations](/docs/keyline/conversations-and-citations) and [multi-file search](/docs/keyline/multi-file-search). --- ## Conversations and Citations Source: https://en.helpfeel.com/docs/keyline/conversations-and-citations # Conversations and Citations Once a document has finished processing, you can ask it questions directly instead of reading it start to finish. This page covers how to ask a question in the Keyline app, how to read the answer, and how to trace an answer back to its source page. The [Keyline API](/docs/keyline/api-quickstart) returns the same answers and references for your own interface to render. ## Asking a question Open a document's page view in the app and type a question into the conversation panel. You can ask about a specific detail, a figure or table, or something that spans several pages. Keyline keeps the conversation history for the document, so you can ask a follow-up question without repeating context. ## Reading citation-backed answers Every answer carries numbered citations, and each one resolves to a single block of the document: a paragraph, a figure, a table, or an equation. A citation gives you the page number and a cropped image of that block, so you verify the claim against the original artwork in seconds. That matters most for technical or numerical details, where the answer lives inside a figure or a table. ## Following a citation back to the source Select a citation in an answer to jump straight to the page it references in the [PDF view](/docs/keyline/reading-pdfs). This closes the loop between an answer and the document it came from, so you always know exactly where information originated. --- ## Searching Across Multiple Files Source: https://en.helpfeel.com/docs/keyline/multi-file-search # Searching Across Multiple Files A single document is easy to search once you have opened it. Multi-file search in the Keyline app lets you ask a question across every document in your library at once, so you do not have to open each file individually to find what you need. Through the [Keyline API](/docs/keyline/api-quickstart) this is the default: a call searches the whole project rather than one file. ## Searching across your library Start a search from your library instead of from a single document. Keyline searches every processed file at once and returns the passages, figures, and tables most relevant to your question, each with a citation back to its source document and page. ## How the multi-file index works Every file you upload is added to a shared search index as part of [processing](/docs/keyline/getting-started). This index is what makes cross-file search fast: instead of re-reading every document at search time, Keyline searches the index and then confirms the most relevant matches directly against the source pages. ## Grouping results into bundles When a search turns up a useful set of files, you can group them into a bundle so you can come back to that same set later without searching again. See [Organizing your library](/docs/keyline/organizing-your-library) for more on bundles. --- ## Organizing Your Library Source: https://en.helpfeel.com/docs/keyline/organizing-your-library # Organizing Your Library As you upload more files, keeping them organized makes it easier to find the right document, or the right set of documents, later. The Keyline app offers two ways to do this: libraries and bundles. A [Keyline API](/docs/keyline/api-quickstart) call is scoped by a project id you copy from the app's project settings page, so how you organize here decides what a call can reach. ## Libraries A library is the main collection your uploaded files live in. Every file you upload belongs to a library, and every file in a library is searchable both on its own and as part of [multi-file search](/docs/keyline/multi-file-search). ## Bundles A bundle is a smaller, named set of files pulled from your library, useful when you want to keep coming back to the same group of documents. Bundles are commonly created straight from a [search](/docs/keyline/multi-file-search) result, so a useful set of matches does not have to be found again from scratch. ## Choosing between a library and a bundle Use your library for everything you upload; it is the full, searchable collection. Use a bundle when you have identified a smaller set of files, for example everything relevant to one topic or project, that you want to revisit without re-searching your whole library each time. --- ## Keyline FAQ Source: https://en.helpfeel.com/docs/keyline/faq # Keyline FAQ Common questions about the documents Keyline reads, the Keyline app, the Keyline API, and what to check when an answer looks wrong. ## Documents, whichever way you reach them These answers hold for both the [Keyline app](/docs/keyline/getting-started) and the [Keyline API](/docs/keyline/api-quickstart), because both read the same documents the same way. ### What file types does Keyline support? Keyline accepts PDF, PNG, JPEG, and Office files (DOCX, XLSX, PPTX, DOC, XLS, PPT). Upload one and Keyline analyzes it and builds a search index automatically; see [Getting started](/docs/keyline/getting-started) for the upload flow. ### Is there a file size or page limit? Per-file size and page limits apply. They are preliminary and get adjusted to fit customer needs, so tell us what you are working with. Your library itself keeps growing as you add files, and if an upload is rejected for size or length, split the document and upload the parts. ### Will Keyline read scanned documents? Yes. Every page is analyzed as an image, so a scan is an ordinary case. Architectural, electrical, and wiring drawings are a confirmed failure case in our own testing, so build elsewhere for those. ### What languages does Keyline support? Keyline can read and answer questions in multiple languages, including English and Japanese. You can ask a question in a different language than the source document. ### What is the difference between the Keyline app and the Keyline API? The Keyline app is the hosted interface you sign in to, where you upload files into a library, read them, and ask questions beside the page. The Keyline API is the same retrieval engine over HTTP with no interface of its own: it returns text, page numbers, and image crops for your own agent, application, or internal tool to render. An API call is scoped by a project id you copy out of the app, so the two work over the same documents. ## Using the Keyline app ### Who can see the files I upload? Uploaded files and their conversations are visible to the members of the library you upload into. They stay inside that library. ## Building on the Keyline API These answers cover Keyline as an API you call from your own code. The landing page for that audience is [Keyline for developers](/keyline), the hands-on first run is the [Keyline API quickstart](/docs/keyline/api-quickstart), and the call-by-call reference is the [agent runbook](/docs/keyline/agent-runbook). ### Does the Keyline API come with a user interface? No. The API returns text, page numbers, and cropped images over HTTP, and your application renders them wherever your users already are: a chat transcript, a ticket view, a panel in your own tool. The Keyline app is what that data looks like when Helpfeel renders it. ### How does the Keyline API differ from a document parsing API? Parsing tools convert a document you hand them, per call. Keyline answers a question across a library you indexed earlier and decides for itself which pages to read. For structured output from one known file, a parsing API is the right tool. ### Do I have to bring my own vector store? Keyline builds and owns its own index over the documents you send it. Embeddings, chunking, and the database are handled for you. ### Can I use my own model? The reasoning that decides what to read is part of the product. Keyline returns text and images to your application, and you use whatever model you like from there. ### How do I know when a document is ready? Poll its status. Documents become deep-readable slightly before the library-wide search index finishes, so there are two states that both mean usable. ### What does a citation look like? A numbered reference that resolves to a file, a page number, and a cropped PNG of the block the claim came from. Region coordinates for that block arrive with the preview. ### Is there a free tier? Public Preview access comes with $50 in free credit, then usage-based pricing. Indexing has a real per-page cost, so pricing stays usage-based after the credit runs out. Rates and published limits live on the [pricing page](/keyline/pricing). ### How do I get an API key today? Sign up for the Public Preview. Everyone on the list gets a key and the documentation the day it opens. ## Troubleshooting ### My upload seems stuck Indexing runs once, on ingest, and a long document takes longer than a short one. Uploads also queue behind the files ahead of them, so a bulk import finishes in the background rather than all at once. Check the file's status before re-uploading it. ### Keyline's answer did not match what I expected Open the citation to see the block it came from: the page number plus a cropped image of the figure, table, or paragraph behind the claim. If that block does not support the answer, rephrase the question more specifically, for example naming the figure, table, or section you are asking about. --- ## Keyline API Quickstart: Your First Call Source: https://en.helpfeel.com/docs/keyline/api-quickstart # Keyline API Quickstart: Your First Call You have a key. This page takes you from an empty shell to a cited answer over HTTP, then to putting your own documents in. The API ships no interface of its own. It answers a question and hands back text, page numbers, and image crops for your agent, your application, or your internal tool to render, so what a user sees is the surface you already ship. The [Keyline app](/docs/keyline/getting-started) is what that looks like when we render it. If you are still deciding whether to bother, [Keyline for developers](/keyline) carries that argument, and the rates live on the [pricing page](/keyline/pricing). ## Three things you need Every call is scoped to a project and authorized by one header. A token cannot discover the base URL or the project id for you, so all three arrive out of band with your Public Preview access. | Value | Where it comes from | | ---------- | -------------------------------------------------------- | | Base URL | Provided with your Public Preview access | | Project ID | Copied from the project settings page in the Keyline app | | Token | Issued with your Public Preview access | ```bash export KEYLINE_API_URL="..." # provided with your Public Preview access export KEYLINE_PROJECT_ID="..." # project settings page in the Keyline app export KEYLINE_TOKEN="plms_..." # issued with your access ``` A token is the prefix `plms_` followed by 43 URL-safe characters. It is shown once, at creation and at reset, so store it as a secret; regenerating one invalidates the old one immediately, with no overlap window. Calls are server to server, with no CORS and no browser SDK. Without a key yet? Sign up for the [Public Preview](/keyline#signup) and we will send one with the base URL and your project id. ## Your first call Agentic search against a project that already has documents in it. `query` is the only required field: the project is the corpus, with no per-call file filter, no `topK`, and no depth cap, so cost scales with library size. Scope a project to the documents that matter rather than holding everything in one. For multi-turn conversations, pass `history` and the `lastRefNumber` you used, so citation numbering continues across turns. ```bash curl -N -X POST \ "$KEYLINE_API_URL/api/projects/$KEYLINE_PROJECT_ID/search/agentic-search" \ -H "Authorization: Bearer $KEYLINE_TOKEN" \ -H "Content-Type: application/json" \ -d '{"query": "What is the maximum operating pressure?"}' ``` ```ts const base = process.env.KEYLINE_API_URL; const project = process.env.KEYLINE_PROJECT_ID; const res = await fetch(base + "/api/projects/" + project + "/search/agentic-search", { method: "POST", headers: { Authorization: "Bearer " + process.env.KEYLINE_TOKEN, "Content-Type": "application/json", }, body: JSON.stringify({ query: "What is the maximum operating pressure?" }), }); const decoder = new TextDecoder(); for await (const bytes of res.body) { process.stdout.write(decoder.decode(bytes, { stream: true })); } ``` ```python import os import httpx url = ( os.environ["KEYLINE_API_URL"] + "/api/projects/" + os.environ["KEYLINE_PROJECT_ID"] + "/search/agentic-search" ) headers = {"Authorization": "Bearer " + os.environ["KEYLINE_TOKEN"]} body = {"query": "What is the maximum operating pressure?"} with httpx.stream("POST", url, headers=headers, json=body, timeout=None) as res: for line in res.iter_lines(): print(line) ``` ## Reading the stream The response is Server-Sent Events with two event names on the wire, `chunk` and `done`, and `chunk` carries a `phase` discriminator. This is where most of your integration time goes. ```text event: chunk data: {"phase":"info","searchScope":{"projectId":"...","libName":"Acme Docs"}} event: chunk data: {"phase":"process","type":"tool_call","toolName":"vector_search","toolArgs":{...},"text":"..."} event: chunk data: {"phase":"process","type":"tool_result","toolName":"resolve_search_hit","toolResponse":{...}} event: chunk data: {"phase":"final","text":"... [1] ... [2]"} event: done data: {"success":true,"refs":[{"refNumber":1,"fileHash":"4c8ae...","vpId":"p12_vp00042", "hitId":"Financial Report\tRevenue\t[Quarterly Trends]","fileName":"FY26-Q3.pdf"}]} ``` | Event | What to do with it | | ------------------------------------------ | -------------------------------------------------------------------------------------------------- | | `chunk, phase: info` | The search scope for this run. Label your view with it, or drop it | | `chunk, phase: process, type: tool_call` | The loop narrating its own steps. Surface it, because a long query should not be silent | | `chunk, phase: process, type: tool_result` | Intermediate analysis, useful for a live view. Buffer by unit identifier, because units interleave | | `chunk, phase: final` | The answer, streamed token by token, carrying inline `[1]` and `[2]` markers. Append it | | `done` | Ends the stream and carries `refs`, the structured reference list. Resolve citations here | Disconnecting aborts the work server side, so abandoning a stream stops the spend. When nothing relevant is found the answer says so, and that is a valid outcome to pass through honestly. Internally the run is a Gemini tool loop of up to fifty turns; you consume its output rather than driving it. ## Getting a citation on screen The `final` text carries inline `[1]` and `[2]` markers, and the `done` event's `refs` array resolves each number to a `fileHash`, a `vpId`, and a `fileName`. The page number comes from the `vpId` string convention rather than a field. Ids are built as `p_vp`, so `p12_vp00042` is physical page 12, and snippet ids follow the same shape, with `p12_s3` being page 12, snippet 3. ```ts const ID = /^p(\d+)_(?:vp|s)(\d+)$/; function pageOf(id) { const match = ID.exec(id); if (!match) throw new Error("unrecognized identifier: " + id); return Number(match[1]); } pageOf("p12_vp00042"); // 12, physical page 12 pageOf("p12_s3"); // 12, page 12 snippet 3 ``` ```python import re ID = re.compile(r"^p(\d+)_(?:vp|s)(\d+)$") def page_of(identifier): match = ID.match(identifier) if not match: raise ValueError("unrecognized identifier: " + identifier) return int(match.group(1)) page_of("p12_vp00042") # 12, physical page 12 page_of("p12_s3") # 12, page 12 snippet 3 ``` ```bash # FILE_HASH and the snippet id both come from the done event's refs array. curl -L -o snippet.png \ "$KEYLINE_API_URL/api/projects/$KEYLINE_PROJECT_ID/files/$FILE_HASH/snippets/p12_s3.png" \ -H "Authorization: Bearer $KEYLINE_TOKEN" ``` The snippet filename is validated against `/^p\d{1,4}_s\d{1,4}\.png$/`, so you build the URL from a snippet id the server already gave you. The response is the cropped region, or a redirect to a signed GCS URL. You get the page and the figure it came from, so you can show the crop; drawing a highlight box on the full page arrives with the Public Preview. ### Never fabricate a citation Cite only identifiers the server returned. Do not construct, guess, adjust, or interpolate one, and do not carry one over from a previous answer. A citation that does not resolve is worse than no citation, because the whole value here is that a claim can be checked. ## Putting documents in This comes after your first query on purpose: a preview key arrives with a project already indexed. Upload returns immediately and everything after it is asynchronous, in four steps. 1. `POST .../upload/init` with `{ fileName, size, mimeType, fileHash }`, where `fileHash` is a client-computed SHA-256 matching `/^[a-f0-9]{64}$/`. You get back a GCS V4 signed PUT URL, or `{ "mode": "legacy" }`, which sends you to `POST .../upload/queue` with a multipart body. 2. PUT the bytes to the signed URL, sending exactly the extension headers you were given. They are part of the signature, and `x-goog-content-length-range` is always one of them. The URL expires, so do not hold it. 3. Poll `GET .../files/:fileHash` every few seconds. There are no webhooks. 4. The status ladder runs `uploading`, `converting`, `converted`, `processing_snippets`, `processing_index`, `processing_index_flow`, `processing_virtual_pages`, `indexing`, `completed`, plus `error`. ```bash FILE_HASH=$(shasum -a 256 manual.pdf | cut -d' ' -f1) # 1. Ask for an upload target. curl -X POST "$KEYLINE_API_URL/api/projects/$KEYLINE_PROJECT_ID/upload/init" \ -H "Authorization: Bearer $KEYLINE_TOKEN" \ -H "Content-Type: application/json" \ -d "{\"fileName\":\"manual.pdf\",\"size\":$(wc -c r.json()); // PUT the bytes to the signed target in init, echoing back exactly the headers it was // signed with, then poll. Stop at indexing. const file = await fetch(base + "/api/projects/" + project + "/files/" + fileHash, { headers: auth, }).then((r) => r.json()); ``` ```python import hashlib, os import httpx base = os.environ["KEYLINE_API_URL"] + "/api/projects/" + os.environ["KEYLINE_PROJECT_ID"] auth = {"Authorization": "Bearer " + os.environ["KEYLINE_TOKEN"]} data = open("manual.pdf", "rb").read() file_hash = hashlib.sha256(data).hexdigest() init = httpx.post(base + "/upload/init", headers=auth, json={ "fileName": "manual.pdf", "size": len(data), "mimeType": "application/pdf", "fileHash": file_hash, }).json() # PUT the bytes to the signed target in init, echoing back exactly the headers it was # signed with, then poll. Stop at indexing. status = httpx.get(base + "/files/" + file_hash, headers=auth).json() ``` ### Two statuses mean usable. `indexing` already means analysis is done and the document is deep-readable. The worker is still building the library-wide hierarchy and title index, which is what `completed` marks. Waiting for `completed` leaves the document unavailable longer than it needs to be. A duplicate name in the processing queue is rejected, and a file already being processed is locked with a 409. Treat both as already in flight. ## The other two retrieval calls Pick by what you know before you ask. | You know | Use | Shape | | --------------------------------------------------- | -------------- | ----------------------------------------------------------------------------------------- | | Nothing. You have a library and a question | Agentic search | Streamed. A reasoning loop that searches, resolves, reads, and decides when it has enough | | Which file or bundle holds the answer | Deep read | Streamed. A fixed pipeline that reads inside that scope in depth | | You only want to know which files mention something | Vector search | One JSON response, no reasoning, cheapest by a wide margin | ### Deep read `POST .../search/deep-read`, streamed, with a body of `{ message, fileHash | (bundleId + fileHashes), isFollowUp, conversationId }`. `message` is required, and one of `fileHash` or `bundleId` is required. In bundle mode `fileHashes` is required, non-empty, and a subset of the bundle. ### Deep read cites differently from agentic search. Its citations are inline `[[p1_s3]]` markers embedded in the markdown, and its `done` event carries only `{ success, timestamp, message, chatLink }`, so a consumer regexes the answer text to recover them. Its progress analyzes several units in parallel, so per-unit output interleaves and has to be buffered by unit identifier. Deep read also leaves persistence to you: inference and saving are separate calls. ### Vector search `GET .../search/vector-search?q=`, plain JSON, no reasoning, cheapest by a wide margin. Hits come back grouped by file, each a topic path through that document's hierarchy with a raw LanceDB `_distance`, top 2 per table and capped at 20 tables. It returns no page numbers, and `hierarchy/resolve_hit` is off the token allowlist, so reach for agentic search when you need the page and let the server run the loop. ## Errors you will actually hit In roughly the order you will meet them. 1. **Everything returns 401.** Every auth failure returns a byte-identical 401, and invalid, revoked, never-issued, and not-permitted are deliberately indistinguishable. Do not write logic that discriminates them, and do not retry hoping for a different error. Check the token prefix and length, then the project id, then whether the endpoint is one a token may reach at all. 2. **A file never becomes ready.** Check the format and the size and page limits. A rejected format fails at upload, before any processing starts. 3. **You uploaded the same document twice.** A duplicate name in the processing queue is rejected, and a file already being processed is locked with a 409. Both mean already in flight. 4. **Interleaved progress looks like nonsense.** It is parallel analysis. Buffer by unit identifier before you display anything. 5. **The conversation was not saved.** Deep read does not persist. Accumulate the final text and save it in a second call. 6. **The answer says nothing relevant was found.** Often correct, and a valid outcome to pass through. Architectural, electrical, and wiring drawings are a confirmed failure case. 7. **Error text arrives in Japanese.** Some server messages are not fully translated yet. Match on status codes rather than on message strings. ## Limits, and what a token cannot do Published, so you can plan against them. | | | | ---------------------- | ----------------------------------------------------------- | | Accepted formats | PDF, PNG, JPEG, and Office: DOCX, XLSX, PPTX, DOC, XLS, PPT | | File size and pages | Preliminary, and adjusted to fit customer needs | | Throughput | Uploads queue and index in the background | | Rate limits and quotas | Be a considerate caller and back off on errors | ### In this preview a token cannot: - Delete anything: not files, not bundles, not sessions. - Rename a file. - Save a conversation attached to a file. Bundle and saved-session conversations can be saved. - Read text-extraction output, page dimensions, or region coordinates. - Resolve a vector-search hit to page content. - Manage members, or grant anyone any access. - Discover which organization or project it belongs to. Every one of those returns the same opaque 401, so probing teaches you nothing. Design around them. Rates and the published limits live on the [pricing page](/keyline/pricing). ## Next [The agent runbook](/docs/keyline/agent-runbook) is this page written for a coding agent to read in one pass: the whole contract, no code blocks. [Keyline](/keyline) covers what the retrieval engine does and how it is built. [Contact us](/contact) when a call does not behave the way this page says it will. Ready for a key? [Sign up for the Public Preview](/keyline#signup). --- ## Keyline API for Agents: The Runbook Source: https://en.helpfeel.com/docs/keyline/agent-runbook # Keyline API for Agents: The Runbook This page is written to be read by an AI agent as well as a person. It is the whole integration on one page: how to authenticate, the three retrieval calls, what comes back, and the failure modes worth knowing before you hit them. If you are a person deciding whether to bother, read [Keyline for developers](/keyline) instead. If you want to run the calls by hand first, start with the [Keyline API quickstart](/docs/keyline/api-quickstart). **Status: the Public Preview opens soon.** Everyone on the signup list gets a key, the command-line tool, and the request and response samples the day it opens. This page documents the contract itself, which is true now and will not change when the samples land. ## What the Keyline API is, for an agent The Keyline API is a retrieval tool. You call it with a natural-language question. It searches a document library that was indexed earlier, reads the pages worth reading, and returns an answer with the page and the cropped image region each claim came from. It renders nothing. Text, page numbers, and image crops come back over HTTP for the calling program to display, so the answer appears inside whatever interface you already have: a chat transcript, a ticket view, a sidebar in your own tool. Helpfeel's own rendering of it is the [Keyline app](/docs/keyline/getting-started). Reach for it when: - The answer is inside a figure, a diagram, or a table, where text extraction loses it. - The question spans a library and you do not know which file holds the answer. - The user will ask where the answer came from, and a page number alone is not enough. Do not reach for it when: - You already have the text and just need a model to reason over it. - You need structured field extraction from one known document. That is a parsing job. - The documents are mostly prose. Ordinary text search is fine and cheaper. ## Core concepts, in dependency order | Term | What it means | | ------------ | -------------------------------------------------------------------------------------------------------------------------- | | Organization | Billing and membership boundary | | Project | The tenancy unit that owns documents. Every data call is scoped to one project. A project is the corpus | | File | One uploaded document, addressed by a content hash | | Snippet | One cropped semantic block of a page: text, figure, table, equation. The atom of a citation | | Virtual page | A logical group of snippets that can span a physical page break, so a figure and the paragraph explaining it stay together | | Bundle | Several files treated as one target for a deep read | | Info | A saved investigation session you can resume later | The two that matter for citations are **snippet** and **virtual page**. A snippet is what gets cropped and returned as an image. A virtual page is what the retrieval loop decides to read. ## Authentication One header. Server to server only. - Credential: a project-scoped bearer token issued to a service account. - Format: the prefix `plms_` followed by 43 URL-safe characters. - The token carries no permissions of its own. Access is read from the live project grant on every request, so a revoked grant takes effect immediately. - The token is shown once, at creation and at reset. Store it as a secret. - Regenerating a token invalidates the old one immediately. There is no overlapping validity window, so plan the swap. Two behaviors to code against: 1. **Bearer never falls back to a session.** If a token is malformed, unknown, or revoked, the request fails. It does not degrade into an anonymous request. 2. **Every failure returns an identical unauthorized response.** Invalid, revoked, never-issued, and not-permitted are deliberately indistinguishable. Do not write logic that tries to tell them apart, and do not retry hoping for a different error. **No CORS. Do not call this from a browser.** There is no browser SDK and browser usage is out of scope by design. ## Configuration your integration needs Supply all three out of band. A token cannot discover them for you. | Value | Where it comes from | | ---------- | -------------------------------------------------------- | | Base URL | Provided with your Public Preview access | | Project ID | Copied from the project settings page in the Keyline app | | Token | Issued with your Public Preview access | The project ID is a real onboarding step. A token cannot enumerate the organizations or projects it belongs to, which is a deliberate isolation property. ## The three retrieval calls Pick by what you know before you ask. | You know | Use | Shape | | --------------------------------------------------- | -------------- | ----------------------------------------------------------------------------------------- | | Nothing. You have a library and a question | Agentic search | Streamed. A reasoning loop that searches, resolves, reads, and decides when it has enough | | Which file or bundle holds the answer | Deep read | Streamed. A fixed pipeline that reads inside that scope in depth | | You only want to know which files mention something | Vector search | One response, no reasoning, cheapest by a wide margin | ### Agentic search The default choice for an agent. One request carries the question, optional prior conversation turns, and the last reference number you used so numbering continues across turns. The project is the corpus and there is no per-call file filter. Internally it runs a tool loop: search the topic index, resolve promising hits into real content, report progress, and stop when the evidence is sufficient. You do not drive the loop. You consume its output. **Cost scales with library size and you cannot cap it from the request.** There is no result-count or depth parameter. On a large library, prefer a project scoped to the documents that matter over one project holding everything. ### Deep read Use when your agent has already identified the file, or when the user named it. Pass the question plus the file, or a bundle plus the specific files within it. Follow-ups pass the root conversation identifier and the server loads the prior turns as context. Two things to handle: - The progress stream analyzes several units **in parallel**, so per-unit output interleaves. Buffer by unit identifier before displaying anything. - **Deep read does not save the conversation for you.** Inference and persistence are separate calls today. If you want the exchange kept, accumulate the final text and save it in a second request. ### Vector search One call, one response, no model reasoning. It returns hits grouped by file, where each hit is a topic path through that document's hierarchy with a distance score. Good for "which of these files talk about this" and for narrowing before a deep read. It does not return page numbers. Resolving a hit to actual page content is not available over a token in this preview, so do not design your own retrieval loop on top of vector search yet. Use agentic search for that and let the server run the loop. ## Consuming a streamed answer Both agentic search and deep read stream. Three kinds of event arrive: | Event | What to do with it | | -------------- | ----------------------------------------------------------------------------------------------------------------- | | Progress | Surface it. A long query should not be silent. This is the loop narrating its own steps | | Working output | Intermediate analysis, useful for a live view. Not the answer. Buffer by unit identifier because units interleave | | Final answer | The synthesized answer, streamed token by token, carrying inline numbered citations | | Completion | Ends the stream and carries the structured reference list | | Error | Ends the stream. Surface the failure, do not silently return a partial answer as if it were complete | Disconnecting aborts the work server side, so abandoning a stream does not keep burning cost. **When nothing relevant is found, the answer says so.** The pipeline does not fabricate an answer from an empty result. Treat a no-information response as a valid outcome and pass it through honestly rather than re-asking with a looser question. ## Citations, and the one rule that matters Answers carry numbered citations that resolve, through the completion event's reference list, to a file and a unit of content. From there: - **Page number** is available for every citation. - **The cropped image of the cited block** is retrievable as a PNG. This is the highest-value thing the API returns and the reason to prefer it over a text snippet. - **Region coordinates** inside the page are not available over a token in this preview. They arrive with the Public Preview. Until then you can show the crop, but you cannot draw your own highlight box on the full page. ### Never fabricate a citation Only cite identifiers the server actually returned. Do not construct, guess, adjust, or interpolate a citation identifier, and do not carry one over from a previous answer. A citation that does not resolve is worse than no citation, because the whole value of this API is that a claim can be checked. ### Rendering guidance that holds up in practice - **Download images to local paths rather than embedding remote URLs.** You cannot control which renderer the user sees, and a local path renders everywhere. Make the download idempotent: if the file is already there, reuse it. - **Put the crop immediately next to the sentence that cites it.** The crop is the argument. Placed at the bottom in a reference list, it is decoration. - **Show the retrieval loop's progress before the answer lands.** Surface which pages are being read as soon as you know, so the user can follow along instead of waiting. - **Do not expose internal vocabulary to end users.** Words like snippet and virtual page are for you. Say "the table on page 212" to the person reading the answer. ## Ingesting documents Upload returns immediately. Everything after it is asynchronous and your integration has to handle that. 1. Request an upload. You get either a direct upload target to send the bytes to, or an instruction to fall back to a multipart upload. 2. Send the bytes. If you were given a direct target with required headers, send exactly those headers. They are part of the signature and the upload fails without them. The target expires, so do not hold it. 3. Poll the file's status until it is ready. **There are no webhooks.** Polling every few seconds is the expected pattern. 4. Note that a document becomes deep-readable slightly before the library-wide search index finishes. Two distinct statuses both mean usable, and treating only the final one as ready leaves the document unavailable longer than it needs to be. Practical constraints to plan against: | | | | ------------------- | ----------------------------------------------------------- | | Accepted formats | PDF, PNG, JPEG, and Office: DOCX, XLSX, PPTX, DOC, XLS, PPT | | File size and pages | Preliminary, and adjusted to fit customer needs | | Throughput | Uploads queue and index in the background | | Region | US region available now | Per-file size and page limits are preliminary and get adjusted to fit customer needs, so do not hard-code a threshold and do not predict how long an import will take. Upload the file, poll its status, and treat a rejected upload as the limit response. A large first import takes real time. Index a meaningful subset, verify it answers a real question, then widen. ## What you are billed for Worth knowing if you are planning a large ingest, because indexing is the cost that scales with your library rather than with your traffic. | Unit | Rate | | -------- | ----------------------- | | Indexing | $0.06 per page | | Query | From $0.05 per question | | Storage | $0.05 per GB per month | Indexing is charged per page on first registration, and again if you update a document. It is a one-time cost per page: querying the same library repeatedly does not re-index anything. Storage is prorated daily and includes the index alongside the file. Public Preview access comes with $50 in free credit, roughly 800 pages indexed with room to ask a few hundred questions. Practical consequence for an agent: prefer indexing a scoped subset that answers a real question over indexing everything available. A fifty-thousand-page library is a meaningful first invoice, and you cannot un-index it for free. ## What a token cannot do The bearer token reaches an enumerated set of endpoints. Anything outside it is refused even when the account's role would otherwise allow it. This is transport hardening layered on top of normal permissions, and it is expected to relax as rate limiting and audit logging land. In this preview a token **cannot**: - Delete anything. Not files, not bundles, not sessions. - Rename a file, or save a conversation attached to a file. Conversations attached to bundles and to saved sessions can be saved. - Read text-extraction outputs, page dimensions, or region coordinates. - Resolve a vector-search hit into page content. - Manage members, or grant anyone any access. - Discover which organization or project it belongs to. Design around all of these rather than probing for them. Every one returns the same opaque unauthorized response, so probing teaches you nothing. ## Failure modes, in the order you will meet them 1. **Everything returns unauthorized.** Check the token prefix and length first, then the project ID, then whether the endpoint is one a token may reach at all. The response is identical for all three causes. 2. **A file never becomes ready.** Confirm the format is one of PDF, PNG, JPEG, or an Office format (DOCX, XLSX, PPTX, DOC, XLS, PPT), and that it is under the size and page limits. A rejected format fails at upload, before any processing starts. 3. **Uploading the same document twice.** Duplicate names in the processing queue are rejected, and a file already being processed is locked. Treat both as "already in flight" rather than as errors to retry. 4. **Interleaved progress output looks like nonsense.** It is parallel analysis. Buffer by unit identifier. 5. **The conversation was not saved.** Deep read does not persist. That is a second call. 6. **An answer says nothing relevant was found.** Often correct, especially on architectural, electrical, and wiring drawings, which are a confirmed failure case. Do not paper over it with a vaguer question. 7. **Error text arrives in Japanese.** Some server messages are not fully translated yet. Match on status codes. ## Known gaps in this preview Stated so you can plan rather than discover: - No self-serve key issuance until the preview opens. Keys come from the Keyline team until then. - The command-line tool ships with the Public Preview. Until then, drive the API directly over HTTP. - No published machine-readable API schema. - No webhooks. - Be a considerate caller and back off on errors rather than retrying tightly. - No region coordinates over a token. Crops yes, rectangles not yet. ## Getting access Sign up for the Public Preview from the [developer page](/keyline). Everyone on the list gets a key, the command-line tool, and the request and response samples the day it opens. --- ## How to Test AI Search on Your Technical Manuals Source: https://en.helpfeel.com/blog/test-ai-search-technical-manuals # How to Test AI Search on Your Technical Manuals Choose real manual questions, then judge the answer and its evidence before you expand an AI search test. A polished demo can hide a weak lookup. Test permitted documents, approved revisions, known answers, and questions the permitted set cannot answer. Keep the test narrow enough to judge. ## Start with one lookup task Pick one product family and one repeatable question type, such as a compatible part number or a table specification. Keep the test small enough for one product specialist to review. Write one sentence that defines the task: > Given an approved set of manuals for one product family, find the answer to a support question and show the exact page and region that supports it. The task does not ask AI to approve a repair, diagnose a hazard, or replace the accountable person. ## Freeze the documents and expected answers Create the test set before anyone runs either method. [NIST's AI Risk Management Framework Playbook](https://airc.nist.gov/airmf-resources/playbook/measure) says to document test sets, metrics, and tool details. It says that documentation enables repeatability and consistency. For each file, record: - file name and document identifier - product model or family - revision, issue date, and approval status - whether the file is allowed in the test - which file wins if two revisions conflict Have a product specialist write the expected answer and mark the page, cell, label, or note that proves it. Do this before seeing either method's answer. If source authority is unclear, stop and resolve it. Search cannot fix an approval problem. ## Record the current lookup method Use the same questions, documents, and qualified reviewer for the current method and the AI test. For each method, record three times: 1. Time to find a candidate answer. 2. Time for the reviewer to verify the answer against the source. 3. Total lookup and verification time. The reviewer-time score below uses verification time. Read total time alongside it. A fast lookup that takes longer to check may only move the work. Do not run the baseline first for every question. Before testing, match questions within each type by expected difficulty. Run the baseline first for one question in each pair and the AI method first for the other. Assign the order before anyone sees results. This reduces order bias, but it does not remove learning effects. If the set is too small, record the run order and label it as a limitation. Run both methods under similar conditions. Give neither method hints that normal users will not have. ## Test five kinds of question Split the set so you can see where each method works and where it breaks. ![A controlled AI search test checks five question types, then scores the answer and its evidence.](/blog/test-ai-search-technical-manuals.png) | Question type | What it tests | Example | | ------------- | --------------------------------------------- | ------------------------------------------------------- | | Prose | Direct retrieval from paragraphs | What condition does the manual state for this warranty? | | Diagram | Labels and visual relationships | Which callout identifies the named component? | | Table | Row, column, unit, and footnote relationships | Which listed item applies to this model and condition? | | Revision | Source authority across similar files | Which value appears in the approved revision? | | Unanswerable | Restraint when evidence is absent | Does the permitted set state this tolerance? | Use questions from real support work. Keep your team's wording and model numbers, but remove unneeded customer details. Unanswerable questions check whether the tool identifies missing evidence instead of guessing. ## Use this scorecard Copy this table into a spreadsheet. Preserve every raw answer, source reference, and timing record. | Field | What to record | | ---------------------------- | -------------------------------------------------------------------------------- | | Test ID | Stable ID for the question | | Question type | Prose, diagram, table, revision, or unanswerable | | Question | Exact wording used in both methods | | Expected result | Approved answer, or "not answerable from permitted files" | | Allowed source | File ID, model, approved revision, and exact evidence page | | Run order | Which method ran first for this question | | Baseline result and evidence | Full answer plus file, page, and source region used | | Baseline times | Lookup time, verification time, and total time | | Baseline scores | Correctness, source support, revision, and unanswerable behavior when applicable | | AI result and evidence | Full answer plus file, page, and cited region or excerpt | | AI times | Lookup time, verification time, and total time | | AI scores | Correctness, source support, revision, and unanswerable behavior when applicable | | Reviewer-time score | 0, 1, or 2 using the predeclared verification-time threshold | | Notes | Failure type, ambiguity, escalation, correction, or order caveat | Score each method separately for correctness, source support, revision, and unanswerable behavior. Use the reviewer-time row to compare AI verification time with the baseline: | Measure | 2 | 1 | 0 | | --------------------- | --------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | | Correctness | Matches the approved answer with every required model, part, value, unit, and condition | Useful but incomplete, with no material wrong detail | Material error, unsupported addition, or wrong conclusion | | Source support | Evidence directly supports the full answer and the cited location is usable | Right document, but the evidence is incomplete or the location is too broad | Missing evidence, wrong source, or evidence contradicts the answer | | Revision | Uses the approved revision and identifies it clearly | Revision appears right but is not identified clearly | Uses or may use a stale, wrong, or unapproved revision | | Reviewer time | Verification time is at least the predeclared threshold faster than the baseline | Verification time is faster than the baseline, but by less than the threshold | Verification time is equal to or slower than the baseline, or cannot be verified | | Unanswerable behavior | Says the permitted set does not support an answer and identifies the gap | Declines to answer but gives a vague or partly wrong reason | Invents an answer or misses evidence that is present | Do not hide a critical miss inside an average. Fail an answerable case if correctness, source support, or revision scores 0. Fail an unanswerable case if its unanswerable score is 0. Then compare pass rates by question type, reviewer verification time, and total time with the current method. ## Run the test without creating an order advantage 1. Lock the files, revisions, questions, answer key, rubric, matched pairs, and run order. 2. Run each question with its assigned method first, then run the other method. 3. Start a clean AI session for each question unless normal use depends on prior context. 4. Save the exact answer, source reference, visible evidence, and all three times. 5. Have the same reviewer score both methods without changing the rubric. 6. Label every miss by cause, such as retrieval, interpretation, source location, revision, or refusal. 7. Record order and learning effects as a limitation. Before repeating, record what changed. If you rewrite a question after seeing results, save it as a new test version. ## Worked example: check a parts-list relationship Graco publishes [manual 310662D for UltraMix and HydraMix displacement pumps](https://www.graco.com/content/dam/graco/tech_documents/manuals/310/310662/310662EN-D.pdf). This parts-list example uses the manual as its reference answer. It does not report a test of Keyline or another AI tool. **Question:** For UltraMix pump part number 248540, which numbered parts in Repair Kit 248438 are also included in Packing Kit 248437? **Expected answer:** In Graco manual 310662D, Revision D, June 2019, reference numbers 3 and 15. **Exact evidence:** Printed page 12 identifies pump 248540. An asterisk marks Repair Kit 248438 parts, and a dagger marks Packing Kit 248437 parts. Rows 3 and 15 carry both. Row 18 carries only the repair-kit mark. Printed page 16 identifies MM 310662 and states "Revision D, June 2019." Printed page 5 says that an asterisk marks repair-kit parts and points readers to pages 12 and 13. It does not identify Packing Kit 248437 or show which rows carry both markers. Under this proposed rubric, a hypothetical answer that gives references 3 and 15, identifies manual 310662D and Revision D, and cites printed pages 12 and 16 earns 2 for correctness, source support, and revision. This is a scoring illustration. It is not an observed AI result. For an unanswerable variant, permit only printed pages 5 and 16 of manual 310662D. Exclude every other page. Those pages identify the repair-kit marker and revision, but not the packing-kit marker or rows with both markers. The expected result is "not answerable from the permitted pages." This narrow case does not claim that the full manual becomes unanswerable when page 12 is removed. ## Set the buying decision before you see the result Before testing, name the minimum pass rate by question type, maximum critical failures, reviewer-time improvement, decision date, and owner. Set a separate rule for unanswerable cases. The useful outcome is a decision grounded in the same questions your team handles every week. If your team has one recurring manual lookup to test, read about [Keyline for technical teams](/keyline/teams) or [contact us to discuss a paid Keyline beta](/contact). Tell us the workflow, the team that owns it, and the document type. Do not send files. --- ## Loop Ownership 101: How to Run Your First AI Loop Source: https://en.helpfeel.com/blog/loop-ownership-101 # Loop Ownership 101: How to Run Your First AI Loop Your new AI agent went live. The launch team moved on. The first exception arrived. Who owns what happens next? When a production agent hits an exception, your org chart gets very real. That first exception is where AI projects become operations. A **loop owner** is the named person who owns the business outcome, boundaries, exceptions, measures, and next improvement decision. Technical teams still own models and infrastructure. The loop owner owns what the workflow does to the customer and the business. ## Launch day starts the real work A demo gets judged in minutes. A production agent gets judged every time the workflow changes around it. And it will change. Source data moves, policies change, tools return unexpected errors, and customers ask questions nobody included in the test set. Sometimes the agent completes its assigned step while the next team still can't use the result. The launch didn't fail. The agent is now working inside a living business. Tim Thijsse put the risk plainly on [CX Heroes](https://www.youtube.com/watch?v=Nl0nB7QQkFE): > "it might also be something in the operation where there's a handover just lacking or is leaking information, your data is not complete or incorrect." That is the handoff sitting inside the opening scene. The agent ran. The workflow still needs a person who can protect the customer, find the cause, and decide what changes next. ## Handle the first exception Start by stopping the normal path and keeping the evidence. The input, the output, the tools used, the handoff, and the failed completion check all belong with the case. Now protect the customer outcome. A fast support answer followed by a repeat contact has failed downstream, even if the agent's own task shows complete. Route the case to a named person with the context and authority to act. "Ask a human" isn't enough. Which human? With what context? By when? Then classify the cause. Model or tool behavior belongs in one group; broken steps, unclear policy, and bad source data belong in another. Human execution may belong in a third. Every failure needs a cause before it gets a fix. The green checkmark can wait. Once the team understands the cause, change one bounded thing. State what should improve, test it against historical cases, release it through the right gate, and compare the new evidence with the baseline. The loop owner decides whether to keep, revise, or reverse the change. ## Run one weekly loop review Bring the first exception into the weekly review as evidence. Then look for the other cases that share its cause. This keeps the meeting focused on removing a class of failure. One memorable example becomes evidence for a wider pattern. ![A loop owner at the center of five decisions: outcome, boundaries, exceptions, measures, and the next change.](/blog/loop-ownership-101.png) Review these five items in order: 1. **Outcome movement:** Did the customer and business measures improve, stay flat, or decline? 2. **Incorrect or incomplete work:** Which outputs failed the quality or completion test? 3. **Stops and human interventions:** Where did the agent stop, escalate, or require correction? 4. **Workflow changes:** Did knowledge, policy, tools, permissions, or source systems change? 5. **One experiment:** Which single approved change will the team test next? Leave with one owner, one change, one expected effect, and one verification date. A long list with no decision means the loop didn't close. ## Your first 30 days as a loop owner Use the first month to narrow uncertainty before you expand authority. ### Week 1: Map and baseline Map the workflow from trigger to verified outcome. Include the point where that first exception left the normal path, the evidence that traveled with it, and the person who received it. Then measure the current process. You need a baseline for quality, time, cost, customer outcome, and human effort before the agent changes them. "Use an AI agent" is an implementation choice. It isn't an outcome. ### Week 2: Shadow and classify Run the agent beside the current process without allowing customer-facing or live-system changes. Compare both the outputs and the paths used to produce them. Classify every disagreement. Don't hide a broken workflow inside a model-quality score. ### Week 3: Launch narrowly and verify readback Choose a low-risk case type with clear rules. Limit the agent's users, volume, tools, or permissions. For every write or external action, read the target back. A successful tool response isn't proof that the business record changed as intended. ### Week 4: Review outcomes and approve one expansion Compare the first production evidence with the baseline. Check customer outcomes, task quality, downstream work, exceptions, human review, cost, and speed. Then make one decision: keep the current boundary, narrow it, stop it, or approve one expansion. Evidence determines whether the workflow earns more authority. ## Related reading - [AI agent operations](/ai-agent-operations): define the full operating model for owned AI workflows. - [Customer support](/customer-support): connect repeatable work, human judgment, and self-service. - [The Helpfeel platform](/platform): see how Helpfeel combines AI agents, knowledge, and continuous improvement. ## Loop ownership changes the human role As AI agents take on more execution, service professionals gain broader responsibility for judgment, customer relationships, system quality, and improvement. That is the human side of Agent Growth Experience. A person who used to process each request can now see patterns across requests, fix the system behind them, and protect the customer when the normal path breaks. The loop owner needs the authority to define what good looks like, stop the workflow, and approve the next change. **Operating loops pair agent execution with human judgment: each run produces evidence that the loop owner uses to improve the next cycle.** --- ## The Five Levels of Service Elevation Source: https://en.helpfeel.com/blog/five-levels-service-elevation # The Five Levels of Service Elevation _Helpfeel note: This guide adapts a Service Elevation draft from [Nate Brown](https://experiencenate.com). The level names, descriptions, AGX framing, and product links are Helpfeel's additions._ A queue can be full and every dashboard green while the customer still carries the hardest part. Service Elevation starts with that gap. Nate describes customer experience as making people's lives better and easier. Customer expectations have risen alongside what service leaders can now design. AI agents can answer, route, act, and learn across workflows. Human service agents can spend more time on judgment, relationships, and improvement. The opportunity grows when those experiences are designed together. Helpfeel calls that combined design approach **Agent Growth Experience, or AGX**. AGX treats the human service agent experience and the AI agent experience as one system. AI agents receive the strongest focus because their growing capability changes service work. That gives human agents more room to learn, guide, and improve the system. Together, they can help customers and the business grow. Cost savings matter. Customer and business growth matter more. The five levels make the next transition visible. ## How to use this maturity model This maturity model helps you identify the dominant shape of service today and choose the next useful transition. It is a planning guide, not a grade. Each level describes observable behavior that you can use for planning. One company can operate at different levels across journeys, products, regions, or channels. A routine account update may sit at a higher level than a sensitive claim or complex repair. The right design matches capability, customer need, and risk. Look for evidence in how people spend their time, what agents complete, where knowledge lives, how decisions get made, and which customer signals reach the rest of the business. Level 0 is the short pre-maturity baseline. Manuals and documents exist digitally. Their knowledge still needs structure before service teams or AI agents can use it reliably. Service Elevation covers the whole service organization. Helpfeel uses the Autopilot Call Center model as a related technical view of AI-ready knowledge, workflows, system actions, supervision, and continuous improvement. Explore that [technical capability view](/platform) separately. Helpfeel's approaches to [customer support](/customer-support) and [knowledge base management](/knowledge-base-management) provide more context for the model's operating foundation. ## The five levels at a glance ![Five levels of Service Elevation from reactive service to unified experience](/blog/five-levels-service-elevation.png) | Level | Public name | What you can observe | | ----- | ------------------------------ | --------------------------------------------------------------------------------------------------------------------- | | 1 | Level 1: Reactive service | The team responds to demand and begins building trusted knowledge for common needs. | | 2 | Level 2: Assisted self-service | AI completes established routine work and assists people during the interactions that remain. | | 3 | Level 3: Consultative service | AI follows defined procedures and takes approved actions while people handle complex needs and create customer value. | | 4 | Level 4: Proactive guidance | Routine service runs end to end where appropriate, and people focus on exceptions, improvement, and guidance. | | 5 | Level 5: Unified experience | Shared customer knowledge drives continuous improvement across the business. | ## Level 1: Reactive service At Level 1, the organization still works mainly in response to customer demand. People monitor queues, follow scripts, move information between systems, and escalate work to a small group of experts. The leadership focus is operational stability. Volume, speed, adherence, handle time, and compliance dominate the dashboard. Those measures can keep a service operation healthy. They give a limited view of whether the customer reached a useful outcome or why the demand appeared. Knowledge often comes from subject-matter experts and moves slowly. Common answers may live in a basic knowledge base, while procedural details remain scattered across manuals, folders, and people's memories. AI search or simple self-service can resolve familiar questions. Human agents still carry most of the context and execution. Level 1 creates a dependable foundation. Repeated questions become visible, and leaders can turn service records into knowledge that people and agents can trust. **Observable evidence** - Most customer needs become live contacts. - The same questions return across the queue. - A few experts receive frequent interruptions for complex work. - People copy information between systems to complete routine tasks. - Leaders can describe workload more easily than customer or business impact. **One practical next move:** Choose one common, well-understood journey. Expand its trusted knowledge from simple answers into the actual procedure, then measure whether customers reach a successful resolution. This creates a clean path into Level 2 without asking the whole operation to change at once. ## Level 2: Assisted self-service At Level 2, human service agents remain the primary operators, and AI takes on a meaningful share of routine work. Customers can resolve straightforward needs through self-service. During live interactions, AI can transcribe conversations, suggest reliable answers, surface procedural knowledge, and reduce after-call administration. Agents spend less time searching and entering repetitive data. They supervise assisted work and step in when knowledge, policy, or system access reaches a boundary. Leaders can reallocate capacity toward work that needs more skill, care, or judgment. Self-service quality becomes crucial here. The customer outcome tells you whether the channel works. Strong teams track successful resolution, first-contact resolution, customer satisfaction, sentiment, and quality. Knowledge now reflects how customers actually ask for help. Agents start contributing what they learn, though ownership usually stays within the service department. Separate tools and data boundaries still create friction. Complex requests may require manual assembly, and cross-system needs can reach a dead end. **Observable evidence** - Routine needs often resolve without a live interaction. - Agents consistently use transcription, answer suggestions, or workflow assistance. - After-call work is smaller and more standardized. - Complex needs still require people to gather context from several systems. - Most knowledge improvement remains inside the service team. **One practical next move:** Connect one trusted procedure to the system where action happens, such as a CRM. Give the AI agent a bounded action it can take with clear approval rules. Train human agents to review exceptions, improve knowledge, and identify root causes. Then bring one recurring customer learning into a regular cross-functional meeting. ## Level 3: Consultative service Level 3 changes the role of service. AI performs beyond novice level on defined work. It can follow conditional procedures, route decisions, and take approved actions through connected systems. Human agents concentrate on complex resolution, sensitive decisions, customer relationships, and improvement. The ticket becomes a source of insight. A strong agent looks beyond the immediate answer and asks what would prevent the next issue, what outcome the customer is trying to reach, and what the wider organization should learn. Service begins to influence product, policy, operations, and customer success. New roles and skills emerge. Service engineers improve workflows and knowledge. Other people focus on consultation, community, quality, or customer learning. Their judgment makes the whole system more capable. Knowledge crosses department and system boundaries. People and AI agents can identify gaps during real work and feed improvements back into the source. Customer learning starts producing documented changes outside the service team. Measures expand toward customer effort, sentiment, quality across human and automated work, renewal, retention, and the outcomes of improvement projects. Leaders should separate useful correlation from proven business causation. **Observable evidence** - AI agents complete defined conditional work and take approved actions in connected systems. - Human agents can find reliable answers for most needs without interrupting an expert. - Exception handling has clear owners and decision rights. - Customer learning regularly causes a product, policy, process, or knowledge change. - Service people spend visible time improving the system and guiding customers. **One practical next move:** Select one routine workflow that is ready to run end to end. Define its stop conditions, exception paths, human approval gates, and readback. Add observability across self-service, assisted service, and human service so the team can see what happened and improve the workflow safely. ## Level 4: Proactive guidance At Level 4, routine service runs end to end where appropriate. AI agents follow approved decision paths, connect information across systems, and take bounded action. Human agents set policy, supervise risk, resolve exceptions, improve the system, and guide customers before a familiar hurdle becomes a problem. The organization becomes smarter with every interaction. Customer learning can reach other teams through shared data flows and operating routines. Leaders measure resolution across service tiers, customer health, future behavior after service, and the effect of improvement work. **Observable evidence** - Routine workflows complete across systems with clear controls. - People focus on exceptions, relationships, proactive engagement, and system quality. - Customer signals move to other teams without waiting for a periodic presentation. **One practical next move:** Give product, service, success, marketing, sales, and leadership access to shared customer context. Assign cross-functional ownership to one experience improvement and measure its before-and-after result. ## Level 5: Unified experience Level 5 is an intentionally ambitious destination. AI agents autonomously run the service work selected for automation, including bounded discretionary decisions and exceptions. People define goals, judgment, ethics, relationships, and the next improvement opportunity. Fresh customer knowledge drives continuous improvement across the business. The lines between support, service, success, marketing, and sales become less important because teams share context, measures, and responsibility for customer outcomes. Knowledge is the core asset, and high-value human input makes every agent more capable. **Observable evidence** - Teams share fresh customer context across traditional department boundaries. - Voice of Customer learning drives controlled, measured improvement across the business. - Human service professionals gain skill, scope, judgment, and influence. **One practical next move:** Choose one customer outcome that crosses several departments. Create shared ownership, connect the relevant knowledge and signals, and define where people retain judgment. Automate the learning cycle only where evidence and risk allow. ## Elevation starts with the next transition Level 5 will look different for every company and customer. Its value lies in giving service leaders a direction to explore while the earlier levels provide practical work to do now. Start with one journey. Find the evidence for its current level. Choose one transition. Improve the knowledge, workflow, and decision rights together. Then help your human and AI agents grow through the work. Helpfeel's AGX perspective adds one working principle: **Service elevation happens when every agent becomes more capable, and that shared capability creates new value for the customer.** If you want a partner for your next transition, [start a conversation with Helpfeel](/get-started). --- ## CX Heroes Podcast #52: Why a Missed Delivery Date Costs You More Than the Order Source: https://en.helpfeel.com/blog/cx-heroes-melanie-edwards-virreira-delivery-date-trust # CX Heroes Podcast #52: Why a Missed Delivery Date Costs You More Than the Order In B2B, what you're really selling is the experience around the product. Melanie Edwards Virreira, who has spent her career in packaging, says the product is only part of what a customer pays for. Miss a date you promised, and that is the fastest way to break the deal. Melanie has led brand and go-to-market strategy at Resource Label Group and at Ball Corporation, with earlier international roles at MCC Label. She joined the CX Heroes podcast to talk about brand, trust, and why manufacturing has been slow to invest in customer experience. Her sharpest point was about delivery dates. When a customer orders bottles or labels for a bottling run, the date is not flexible. As she put it, price can be what it is and quality can sit inside a range, but you cannot miss the delivery date. If the shipment is late and the buyer misses their bottling window, the buyer is the one who gets fired. That changes what a good supplier owes the customer. The product matters, but the buyer also needs to know where the order is at every step. What buyers really want, in Melanie's words, is reassurance at every step of the process, so they can sleep at night. A supplier who provides that becomes hard to replace. This is a customer support and knowledge problem as much as a logistics one. Every status check is a question. When a customer has to call or email to find out where their order stands, that is a signal you could have answered the question before they asked it. Melanie also had a practical reframe for supplier meetings. Instead of walking in with a slide of everything you can make, ask the customer for a challenge statement. What is going on with your brand? What problem are you trying to solve? She pointed to her time around the beverage industry, where the strongest supplier programs started with a customer problem and built the solution backward from there. The last piece was about data. For years, manufacturers treated connected systems as optional. CRM and ERP that do not talk to each other, customer names that do not match after an acquisition. None of it was required to keep the lights on, so it kept getting pushed off. Melanie's view is that AI is finally making the cleanup fast enough to be worth doing, which lets the industry skip stages it avoided for years. Melanie's advice comes down to this: figure out which customers actually value what you do, then solve the one problem they care about most before anything else. In an industry where the bar for customer experience is still low, doing that well is one of the clearest ways to stand out. You can watch the full conversation with Melanie Edwards Virreira on the CX Heroes podcast. [Watch: CX Heroes: Melanie Edwards Virreira on why a missed delivery date costs you more than the order](https://youtu.be/oTN7Nq5cyng) [Watch the full episode of CX Heroes with Melanie Edwards Virreira →](https://youtu.be/oTN7Nq5cyng) --- ## Context Boundaries: When to Start Fresh With AI Source: https://en.helpfeel.com/blog/context-boundaries # Context Boundaries: When to Start Fresh With AI The single most useful thing I have learned working with AI every day is not a prompt. It is knowing when to stop the current chat and start a clean one. Most people treat a session as one long thread. They pour everything into the same window and keep going until the answers get worse, then blame the model. The real problem is usually a boundary they never drew. ## What a context boundary is A context boundary is the line where you decide the next task gets a fresh start. A new session. A different agent. A different set of inputs. A prompt written from scratch instead of piled on top of the last ten. You draw one every time you sense that what you have already said is pulling the answer in the wrong direction. It is a design decision, not an accident, and once you start making it on purpose the quality of your work with AI jumps. ## Why the same chat quietly fails An AI predicts the next token from everything already in front of it. That is the whole trick. It reads the window and produces the most likely continuation of what is there. That is a strength when you want continuity. It is a trap when you want a break. A long session builds momentum. Twenty messages of you praising a draft, refining a draft, defending a draft, and the model has learned one thing above all: this draft is good, keep going in this direction. Ask it to now tear that draft apart and it will predict the polite continuation of a conversation that has been agreeing with itself the whole time. The context is not neutral. It has a gravity. And the more you put in it, the harder it pulls toward more of the same. ## The writer and the editor Here is the case where the boundary matters most. A writer and an editor do different jobs. The writer commits. Their job is to produce, to believe in the thing, to push it out the door. The editor doubts. Their job is to find what is broken and say so without mercy. Now put both jobs in one AI session. You ask it to write, it writes. You ask it to edit, and it edits as the same voice that just wrote the piece, gently, protectively, invested in the choices it already made. You get a proofread, not an edit. The critical task loses to the momentum of the creative one. Break the context and the whole dynamic changes. Open a new session. Give it only the draft, no history of how the draft was made, no record of you loving it. Tell it: you are a skeptical editor, this piece is not working, find out why. With none of the writing session's gravity to fight, it reads the words on the page and says the hard thing. A great human editor does exactly this. They did not sit next to you while you wrote. They come to the page cold, and the distance is the point. ## Where the boundaries go Once you see it, you find the line in a lot of places. Four practical ones. **New session.** When the next task has to contradict the current one. Critique after creation. A second opinion that must not be charmed by the first. Anytime you want the model to react to a result rather than to the story of how you got there, cut the thread and hand over only the result. **New agent.** When the mindset itself should be different. A researcher who gathers wide and a decision-maker who cuts hard are not the same person and should not share a head. Give each one its own instructions, its own role, its own window. Real teams separate these jobs for a reason. Copy the reason. **New inputs.** When the old context is quietly biasing the read. If you want a clean judgment on a piece of data, do not feed the model your running commentary about what you hope it shows. Give it the data and the question, nothing else. Your framing is a thumb on the scale. **New prompt.** When you are patching a conversation that went sideways three turns ago. Stop adding corrections on top of corrections. The window is now full of the wrong direction and your fixes are fighting all of it. Start clean, say what you actually want in one clear prompt, and let it predict from a blank page instead of a mess. ## The heuristic Continuity is a feature when each step builds on the last. Keep the thread when you are refining, extending, or going deeper on the same task in the same mode. Break it the moment the next task needs to escape the last one. New mode, new job, contradicting judgment, or a read you want kept honest. If you are about to ask the model to argue against something it just argued for, do not. Give that job a clean window. The tell is simple. When you catch yourself typing "no, actually, forget what I said," that is not a correction. That is a boundary you should have drawn one message ago. ## Draw the line on purpose The people who get the most out of AI are not the ones with the longest chats. They are the ones who know when to end one. A boundary is not lost work. It is the thing that lets the next task be honest instead of agreeable. Set it before the momentum sets it for you. If you want to see what drawing these lines on purpose looks like inside customer support, [see how Helpfeel works](/platform). --- ## What an AI Customer Support Deployment Actually Looks Like Source: https://en.helpfeel.com/blog/inside-an-ai-support-deployment # What an AI customer support deployment actually looks like Most AI support pitches end at launch day. The real work is what happens next. One retailer's self-resolvable inquiries fell 78% within weeks of go-live, so fast the team assumed their own tracking had broken. What actually happens in the 90 days after an AI support tool goes live? Here's the actual 90-day arc behind four real deployments, told in the order it happened, not the order it got pitched. You've watched a knowledge base project launch loud and go quiet. Here's why some don't. ## Day one: the data starts talking The moment search goes live on your existing help center, every question typed in becomes data: what people searched, what they clicked, where they gave up. It's the first real signal anyone's had. More often than not, it surfaces something the team wasn't expecting. Alpen, one of Japan's largest sporting goods retailers, spent a year evaluating chatbots before it changed course. Once real search data started coming in, self-resolvable inquiries fell 78%. The number was so steep, the internal team assumed something had broken. Same customers. Same questions. Same articles, sitting on the same help center. The only thing that changed was whether search could actually find the answer already there. That's the pattern, over and over. The articles were usually fine. What was missing was the match between how customers ask and how the answers were written. ## Week one to two: the easiest win is the one nobody expected Most teams expect the first big win to come from the help center. It rarely does. The bigger win sits wherever a customer is already about to write in. One mail-order catalog retailer, fielding over ten thousand inquiries a month during a peak surge, embedded live search directly into its contact form. As a customer typed their subject line, relevant articles surfaced in real time, before they ever hit send. The team's own name for it: the "chasing FAQ." Inquiries fell 50%. Help center sessions doubled. Nobody changed their behavior. Customers still opened the contact form and started typing. They just got the answer before they hit send. That's the shape of the first quick win, almost every time. Find the moment right before a customer commits to writing in. Put the answer there instead. ## Month one: the expert bottleneck becomes visible When knowledge lives in one person's head, the first month of usage data makes that impossible to ignore. Fujimak Corporation runs 218 service engineers across roughly 70 locations, all needing answers about commercial kitchen equipment out in the field. For years, a hard question meant one phone call, to one product expert. If he was in a meeting, or off for the day, the field waited. > "The confidence that comes from being well-prepared beforehand has become a major source of support." > > Kenya Karita, Service Engineer, Fujimak Corporation Search usage climbed from roughly 200 accesses a month at launch to over 1,000 within the year, a five-fold increase. And that one expert's own inquiry handling load dropped about 80%. His workload didn't shrink because he offloaded it. The same answer, written down once and findable, stopped needing to be repeated one phone call at a time. ## Month two: the monthly review becomes the actual engine Every one of these deployments shares one structure that isn't optional: a recurring review of what the usage data shows, followed by real edits before the next cycle starts. This is where the technology stops being the story. Fujimak's team put it plainly when asked what would have happened without that managed review cycle. Without it, "the system would have ended up like previous failed implementations, gradually going unused." The search engine itself is maybe a fifth of what makes this work. The other four-fifths is the discipline: look at what people actually searched for last month, then fix the gaps before they compound. That's the core mechanic behind what we call [The Loop](/customer-support). Agents handle the volume. Humans read what the agents couldn't answer. Someone updates the knowledge. And next month's data comes in a little cleaner than the last. ## Month three: proving it holds up under pressure The real test of any support setup isn't a quiet Tuesday. It's the traffic spike. MUJI runs support across more than 1,300 stores and an EC business selling everything from a fifty-cent eraser to furniture. Its old FAQ belonged to a separate team, took up to two weeks to update, and missed seasonal moments outright. Sunscreen questions were still going live in early spring, well after summer had already started. After rebuilding around real-time publishing, the team could ship a seasonal update the same day, not two weeks later. During its highest-volume campaign period, historically a stretch that spiked inquiries hard, sales grew while inquiries stayed flat. > "Our perception of what an FAQ can be has fundamentally changed. It used to feel like a supporting role in customer service. Now we see it as equally important as phone and email, and as something worth continuously improving." > > Mami Masumoto, CX Promotion Division, MUJI ## Why the same 90 days keeps producing the same shape of result None of these companies look alike. A kitchen equipment manufacturer. A sporting goods retailer. A nationwide EC brand. A catalog retailer surviving its own seasonal surge. What they share is the sequence. Launch with real data flowing. Catch the easiest quick win in the first two weeks. Surface the hidden bottleneck in month one. Lock in a monthly cycle that makes the knowledge underneath a little more accurate than it was the month before. [Read how AI agents fit into a support methodology built to compound](/customer-support). The payoff shows up after launch day, in the monthly work that keeps compounding. See [how Helpfeel's done-for-you model works](/platform). --- ## Your CX Problem Is Probably a Knowledge Problem Source: https://en.helpfeel.com/blog/cx-problem-knowledge-problem # Your CX Problem Is Probably a Knowledge Problem The launch went fine. New chatbot, refreshed help center, an AI pilot the executive team is excited about. Six months later, ticket volume is flat, CSAT is flat, and everyone is quietly wondering what went wrong. Here's what's actually happening. Every tool in that stack does the same job. It takes a question and returns an answer. The chatbot, the help center search bar, the agent on the phone: different doors, same kitchen. If the kitchen can't find its recipes, a nicer door doesn't change what comes out of it. That's the contradiction sitting inside most customer experience programs: the ambition and the budget both point at the experience layer, while the knowledge every layer depends on stays exactly as messy as it was the day the project kicked off. ## The fallacy: experience lives in the interface Ask a customer to describe a great support experience and they'll describe an answer. Right, fast, first try. They won't mention the widget. The interface barely registers when it works, which is exactly what makes it so tempting to buy. An interface is visible. You can demo it to the board, screenshot it in the all-hands deck, put a launch date on it. Knowledge work is the opposite. Four hundred articles reviewed for accuracy doesn't make a slide anyone claps for. So the visible layer gets funded and the invisible layer gets inherited. Whatever state the knowledge was in before the project, that's the state the new tools now serve at higher speed. ## Where the mess actually comes from Nobody set out to build a messy knowledge base. That matters, because the usual story teams tell themselves is some version of "we got lazy," and it's wrong. The articles got written during a launch push or an onboarding sprint, by whoever had time that month, describing the product as it existed that quarter. Then the product changed. Pricing changed. A policy changed. The person who wrote the articles changed roles. Every one of those changes made some article somewhere quietly wrong, and no alarm went off. One customer success director at a national office-equipment company put it plainly: "We are running extremely lean. There's nobody dedicated to even keeping the knowledge base updated." That's not an unusual confession. In my experience it's the default state of the industry. Meanwhile, the real knowledge kept accumulating where knowledge naturally accumulates: closed tickets, Slack threads, and the heads of your two most senior agents. Martin Hobratschk, who spent years running knowledge management at Apple, described the result on our podcast: "Having lots of silos is like having a giant digital landfill, nobody can find anything." The effort was never the problem. The system was. Maintenance is invisible work in an organization that celebrates launches, and [knowledge goes out of date](/knowledge-base-management/why-knowledge-bases-go-out-of-date) for structural reasons, whether or not anyone is slacking. ## Why the gap is widening now For years this arrangement was survivable. A stale article cost a customer five minutes and a sigh, then they emailed support and a human absorbed the miss. AI changed the price of the miss. An AI tool built on top of a messy knowledge base doesn't hesitate the way a human does. It answers with full confidence, and a confidently wrong answer costs more trust than a slow human ever did. Customers forgive "let me check on that." They remember being misled. Each new channel makes this heavier, and that's the part that stings. Every door you add, chat, search, email, voice, is another route to the same shelf of answers. Adding doors multiplies whatever is on the shelf. If the shelf is current, you've multiplied something great. If it's a landfill, you've built five new entrances to it. ## The question this forces Before the visible layer can get better, someone in your organization has to be able to answer one question: which of our answers are true today? Sit with how hard that is. Most teams can tell you their ticket volume to the decimal and can't tell you what fraction of their published answers still match the product. The measurement everyone watches sits downstream of the thing nobody measures. That's the real starting line for a customer experience program. It's less glamorous than a chatbot launch, and it decides whether the chatbot launch means anything. This exact problem is the reason Helpfeel exists. If you want to see how we think about it, [start here](/platform). --- ## Executives Think Customers Trust Them. They Don't. Source: https://en.helpfeel.com/blog/customer-trust-perception-gap # Executives Think Customers Trust Them. They Don't. Ninety percent of business executives believe customers highly trust their company. Only 30% of customers actually do. That 60-point gap, from PwC's Trust in US Business Survey, has been widening every year it's been measured. This isn't a one-off finding. Every major study that asks both sides the same question finds the same pattern: companies believe they're closer to their customers than they are. ## The gap, by the numbers | Study | What it measured | The gap | | ----------------------------------- | --------------------------------------------------------------------------------------- | --------------------------------------------------------------- | | PwC Trust in US Business Survey | Executives who think customers highly trust their company vs. customers who say they do | 90% vs. 30% (a 60-point gap, up from 57 in the two prior years) | | PwC 2025 Customer Experience Survey | Executives who say customer loyalty has grown vs. consumers who agree | ~90% vs. 40% | | Forrester Brand Experience Index | Customer vs. noncustomer perception of the same brand | 5 to 30 points, depending on industry and region | Forrester's version of the gap works a little differently; it compares how a brand's existing customers see it against how people who've never bought from it see the same brand. But the underlying story matches PwC's: the people inside a company, and the people who already like the company, both see it more favorably than the people who don't yet have a reason to. ## Why the gap keeps growing None of this is executives lying to themselves on purpose. It's a measurement problem. PwC's researchers found that most companies that say they track trust are actually tracking something adjacent to it, like CSAT or NPS. Those numbers are related to trust. They aren't the same thing. [CSAT](/glossary/csat) tells you if someone was happy with one interaction. [NPS](/glossary/nps) tells you if someone would recommend you to a friend. Neither one asks the question that actually predicts whether a customer sticks around when something goes wrong: do they believe you'll come through for them next time? Trust is what's left when you strip out the interaction that just happened and ask about the relationship. Most companies don't have a number for that, so they default to the number they do have and assume it's close enough. ## Where the gap actually gets built The PwC 2025 Customer Experience Survey found that 29% of consumers stopped buying from a brand specifically because of a bad customer experience, separate from a bad product. That's not one dramatic failure. It's an accumulation of small moments where the company didn't show up the way it said it would. A support search that comes back empty. A chatbot that gives a confident wrong answer. A [customer effort score](/glossary/customer-effort-score) that keeps climbing because every issue takes three tries to resolve. Each one is small. Customers remember the pattern, not the individual event. That pattern shows up again in how consumers feel about AI specifically. The same PwC survey found that 58% of consumers are only somewhat or not at all comfortable using AI tools to engage with brands. People aren't rejecting AI on principle. They're reacting to the version of AI that guesses instead of knowing, the one built on a knowledge base nobody's kept current. Get an AI-driven answer wrong once, and it costs more trust than a slow human ever would. ## Closing the gap starts with measuring the right thing The fix isn't a bigger CSAT number. It's building a system where the answer a customer gets is the one you'd actually stand behind, every time, whether a human gives it or an [AI agent](/customer-support/ai-customer-support-agent) does. That means treating your knowledge base as the thing customers actually judge you on, not a back-office reference doc. It means the AI answering a question at 9pm is drawing from the same verified source your best support rep would use, not improvising. Teams further along this path, using frameworks like the [5 levels of AI maturity in customer service](/customer-support/ai-maturity-customer-service), already know the jump from "answers tickets" to "customers trust the answer" isn't a tooling upgrade. It's a decision to make the knowledge underneath everything accurate first. The companies closing this gap aren't the ones with the most confident brand messaging. They're the ones whose customers, when asked directly, say the company actually does what it claims. That's a much harder number to fake, and a much better one to chase. If you want to see what a verified, consistently accurate knowledge base looks like in practice, [see how Helpfeel works](/platform). --- ## Helpfeel Launches AI-Ready Consulting for Call Centers Source: https://en.helpfeel.com/blog/helpfeel-ai-ready-consulting-autopilot-call-center # Helpfeel Launches AI-Ready Consulting for Call Centers Japan's call centers answered fewer calls every year for the last three years, and hiring hasn't gotten any easier. Helpfeel just opened a consulting practice to fix the reason why, and the first customer is already live. ## The call center is running out of road Japan's call center market is worth more than one trillion yen, and it's under real strain. In a recent industry survey, about 80 percent of companies said hiring operators is difficult at some or all of their sites. That shortage shows up directly in service: answer rates have fallen for three straight years, and "I can't get through" is becoming a common complaint. At the same time, customer expectations are climbing. People want help on their own schedule, not the call center's, and they still want it to feel personal. Round-the-clock, high-quality support used to be a staffing problem. Now it's a staffing problem that fewer people are available to solve. ## AI adoption is up. Results haven't caught up. Companies have turned to AI to close that gap, and adoption in Japan now roughly matches other countries. But results haven't followed. A recent PwC Japan survey found that Japanese companies were four times less likely than their US and European counterparts to say their AI results beat expectations. Helpfeel's read on that gap: most companies are missing the knowledge layer that AI actually depends on. Manuals, FAQs, and call logs are only part of the picture. The rest lives as tacit know-how in experienced operators' heads and in workflows nobody wrote down. Until that knowledge gets organized into something AI can reference and reason over, adding an AI tool on top doesn't change much. ## What Helpfeel means by "Autopilot Call Center" Helpfeel is calling its answer the Autopilot Call Center: a call center where AI carries a request from the moment it comes in through to resolution, not just a chatbot bolted onto the front end. **Traditional call center:** an operator listens to the customer, then manually checks FAQs, manuals, and internal systems to answer or process the request. **Autopilot call center:** AI runs the whole path from intake to resolution, including system lookups and judgment calls on complex requests, not just scripted responses. Helpfeel frames this the same way the industry talks about self-driving cars: autonomy arrives in stages, not all at once. It maps the path in six levels, from no usable AI knowledge base at all up to full autonomous operation. | Level | Focus | What it looks like | | ------------------------ | ---------------------------------------- | ----------------------------------------------------------------------------------------------------- | | Level 0 (0% automated) | No usable knowledge base | AI tools exist but the underlying data isn't structured for them, so search and reuse are weak | | Level 1 | A knowledge base built for people and AI | Customers self-serve more, call volume drops, and the knowledge base keeps growing | | Level 2 | AI that supports the operator | AI assists with answers, making operators faster | | Level 3 | AI operators | AI handles specific inquiry types and procedures on its own | | Level 4 | Full automation of routine work | Every routine task runs on AI, freeing people for exceptions | | Level 5 (100% automated) | Full autonomy | An AI agent handles judgment calls and exceptions too, and people manage the AI rather than the queue | ## Proof in production: Tohoku Electric Power Tohoku Electric Power is the first company running this playbook with Helpfeel, and the early numbers back it up. The utility launched its self-service AI, built on Helpfeel, in March 2025. Ten months in, it was drawing 160,000 page views a month, call volume had dropped by about 20 percent, and the center's answer rate had improved. Tohoku Electric didn't stop at cutting call volume. It's now redesigning its support operation around AI from the ground up rather than swapping AI into the old process. The company has completed Level 1, building the shared knowledge base, and is now moving into Level 2, using AI to support its operators directly. Helpfeel is the consulting partner behind that transition, structuring the utility's complex workflows into data AI can actually use. ## The new consulting business Off the back of that engagement, Helpfeel is formally launching an AI-Ready consulting practice, starting in customer support and built to extend across a company's operations, systems, and knowledge base. It runs in two steps. **Step 1: AI-Ready strategy.** Helpfeel starts from the business outcome, not the tool. It works with a company to define what AI should actually change (revenue, cost, or productivity) and builds a roadmap around that P&L impact. **Step 2: execution support.** Helpfeel then runs the buildout, drawing on lessons from more than 900 site deployments as of April 2026. Every Helpfeel customer already gets a dedicated customer success partner, and that same team applies what it has learned to structuring knowledge, redesigning workflows, and adjusting how the organization operates around AI. The practice isn't limited to call centers. Helpfeel plans to apply the same approach to internal knowledge, sales enablement, marketing, and back-office work, wherever a company's know-how needs to become something AI can use. ## The CEO's take "Building the foundation for AI-autonomous customer support, that's the real substance of what we mean by the Autopilot Call Center," said Isshu Rakusai, CEO of Helpfeel. "We're drawing on more than 900 site deployments to lead the knowledge-building work behind every company's AI-Ready transformation. It's a lot like the path to self-driving cars: autonomy didn't arrive overnight there either, and it won't here. The direction is clear. Routine work goes to AI, and people move toward the creative, high-value work only they can do." ## Why this matters if you run support The gap between "we bought AI tools" and "AI actually improved our numbers" almost always comes down to knowledge, not the model. If your team's know-how is still scattered across FAQs, tribal knowledge, and undocumented workflows, adding AI on top just automates the guesswork. If you're mapping out where your own support operation sits on that same path, from Level 0 to full autonomy, [book time with our team](/get-started) to talk through it, or see [how Helpfeel works](/platform) today. --- ## CX Heroes Podcast #53: AI Can Write the First Draft. The Revision Is Still Your Job. Source: https://en.helpfeel.com/blog/cx-heroes-elise-jeanmaire-ai-needs-an-editor # CX Heroes Podcast #53: AI Can Write the First Draft. The Revision Is Still Your Job. Elise Jeanmaire has spent eighteen years building and scaling support teams at Sonos, Reebok, Nestle, SeatGeek, and Dutchie. Today she runs customer experience, loyalty technology, and retention at SimpliSafe, and she holds a COPC certification. She also writes fiction. That last detail shaped the most useful part of our conversation on CX Heroes. We talked about AI in the support workflow. Most of that conversation goes straight to speed and ticket volume. Elise took it somewhere more interesting. She talked about voice. ## The problem she sees every day AI can produce a knowledge base article, a customer reply, or an executive summary in about a second. Many teams take that output and ship it as is. Elise can tell which content was handled that way. It reads flat and generic, and sometimes the facts are wrong. Here is how she framed the risk: > "If we have AI writing all of our stories for our business, then it's no longer what we believe. It's kind of like what this tool believes." That is the quiet cost of shipping the first draft. Your help center, your replies, and your knowledge base carry your company's voice. Hand all of it to a model and you slowly hand over the voice too. ## Her fix comes from fiction Elise treats AI the way a novelist treats a first draft. The first draft is always a mess, whether a person wrote it or a model did. The work happens in the revision. > "The hard work always as a writer is the revisions, the continual revisions. As long as you trust AI to provide you something that you can start to sculpt down, great." So she lets AI write the messy first version, then edits it down to what she actually means. The model gives her raw material to work with, and she supplies the judgment. ## Revision is how you learn your business This is the insight worth keeping. Elise does not treat editing as a chore that slows her down. She treats it as the moment she learns the most. > "I find that I'm learning so much about my business by rereading and by revising, by just going through that work." When she rereads an AI draft of an executive summary or a test-and-learn report, she catches the places where it drifts off theme, surfaces something she would not say, or gets a detail wrong. Fixing those is how she stays close to her customers and her data. Skip the revision and you skip the learning. ## Why this matches how we think about AI support At Helpfeel, we build AI customer support with a managed service around it, and Elise's point is close to why. Good AI support is only as good as the content behind it and the people keeping that content honest. Helpfeel answers up to 70 percent of incoming questions and returns a relevant answer to 98 percent of searches, across more than 800 sites. Those numbers hold up because a human editor tunes the answers against what customers actually ask, using the voice of the customer. The AI drafts and answers at scale. People keep it accurate and keep it sounding like you. That is the same balance Elise describes, built into the product. Her episode is a good reminder for any team leaning on AI to write support content this year. Let it draft. Then revise it, and keep the voice that makes the business yours. [Watch: CX Heroes: Elise Jeanmaire on why AI needs an editor](https://youtu.be/B0GfIfzwpGU) [Watch the full episode of CX Heroes with Elise Jeanmaire →](https://www.youtube.com/watch?v=B0GfIfzwpGU) --- ## Inside HACK Lab: Helpfeel's Silicon Valley Rotation Source: https://en.helpfeel.com/blog/hack-lab-silicon-valley-outpost # Inside HACK Lab: Helpfeel's Silicon Valley Rotation In July 2026, Helpfeel kicked off HACK Lab, a rotation that brings our engineers and PMs to Silicon Valley for short, hands-on stretches. Our CEO, Isshu Rakusai, and I sat down to talk about why, and what we're actually doing there. ## Nobody asks what you do here. They ask what you're building Most places, meeting someone new starts with small talk and a business card. Not here. The standard opener is "Hi, what are you building?" or "Are you working on anything interesting right now?" You'll hear it at a bar, a coffee shop, an art opening. It doesn't matter where. Almost everyone in the room works in tech, so the conversation drifts toward the latest models and tools without anyone forcing it. Rakusai has visited Silicon Valley many times, and this trip he described it as a whole city functioning as one engineering lab. The billboards along the highway back him up: nearly all of them are ads for developer tools like Vercel and Intercom, aimed at engineers rather than consumers. And the density is real. Google, Apple, Meta, Nvidia, OpenAI, and Anthropic all sit within about an hour's drive of each other, packed into a metro area roughly the size of Japan's Kansai region. Being surrounded by that concentration of frontier work does something specific: it shows you exactly where the edge of the field currently sits, and where your own product is sharp versus where it's soft. That clarity is hard to get secondhand. ## Reading about AI online isn't the same as standing in the room You can follow every AI paper and product launch from anywhere with an internet connection. What you can't get from a feed is the pressure. Share a big AI news story in Slack from outside the Bay Area and you'll usually get a polite "huh, neat." Share the same story here and people start planning a meetup within the hour. Events move at a different speed. On the event platform Luma, someone can post a meetup three days out and still pull in a few hundred engineers. One recent AI event drew over a thousand signups within 24 hours of the announcement, more than the venue could hold. That "something big just dropped, let's get together tomorrow" reflex is just how people operate here. I recently sat in on a knowledge-graph meetup that would be a niche topic almost anywhere else. Here it pulled 200-plus people, from Stanford faculty to junior engineers at seed-stage startups, all talking as equals. There's also a "move fast and break things" instinct running underneath all of it: fewer people trying to plan the perfect answer, more people just shipping something and seeing what happens. And a lot of the most useful information never makes it onto a conference stage or a blog post. It shows up at a house party, over a beer, when a founder or investor tells you honestly what didn't work for them, or what GTM approach is quietly outperforming everything else right now. Getting access to that kind of conversation is one of the clearest reasons to actually be here rather than watch from a distance. ## What we're researching: agents as customers, and why AI projects fail The industry is shifting from copilots, AI that assists a person, to autopilots, AI that completes work on its own. I've watched this play out firsthand with Tesla's Full Self-Driving: door-to-door trips with no human intervention, something that felt theoretical not long ago and now doesn't. That shift raises a question we think about constantly: once agents act autonomously on someone's behalf, what does the org chart look like, and what should a person actually spend their day doing? That's an open, active debate here, not a solved problem. It also changes what a knowledge base is for. Search boxes and chat are the current interface, but the next one probably lets an agent understand intent and complete the action behind the scenes without being walked through it step by step. That points to something we call B2A, business to agent: if an AI agent is going to act as your customer's proxy, doing the researching and deciding on their behalf, it needs a knowledge foundation built to serve an agent, not just a person clicking through pages. That's the core research question for HACK Lab. The other half of the work is less glamorous but just as important: studying why AI projects fail. MIT research has found that the large majority of generative AI pilots don't deliver a return. We want to understand why firsthand, by talking directly to teams that tried and hit a wall, and feed those lessons back into our own product and go-to-market decisions. ## How HACK Lab actually runs HACK Lab is a rotation, not a permanent relocation. Engineers and PMs from Japan come out for a few weeks at a time. Monday through Thursday, they keep working their existing projects remotely. Friday is set aside entirely, no regular work, for hands-on exploration of whatever's new. It's modeled loosely on Google's old 20% time: protected space to try something with no immediate deliverable attached. That unstructured time matters. Some of the best ideas won't come from a scheduled meeting; they'll come from a hallway conversation in a shared kitchen during a short, intense stretch together. Rakusai is aware of the obvious risk: when leadership spends real time somewhere else, it can widen the gap in visibility and pace with the team back home. He's planning to address that with more deliberate internal communication. But he'd rather take that risk than manage the effort from a distance. He wants to be the one at the front of this, not the executive telling everyone else to move faster while staying put. There's a reason to believe the combination works. Helpfeel's engineering team has spent years meeting the standards of Japan's most demanding enterprise customers, banks, retailers, manufacturers, where a wrong answer is worse than no answer. That discipline is already in our foundation. Pair it with Silicon Valley's willingness to ship an imperfect version and learn fast, and you get something neither culture produces alone: quality and speed together, not quality or speed. If you want more on where we're headed as a company, read [why Helpfeel exists](/blog/why-helpfeel). --- ## Things That Are Knowable Today (And Weren't a Few Years Ago) Source: https://en.helpfeel.com/blog/things-knowable-today # Things That Are Knowable Today (And Weren't a Few Years Ago) A few years ago, most of this list was either too expensive to build or technically out of reach. Now it's a weekend project. Here's what changed, category by category. ## Support and product You can now see the root cause of every ticket, not a sample of them. You can tell which help articles get read but still don't solve the problem. You can catch, in real time, every search that comes back empty. You can measure doc staleness against what the product actually does today, not what it did at launch. You can read sentiment across 100% of interactions instead of a survey with a 4% response rate. None of this required new ideas. It required someone to write the code cheaply enough that reading every ticket costs less than reading a sample used to. ## Sales and customer signal Talk ratio, filler words, and question density per call used to live in a sales coach's head. Now they're a transcript away. You can spot deal risk from the language in an email thread before anyone flags it manually. You can predict churn from usage decay weeks before the cancellation email arrives. You can search every call transcript your whole team has ever had, in plain English, and find the pattern instead of hoping someone remembers the call. ## Content and marketing You can see which sentences actually get read versus skimmed, not just which page got a pageview. You can run a simulated reader against a draft and get a real comprehension score before it ships. You can measure content performance for an audience of one person, instead of averaging across a cohort and losing the signal. You can catch brand voice drift across thousands of pages automatically, instead of finding out from a customer complaint. ## Engineering and ops You can estimate the blast radius of a code change before it merges, not after it pages someone at 2am. You can price out cost to serve per customer per channel, down to the interaction. You can query every log line your systems have ever written, in plain English, and get an answer instead of a grep session. ## What this looks like in manufacturing Every category above holds inside a manufacturing company too, but the department mix looks different than it does at a software company. Here's the same shift mapped onto a plant floor and the offices around it. ### Marketing You can see which technical datasheets and spec pages an engineer actually reads mid-evaluation, versus which ones they open once and abandon. You can pinpoint where a distributor RFQ (request for quote) form loses people, at the sentence level, instead of settling for a conversion rate. You can catch brand and spec inconsistency across product pages localized into a dozen languages for regional distributors, automatically, instead of hearing about it from a distributor complaint six months later. You can comprehension-test technical content before it ships, for readers who are engineers first and a marketing audience second. ### Operations You can correlate PLC and SCADA log lines with maintenance tickets and technician notes in plain English, instead of paying a specialist to read raw logs line by line. You can find the root cause across every defect on the line, not a sampled batch, so a recurring failure mode surfaces after ten units instead of ten thousand. You can estimate the blast radius of a firmware or software update to a fleet of machines in the field before it ships, instead of finding out after a service call. You can price cost to serve per SKU per distribution channel, down to freight and warranty cost, not just landed cost. ### Support and service You can find the root cause of every field service ticket, not a sample, so the third recurring failure on a machine gets caught at three, not thirty. You can tell which pages of the service manual and parts catalog actually answer a technician's question in the field, versus which ones send them straight to a phone call with engineering. You can spot warranty claim patterns across every claim submitted, in every language a distributor operates in, not just the ones a regional team happens to read. You can measure support content staleness against the actual firmware or hardware revision a customer has, not the revision the manual was written for. ### Elsewhere in the plant Sales can see deal risk in the language of an RFQ back-and-forth before a quote goes cold. Quality can query the full log of inspection notes in plain English instead of exporting everything to a spreadsheet first. Engineering can see the blast radius of a design change against every open project that depends on that part number. None of this is hypothetical. It's the same shift as the general list above: the cost of reading and understanding unstructured text (log files, service tickets, spec sheets, inspection notes) dropped enough that reading all of it now beats reading a sample. ## Why this list didn't exist five years ago Every one of these was possible in theory before. What changed is the cost of writing the code and building the tooling. Reading and understanding unstructured text at scale used to require a team of engineers and months of work. Now it requires a prompt and an afternoon. The bottleneck moved from "can we build this" to "did we think to ask." That's the real shift. The technical ceiling didn't just move up, the floor to reach it dropped through the basement. Things that would have been a full quarter's roadmap item are now something one person tries before lunch. If you're running a support or knowledge operation, manufacturing or otherwise, and want to see what this looks like applied to your own tickets and articles, [see how Helpfeel works](/platform). --- ## Don't Produce Noise, Produce Music Source: https://en.helpfeel.com/blog/dont-produce-noise-produce-music # Don't Produce Noise, Produce Music Every support team is sitting on a pile of data. Ticket counts, contact volume, CSAT scores, raw search logs. Most of it is noise. Noise is sound with no pattern. It tells you something happened, but not what to do about it. "Tickets went up 12% last month" is noise. It does not tell you which question to answer, which article to write, or which word your customers are actually typing. Music is different. Music has rhythm, melody, and a structure you can follow. The same raw signal that sits in your logs as noise can become music, a steady rhythm of customer intent that tells you exactly what to play next. Here is how Helpfeel makes the switch. ## The instrument matters Most search bars hear a few keywords and guess. A customer types "can I send it back after the holidays" and the system strips it down to "return," then serves whatever article happens to match that one word. The full phrase, the real intent, gets thrown away before anyone can read it. Helpfeel's intent-prediction search captures the whole phrase. It is built to match the way people actually ask, in their own messy, specific words. That single design choice changes what lands in your logs. Instead of a stripped keyword, you get a recording of the real question. That recording is your raw material. You cannot make music from a broken instrument, and you cannot read voice-of-customer from a search engine that discards the voice. ## Reading the score Once the question is captured cleanly, a handful of metrics turn into a score you can read. Each one is a different line in the music. **Search keywords are the melody.** The actual phrases customers type, ranked by frequency, are the clearest signal you will ever get about what your audience wants. Not what you assumed they want. What they typed. **The no-hit rate is the missed note.** Every search that returns nothing is a customer who asked and got silence. A rising no-hit rate is the most actionable metric in support, because each gap is a specific article you have not written yet. **The self-service rate is the melody landing.** It tells you how often the song resolves, how often a customer found their answer and never needed to open a ticket. Watch it climb as you fill the gaps the no-hit rate exposed. **User-behavior signals are the rhythm underneath.** What people search at 9am versus 9pm. The phrases they use that do not match your official product names. The second and third searches that reveal a question your first answer did not fully close. Put those lines together and you stop guessing. You are reading a score. ## When the words don't match your names The richest signal is usually the gap between what you call something and what your customers call it. When LUSH Japan looked at its help center search data, the team found something simple and important. LUSH runs a container-return program called "Bring It Back" and a subscription called "Fresh & Flowers." Customers were not searching for those names. They were typing "container exchange" and "subscription." The official name lived in the marketing department. The customer's word lived in the search bar. Helpfeel's Customer Success team helped LUSH map the informal phrases to the right articles, so customers found answers using the words that were already familiar to them. Ecommerce contacts dropped and wrapping-paper sales climbed. None of that happens if the search log only stores "Bring." Hamee saw the same pattern from a different angle. "Cancellation" kept surfacing as a top search term. That one recurring note made a hidden demand visible, the team built a new self-service workflow within a month, and post-dispatch cancellations that had been driving excess shipping costs went away. The search log was not noise. It was a customer telling them, over and over, what to fix. ## Music takes a band A score on paper is not a performance. The reason Helpfeel data becomes music and not just a prettier pile of numbers is the cadence around it. Customers like KOMEHYO run a monthly improvement cycle with the Helpfeel Customer Success team. Expand the target search keywords. Enrich the content to match what people are actually asking. Refine based on user-behavior analysis. Then meet again and check whether the new content resolved the question or just got viewed. That is the rhythm section. Hokkaido Bank made the no-hit rate a priority KPI and treated every gap as something to address immediately, not next quarter. The report is not a trophy you look at once a month. It is sheet music you play from. This is the part that does not show up in a feature comparison. You get a clean instrument, a readable score, and a band that already knows how to play. The work of reading the signal and turning it into the next article is done with you, not dropped on your desk as a dashboard you have to decode alone. ## From noise to something you can hum Self-service really is service when it is done right. A help center that answers the question your customer actually typed, in the words they actually used, is one of the most respectful experiences you can offer. That only happens when the signal stays intact from the search bar to the report to the next article you write. Capture the real question. Read the rhythm. Fill the missed notes. Then do it again next month. Stop producing noise. The music was in your search logs the whole time. If you want to see what your own customers are really asking, [take a look at how Helpfeel works](/product). --- ## CX Heroes Podcast #50: Everything In Support Can Be Automated. That Doesn't Mean It Should Be. Source: https://en.helpfeel.com/blog/cx-heroes-tim-thijsse-when-not-to-automate-support # CX Heroes Podcast #50: Everything In Support Can Be Automated. That Doesn't Mean It Should Be. Ask Tim Thijsse whether the support function will ever be fully automated and he gives you a one-word answer. Yes. Then he spends the next two minutes explaining why that is the wrong thing to aim for. Tim is a UX and customer experience strategist operating from the Digital Experience Collective and the author of Maturing in Customer Experience Optimization. He is a top 50 CX leader in 2025 and again in 2026, and he has led digital optimization programs for brands like Heineken and E.ON. This is someone who has spent his career on the technical, quantitative side of CX. So when he draws a line around automation, it is worth paying attention to where he draws it. ## The example that makes the point Picture an insurance customer whose partner has just died. They need to cancel a policy. You can automate that. Send a link, point them to a chatbot, let them close the account in thirty seconds. It is fast, it is efficient, and it is the wrong call. Here is how Tim puts it: > "If somebody's passed away, it might be best to send a handwritten card, for example, or not send somebody to a chatbot, but ask for a certain time so you can have a call with that person, because that person just lost someone." The interaction is simple to automate. The moment is not simple at all. And confusing those two things is how companies end up with support that runs smoothly and still leaves people feeling processed. ## The skill is knowing which lever to pull Tim's framing flips the usual order of operations. Most teams automate by default and try to add the human touch back in later, usually after a customer has already had a cold experience. Tim decides where empathy has to live first, then automates everything around it. > "This doesn't mean that you have to be automated at 100%. This means that you really understand what you should pull and when you should use that empathy of your employees and let them connect to the customer." That is the part most automation projects skip. The question is not how much you can automate. The question is which interactions earn a person, and how you design the system so those moments are protected instead of swallowed. ## Why this matters more as AI gets better The better automation gets, the easier it becomes to route everything through it. That is exactly when the line Tim draws becomes valuable. A customer canceling a duplicate subscription and a customer canceling after a loss can look identical to a system and could not be more different to a human. This is the worldview behind how we build Helpfeel. Good AI support is not about removing people from every interaction. It is about answering the high-volume, repeatable questions so reliably that your team has room for the moments that actually need them. Helpfeel resolves up to 70 percent of incoming questions and answers 98 percent of searches with a relevant result, which is what gives a support team the time to show up where a person belongs. The automation handles the routine so the humans can own the moments that carry weight. Tim's episode is a good reminder that the boundary between automated and human is a design decision, not an accident. Draw it on purpose. [Watch: CX Heroes: Tim Thijsse on when not to automate support](https://youtu.be/Nl0nB7QQkFE) [Watch the full episode of CX Heroes with Tim Thijsse →](https://www.youtube.com/watch?v=Nl0nB7QQkFE) --- ## Let Your Customers Write Your Knowledge Base Source: https://en.helpfeel.com/blog/voice-of-customer-knowledge-updates # Let Your Customers Write Your Knowledge Base This is an opinion piece, not a how-to. No step-by-step, no checklist. Just a few things we keep seeing, and what we think they mean. Each section is the same shape: here is a fact we observe, and here is where we land on it. ## The audit is already out of date the day you finish it **What we see.** Most support teams treat the knowledge base like a project. You scope it, you write it, you ship it, and you put a recurring event on the calendar to review it next quarter. Between those reviews, it sits there. Search behavior moves. The product ships three releases. A pricing page changes. The articles do not. **Where we land.** A quarterly audit is a snapshot of a moving target. By the time you sit down to do it, you are correcting three months of drift at once, from memory, in a single afternoon. That is not maintenance. That is archaeology. The cadence is the problem, not the effort. ## Your customers are already telling you what to write next **What we see.** Every day, people search your help center and find nothing. They click "was this helpful?" and choose no. They give up and open a ticket about the same thing your last twelve customers asked. Each of those is a signal with a timestamp and a clear instruction attached. **Where we land.** That stream of unanswered searches and repeat tickets is the most honest content brief you will ever get. It is voice of customer, and it is free, and it is sitting in your tools right now. We think the best knowledge base is not the one written by the smartest person on your team. It is the one written by the questions your customers actually ask, in the words they actually use. ## A help center should update on the signal, not on the calendar **What we see.** The gap between "a customer hit a dead end" and "someone wrote the article" is usually measured in weeks or quarters. Often it never closes at all, because the person who noticed the gap is not the person who writes the docs, and the handoff dies in a backlog. **Where we land.** The unit of work should be the signal, not the audit. A failed search should be able to trigger a draft. A spike in repeat tickets should surface the missing article on its own. The loop we believe in is simple: signal comes in, the gap gets detected, a fix gets drafted, a human approves it, it goes live. Continuously. The customer asks, the answer appears, the next customer never has to ask. ## "Self-improving" should still have a human in it **What we see.** Plenty of vendors will now promise a knowledge base that maintains itself. Point an AI at your tickets, walk away, done. We have watched enough auto-generated content go subtly, confidently wrong to be skeptical of "walk away." **Where we land.** We want the machine to do the noticing, the drafting, the routing, all the parts that do not scale by hand. We do not want it to do the approving. A person who owns the voice and knows what is true should still say yes before anything publishes. The goal is not a knowledge base with no humans. It is a knowledge base where humans stop doing the boring 80% and spend their judgment on the 20% that needs it. That is the next hire you will not need to make, not the team you replace. ## Where we land, in one line Stop scheduling the audit. Start listening to the signal. Your customers have been writing your knowledge base all along. The only real question is whether you are reading it. If that is the way you already think about your help center, [see how Helpfeel works](/product). It was built around exactly this loop. --- ## Your Customers Stopped Starting at Your Help Center Source: https://en.helpfeel.com/blog/customers-start-support-on-google-not-your-site # Your Customers Stopped Starting at Your Help Center More than half of customer service journeys now start somewhere you do not control. Gartner puts it at 51%. In a 2025 survey of 5,801 consumers, Gartner found that the majority of service journeys begin on third-party platforms like Google, YouTube, and ChatGPT, with search engines the most common first stop. Your help center is no longer the front door. It is the second or third room people reach, if they reach it at all. That is a quiet shift with loud consequences. ## The moment of need moved upstream Picture the actual behavior. A customer hits a problem at 9pm. They do not log into your portal. They open a search bar or an AI assistant and type the question in plain words. Whatever answers first, wins the moment. If the answer comes from a forum post, an old review, or a confident-but-wrong AI summary, that is the version of your product the customer now believes. You were not in the room. The answer happened to you. ## Your knowledge is either extractable or invisible Here is the part most teams miss. Search engines and AI assistants do not reward the company with the biggest help center. They reward the one whose answers are clean, structured, and easy to lift out of context. A buried PDF, a login-gated portal, a wall of legacy FAQ pages written for an internal audience: none of that gets surfaced. The engines cannot read it, so the customer never sees it. You can have the right answer and still lose, simply because it was not in a form a machine could quote. This is what answer engine optimization actually is. Not a new acronym to chase. A recognition that the question now gets asked outside your walls, and your job is to make sure the correct answer travels back in. ## Your old metrics measured the wrong building For a decade, support leaders optimized the help center as a destination. Sessions, search success, self-resolution inside the portal. Useful numbers. They just measured a building most customers now walk past. The new question is harder and more honest: when someone asks about your product anywhere on the open web, does the right answer show up? That depends on whether your knowledge is verified, current, and structured for extraction. It is a content and architecture problem before it is a chatbot problem. ## What good looks like now Three things separate the companies winning the first answer from the ones getting answered for. Their knowledge is **verified**, so the answer a machine lifts is the answer they would stand behind. Their content is **structured** in self-contained, question-shaped chunks an engine can quote without mangling. And it stays **current**, because a stale answer surfaced confidently is worse than no answer at all. Get those right and the same knowledge base that serves your help center also feeds the search results and AI assistants where the journey now begins. You stop choosing between owning the destination and owning the moment of need. You own both, from one source. ## The takeaway The help center did not die. It stopped being the starting line. Your customers already moved upstream to wherever they can ask a question in their own words and get an answer in seconds. The only choice left is whether that answer is yours. If you want to see what a verified, extractable knowledge core looks like in practice, [take a look at how Helpfeel works](/product). --- ## 10 Overpowered AI Prompts I Actually Use Every Day Source: https://en.helpfeel.com/blog/10-overpowered-ai-prompts # 10 Overpowered AI Prompts I Actually Use Every Day Most "best AI prompts" lists are padding. Long templates nobody types twice. These ten are short, I use them daily, and each one earns its place because of a small phrasing trick that changes the answer you get back. The pattern across all of them: you are not writing a clever incantation. You are telling the model what mode to think in, and then getting out of the way. ## 1. Errors ``` (your error or issue), rca and fix ``` RCA means root cause analysis. Those three letters do the heavy lifting. Instead of pattern-matching to the nearest common fix, the model re-examines its own assumptions and usually finds the thing it skipped the first time. You get the actual cause, not a band-aid on the symptom. ## 2. Planning ``` (your plan ideas here), be skeptical, evaluate the plan, ask follow-ups ``` "Be skeptical" strips out the buttery agreement and surfaces what you are about to miss. "Ask follow-ups" is the quiet hero: it forces the model to tell you what context it still needs before it commits, instead of confidently guessing and being wrong. ## 3. Ship or not ``` (your idea or draft), ask questions, then summarize the new plan ``` This pulls the unknowns out before anything gets built, not after. The summarize step plays the plan back to you in its own words, so you catch the spot where it misread you while a fix still costs nothing. ## 4. Autonomy ``` (the task), use your best judgment and one shot it ``` When you trust the call, this removes the check-in loop. The model commits to a finished deliverable instead of stopping to ask permission halfway through. Use it for the low-stakes work you would rather just have done. ## 5. Distill ``` (messy notes or a long doc), boil it down to three specific takeaways ``` A hard number kills the wall of text. "Three" forces a real cut. "Specific" blocks the vague mush, because each takeaway now has to stand on its own. Swap in ideas, options, or risks depending on what you need. ## 6. Reconsider ``` this is a bad solution. reconsider as a skeptical senior engineer. (why it is wrong) ``` This is the nuke for a lazy first answer. Naming the failure mode is what makes it work. "You copied the file once, but I need it to sync continuously" sends the model back to the architecture instead of patching the surface. Works outside code too, any time the first answer solved the wrong problem. ## 7. Voice pass ``` (your copy), strip the AI tells and report what you caught ``` Run your draft back through and have the model remove its own fingerprints: the "it is not X, it is Y" cadence, the throat-clearing, the over-balanced hedging. Asking it to report what it caught is the bonus. You see the edits, so it doubles as a running lesson in your own taste. ## 8. The panel ``` (your challenge), convene a panel of three expert personas to debate it ``` Pick three people whose track records actually fit the problem and have the model argue the question from each of their points of view, brutally honest, one short verdict each. You get disagreement instead of one averaged opinion, and disagreement is where the real signal lives. Have it list where they agree and where they split. ## 9. Steelman ``` (a plan you are leaning toward), argue the strongest case against this, then tell me if it survives ``` This separates "I like this" from "this is right." The model builds the best possible attack on your own idea, then rules on whether the idea is still standing. If it survives a real assault, you ship with conviction instead of hope. ## 10. Lean it ``` (a full plan), now give me the 20 percent version. one person plus AI, skip the nice-to-haves ``` You were going to ask for the leaner cut anyway. This gets there in one move. It forces the ruthless question every plan needs: what actually moves the needle, and what was just there to look thorough. ## The thing they have in common None of these are long. The leverage is never in the word count. It is in naming the mode you want the model to think in: skeptical, root-cause, distill, steelman. Say that part out loud and the quality jumps. Steal all ten. Keep the ones that fit how you work, and rewrite the rest in your own words. The best prompt is the one you will actually type tomorrow. If you want to see what this kind of thinking looks like applied to customer support, [see how Helpfeel works](/product). --- ## CX Heroes Podcast #48: The Support Agent Role Is Being Rewritten. Most Teams Aren't Building For It Yet. Source: https://en.helpfeel.com/blog/cx-heroes-amanda-carden-chamberlain-support-role-2028 # CX Heroes Podcast #48: The Support Agent Role Is Being Rewritten. Most Teams Aren't Building For It Yet. The support agent job description is being rewritten right now, and most support leaders have not noticed yet. Amanda Carden-Chamberlain has been in customer experience for 17 years, across food service, e-commerce, wine distribution, and enterprise technology. She is the Director of Customer Success at RJ Young, a 70-year-old Southeast US business technology company that grew from printers and copiers into full managed IT infrastructure during COVID. She manages 30 products and hit 95 NPS last month. When she describes what the role looks like in 2028, it is not a support agent anymore. ## What is left when AI takes the routine work Amanda's logic is simple. When AI absorbs the password resets, the order lookups, and the ticket entry, what remains for the human on the team is the work that actually requires a person. The emotionally charged interaction. The problem with no obvious solution. The customer who needs to feel genuinely heard by someone who understands the situation. She has a name for the person who does that work. A CX champion. Someone whose entire job is the high-stakes conversation. The implication is significant, and it is where most teams fall behind. If the repetitive work disappears into automation, the people who remain have to be equipped for a completely different kind of interaction. Hiring criteria change. Training changes. What success looks like changes. Most support teams are not building toward that yet. They are automating the easy half of the job and assuming the people will sort themselves out around it. ## The structure is the strategy Ask Amanda what makes 95 NPS possible and she points at the org chart, not a tool. RJ Young runs three separate support teams, each built around a specific product set, each accountable for becoming genuinely expert in its area. > "A master of all is a master of none. We try to have it so that each team has their focus so that they can truly become experts in that area." The structure is the strategy. The score is the evidence. That focus is also exactly what makes a CX champion role workable: you cannot ask someone to handle the hardest conversations in a category they only half understand. ## What a credible AI pitch has to clear Amanda deletes most AI vendor pitches before finishing the first message. She knows the landscape better than most of the people selling to her, and she has a fast filter. > "Any time somebody comes to me with a right off the bat first message pitch of 'This is going to solve all of your problems,' I am not going to believe that, and I will probably not even look into it." Her evaluation starts with one question: tell me about a company in my space where this worked. Most vendors cannot answer it. There is pressure from above, too. The same pitch landing in her inbox is also landing with her CEO and her C-suite. Because they do not always know the inner workings of what it takes to take care of a customer, the pressure comes back down as a question: why are we not already doing this? Amanda has had to build the case from the inside, and credibility is the currency. RJ Young already runs Agent Assist, surfacing resources for agents mid-conversation in real time, and she is actively evaluating the next layer. ## The takeaway for CX leaders If you are planning your support org for the next three years, Amanda's view is a useful gut check. Automation is not the finish line. It is the thing that changes what your remaining people need to be. The teams that win will redesign the human role on purpose, hire and train for the high-stakes conversation, and refuse to let the easy wins of automation define the whole strategy. That is also how we think about Helpfeel. The point of taking routine questions off your team's plate is not a smaller team. It is a team pointed at the work that only people can do. The next hire you make should be a CX champion, not a ticket router. The full episode covers how Amanda builds the internal case for AI, what 17 years across four industries taught her about reading a vendor, and the role she believes every support team will be hiring for by 2028. [Watch: CX Heroes: Amanda Carden-Chamberlain on the support role of 2028](https://youtu.be/00Pq7Cg2ws8) [Watch the full episode of CX Heroes with Amanda Carden-Chamberlain →](https://www.youtube.com/watch?v=00Pq7Cg2ws8) --- ## CX Heroes Podcast #49: Your Customer Can't Be Honest With Your Sales Rep Source: https://en.helpfeel.com/blog/cx-heroes-brian-porter-honest-customer-feedback # CX Heroes Podcast #49: Your Customer Can't Be Honest With Your Sales Rep A sales rep spent three years taking his customer to ballgames and dinners. Brian Porter sat down with that same customer for one 45-minute conversation, no agenda and nothing to sell, and walked out knowing more than the rep had gathered in three years. The rep called him afterward, half in disbelief. > "You found out more in 45 minutes than we have in three years." Brian is the chief customer officer at Ernest, a packaging and distribution company he has helped grow from about 65 people to nearly 800. Last October his CEO handed him a new title and a loose mandate: go meet our top customers, get the highest-level decision maker in the room, and find out how we are really doing. No playbook. Figure it out. He has now done 60 of those meetings. What he learned says something uncomfortable about how most companies collect customer feedback. ## The relationship is the filter Here is the part most teams miss. Your customer is not withholding the truth because they are difficult. They are protecting the relationship. When the person across the table has a quota and years of golf invitations riding on the answer, the customer softens it. They round up. They leave the real frustration unsaid, because saying it out loud feels like betraying a friend. The dynamic that makes a rep effective at closing is the same dynamic that keeps honest feedback from ever reaching them. This is structural, not personal. No amount of rapport fixes it, because rapport is the thing causing it. Brian carries none of that weight into the room. No deal on the line, no number to hit. He shows up, asks one open question, scores the relationship one to ten, and lets the silence do the work. The decision maker finally says the thing they would never say to the rep. ## He built the whole program with AI before he ran a single meeting Before his first meeting, Brian sat down with ChatGPT, which he named Sally, and asked the basic question: what does a chief customer officer actually do? Sally pointed him to the research. He worked through "Customer Centricity" by Peter Fader and "The Ultimate Question," the book behind the net promoter score. Then he went into the real world. He asked a CEO friend what a great CCO meeting felt like, and what a bad one looked like. The bad one was simple. The CCO tried to sell, and that account never took a second meeting. Every detail went back to Sally. What to wear. Whether to carry an iPad or plain paper. How to take notes. AI did the prep. Brian did the room. That line is worth sitting with, because it is the opposite of how most companies are deploying AI into customer conversations right now. ## The one rule that earns a second meeting A chief customer officer who tries to sell in the meeting will never get invited back. So Brian does not. He asks how they are doing, scores it, asks for the next step, and follows up. When it is an owner or an executive, the follow-up is a handwritten note by mail. Old school, and it works every time. ## What 60 conversations surfaced Problems that had been sitting for months are finally getting fixed, because someone with no agenda elevated them. One rep had been stuck on an issue for two and a half months with no movement. Brian's meeting surfaced it, and within days it was handled. His goal is 250 of these conversations by the end of the fiscal year. Everyone tells him that is crazy. He is doing it anyway. There is a lesson here for any support or CX leader, and it is not "go hire a CCO." It is that the channel you use to listen shapes what you are allowed to hear. A rep hears a filtered version. A survey hears a polite version. A customer talking to someone with no stake in the answer finally tells you the truth. That is also the quiet case for getting your self-service layer right. When customers can answer their own routine questions instantly, the human conversations that remain get to be the high-value ones: the honest, unfiltered, relationship-deepening kind that Brian is having 60 times over. Helpfeel exists to take the repetitive load off your team so the people on it are freed up for exactly that work, not removed from it. The full episode covers the Toyota account Ernest lost over a forklift and a parking lot, how Brian uses AI without ever letting it touch the customer relationship, and why a packaging company rebranded around its people. [Watch: CX Heroes: Brian Porter on honest customer feedback](https://youtu.be/zXVunGmb0n8) [Watch the full episode of CX Heroes with Brian Porter →](https://www.youtube.com/watch?v=zXVunGmb0n8) --- ## Why Helpfeel Exists Source: https://en.helpfeel.com/blog/why-helpfeel # Why Helpfeel Exists Hello from Helpfeel. This is the first real post on our US blog, so it should answer the first fair question: who are you, and why are you here? [Helpfeel Inc.](https://corp.helpfeel.com/) is a Japanese software company with a mission we take literally: human empowerment through technology. For over a decade we've built tools that help people find, share, and use knowledge, including Gyazo, used by millions to capture and share what's on their screen, and Cosense, a knowledge base where teams think together. Helpfeel, our AI support platform, is where all of that converges. Today, 625+ companies run their customer support on it, including names like MUJI, Honda, LUSH, and Panasonic. 99% of them renew every year. ## The story we kept seeing We came to the US because we kept seeing the same story here. A support team launches a help center, configures a bot, writes the articles. Six months later the content is stale, the bot answers wrong, customers stop trusting it, and tickets climb right back up. The team did everything right. The tools just had no way to hold onto their effort. That story is the reason Helpfeel works the way it does. ## What we believe We believe AI support should be prepared, never improvised. Most AI tools are switched on and left to guess from whatever content exists. Helpfeel's AI answers only from knowledge that humans have reviewed and approved, which is why banks and insurers in Japan trust it with questions where a wrong answer is worse than none. We believe the work should compound. Every search, click, and contact message teaches the system what customers actually need. Gaps become drafts, drafts become verified answers, and month twelve performs better than month one. We call this model The Loop, and we'll be writing much more about it here. And we believe support teams deserve a partner, not a login. Every Helpfeel plan includes experts who build, tune, and improve your knowledge base continuously. Your team stays focused on customers. Think of us as the next hire you won't need to make. ## Save the day Support people save the day for customers, every day, usually without much recognition. Our job is to hand them better equipment. That's why we do what we do. If you want to see it working, take a look at [what the platform includes](/platform) or [book a demo](/book). And if you just want to follow along, this blog is where we'll share what we're learning as we bring Helpfeel to the US. Talk soon. --- ## 0.47-Point Drop in Inquiry Rate: How SpaceMarket Built a FAQ Users Actually Use: SpaceMarket Inc. Source: https://en.helpfeel.com/case-studies/0-47-point-drop-in-inquiry-rate-how-spacemarket-built-a-faq-users-actually-use # 0.47-Point Drop in Inquiry Rate: How SpaceMarket Built a FAQ Users Actually Use - Inquiry rate decreased 0.47 percentage points after Helpfeel introduction - Self-resolvable inquiry ratio reduced 2.3 percentage points - Six-person CS team went from spending all day on inquiries to strategic work - Guest vs. Host content clearly tagged for targeted self-service ## The Challenge SpaceMarket operates a platform where users rent and lend spaces by the hour. As the product grew, so did inquiry volume. The six-person Customer Success team was spending virtually all day, every day handling contacts. Most inquiries came from Guests asking about reservations, questions they could have resolved themselves by checking the FAQ or that only required a template response. The team had already launched an internal project to optimize FAQ content within their existing tool. The project stalled. They were looking at page view metrics through analytics, but without specialist knowledge to interpret the data and act on it, improvements couldn't be sustained or properly evaluated. Reducing self-resolvable inquiries became a key KPI, but the path forward wasn't clear. ## Why Helpfeel SpaceMarket considered chatbot solutions but ruled them out. A chatbot risked redirecting questions that should go directly from Guests to Hosts (questions about the specific space) to the platform operator instead. Helpfeel met the team's non-negotiables: superior search performance and the ability to clearly tag each piece of content as Guest-facing or Host-facing, making it easier for users to find information relevant to their role. The monthly review meetings and ongoing support were equally important. The team lacked in-house FAQ management expertise, and Helpfeel's accompaniment model ensured the tool wouldn't just be implemented and left alone. ## What They Did - **Redesigned the FAQ structure from instruction manual to one-question-one-answer format.** Early in implementation, Helpfeel's team pointed out that SpaceMarket's original pages were dense with information the team wanted to convey, resulting in long-form content users would abandon without reading. The shift to one-question-one-answer with answers visible at a glance changed the team's entire mindset around content creation. - **Refocused from "getting users to the FAQ" to "getting users to resolve their questions."** The monthly data-driven review sessions helped the team recognize that page visits don't matter if users aren't resolving their questions. Understanding who was using the FAQ, where they were dropping off, and why refocused efforts on content quality and problem resolution. - **Concentrated improvements on high-traffic Guest-facing pages.** Data analysis revealed that most traffic came from Guest inquiries, so the team prioritized those pages for redesign. - **Created targeted content to route appropriate questions to Hosts.** Search data showed many Guests were looking up day-of-visit logistics like garbage disposal or parking availability. These questions can only be answered by the individual Host, not the platform operator. The team created a dedicated page explaining what types of questions should be directed to the Host and engineered it so that various search keywords related to day-of logistics would surface this page. ## Results SpaceMarket measures inquiry load as an "inquiry rate," the number of inquiries generated per number of users who completed a booking. After introducing Helpfeel, the inquiry rate decreased by 0.47 percentage points. The share of total inquiries concerning topics users could have resolved on their own decreased by 2.3 percentage points. At the time of the interview, the team noted that approximately half of all inquiries were still on topics users should be able to self-resolve, indicating significant room for further improvement. ## Looking Ahead SpaceMarket's near-term goal is reducing the share of self-resolvable inquiries still coming in as contacts. The key is designing a site structure where users can reach the answer they need on their first search attempt, and where even if the first search doesn't immediately succeed, users can navigate the FAQ and find their way to a resolution independently. A longer-term ambition is making the FAQ useful for new employees when they join. New staff are closer in perspective to users than veterans. If even they can resolve questions quickly and easily with the FAQ, it becomes a genuinely user-friendly resource. The team plans to continue working with Helpfeel's ongoing support to build the FAQ structure best suited to SpaceMarket as a service. --- ## 1.2x Member Growth, Fewer Inquiries: How Mogumo Tripled Priority Subscriptions: Oxxx Inc. (Mogumo) Source: https://en.helpfeel.com/case-studies/1-2x-member-growth-fewer-inquiries-how-mogumo-tripled-priority-subscriptions # 1.2x Member Growth, Fewer Inquiries: How Mogumo Tripled Priority Subscriptions - Member count grew 1.2x while email inquiries decreased - Priority subscription contracts tripled compared to prior year - Response time improved to under 16 hours, down from over 20 - Zero additional headcount needed despite growth ## The Challenge Oxxx Inc.'s frozen toddler meal delivery service "Mogumo" grew rapidly after launching in 2021, quickly reaching 2,200 emails and 800 phone calls per month. The team couldn't keep pace. With just 2 staff handling email and 3-5 on phones, they routinely missed their one-business-day response target. Their customers, busy parents juggling childcare and housework, contact support in short windows of free time after children are asleep or during lunch breaks. But the call center closed at 5 PM, and email responses often waited until the next day. The team knew their customers needed a self-service option that worked on their schedule, not the support team's. The FAQ existed, but customers weren't finding it. ![Mogumo frozen baby-food products](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/mogumo_body.jpg) ## Why Helpfeel The team's predecessor selected Helpfeel based on its intent-prediction search, which surfaces relevant articles even when customers use approximate or imprecise terms. For a customer base searching in moments of distraction and exhaustion, this was critical. Case studies from similar companies that had solved comparable challenges gave the team confidence the solution would work for their audience. ## What They Did - Created visual, screenshot-based FAQ articles for common member page operations, directly addressing "how to use" questions that had previously required typed step-by-step email responses - Pinned the "Ganbaru Kazoku Ouen Course" (Family Support Course, their highest-priority subscription plan) to the top of the FAQ homepage and frequently viewed terms section to increase awareness - Strengthened traffic routing to the shipping fee page, which links to the priority course - Redirected searches for specific meal quantities ("18 meals," "12 meals," "8 meals") to surface the shipping fee and course pages - Corrected a visual similarity issue flagged by Helpfeel's Customer Success team where customers searching for "defrost" accidentally landed on the "cancellation" page ## Results Member count grew 1.2x after implementation, yet email inquiry volume decreased. The response time KPI of "within 20 hours" had been consistently missed before Helpfeel. The most recent month achieved response time within 16 hours, a reduction of more than 4 hours on average. Phone inquiries remained roughly flat at approximately 800 calls per month, and the team required no additional headcount, contributing to labor cost containment. Staff who previously typed out step-by-step instructions in every email now simply attach a link to the relevant FAQ article, reducing handling time considerably. The priority subscription course saw dramatic results. Contracts for the "Ganbaru Kazoku Ouen Course" tripled compared to March of the previous year. At Helpfeel launch in October 2024, this course's page accounted for just 6.9% of total FAQ page views. By December 2024, it grew to 14.6% of total FAQ traffic. By January 2025, it reached 16.7%, becoming the most visited page in the entire FAQ. This coincided with a PR push, and the combined effect was clearly reflected in membership growth. The team has been making steady progress toward their KPI of a 9% cancellation rate, attributed to the improved self-service environment and faster resolution of issues. ## Looking Ahead The Helpfeel Customer Success team sends detailed improvement suggestions after every monthly meeting, enabling same-day FAQ updates in the fastest cases. For Mogumo, a recently launched service with limited time and resources, the partnership continues to deliver measurable business impact while keeping the support experience simple for the parents they serve. --- ## 1.4x Growth in One Year: Building an Internal FAQ to Scale 100+ Stores Without Adding Headquarters Staff: KOMEHYO Inc. Source: https://en.helpfeel.com/case-studies/1-4x-growth-in-one-year-building-an-internal-faq-to-scale-100-stores-without # 1.4x Growth in One Year: Building an Internal FAQ to Scale 100+ Stores Without Adding Headquarters Staff - ~1,100 employees (including part-time staff) using the internal FAQ - 25% of all FAQ views come from mobile devices, higher than typical internal FAQ usage - 1.4x article growth in the first year, with reduced inquiry burden on headquarters ## The Challenge KOMEHYO opened more than 100 new brand resale stores in three years, with another 90 planned over the next three. That pace of growth turned information sharing into a bottleneck. Stores needed answers on everything from campaign transaction processing to paid leave policies, but the existing setup (shared folders organized by whoever happened to create them, with stale permissions and unreliable ownership) made it faster to just call headquarters. The result: headquarters teams across multiple departments were fielding the same questions repeatedly. Worse, store staff who didn't know which department to call would phone whichever team they knew, requiring the person who answered to identify the right group and transfer the call. It wasted time on both sides of every interaction, and it pulled floor staff away from what mattered most: serving customers. ![A KOMEHYO team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/komehyo-staff_body.jpg) ## Why Helpfeel The team concluded that trying to maintain folder structure rules and communicate which department handles which inquiry would require constant effort for limited payoff. A single unified self-service hub, a place you check first whether the question is about the shop floor or about HR, was seen as simpler and more sustainable. Helpfeel was selected primarily for the quality of its ongoing support: monthly improvement recommendations based on actual usage data. The simplicity of the interface for both searching and creating content was also noted as a meaningful factor. ## What They Did - Launched the internal FAQ in May 2023, just one month after the adoption decision, seeding it with approximately 100 articles created from informal Q&A notes Tsuruta had been maintaining for EC-related inquiries - Announced the FAQ to all staff via company-wide chat, making it accessible to approximately 1,100 employees (including part-time staff) via company-issued PCs and smartphones - Expanded monthly review meetings from April 2024 to include representatives from multiple headquarters departments beyond the core team, working toward a truly company-wide unified hub - Enabled part-time staff without individual employee accounts to access the FAQ via a shared account, ensuring universal first-stop coverage ## Results Article count grew to 1.4x the initial launch volume within approximately one year, with content expanding to cover a broader range of topics. The system now provides at least one relevant answer for the vast majority of keywords that are actually searched. Headquarters departments have begun to notice a reduction in the burden of repetitive inquiries, and are proactively requesting that new topics be added. For example, during a recent sale period, the marketing department requested an article on coupon handling procedures, which was added and confirmed to be actively accessed by store staff during the campaign. 25% of all views come from mobile devices, notably higher than the typical rate for internal FAQs. This suggests that store floor staff are actively consulting the FAQ on their phones mid-shift without interrupting their workflow. ## Looking Ahead The FAQ is accessible to all employees including part-time staff via a shared account. For topics that should be restricted to full-time employees or management, the approach will be to link to those resources and manage access permissions at the destination, rather than excluding the FAQ entry entirely. This way, the internal FAQ can serve as a universal first stop for every employee across the organization. Usage is growing steadily, but data indicators suggest significant room for further penetration. Part-time staff in particular, who have higher turnover and frequent onboarding needs, may not yet be fully aware the FAQ exists. The team is exploring incorporating Helpfeel into new employee orientations so that the habit of "checking the FAQ first" is established from day one. The goal remains giving every store the tools to resolve questions on their own, so that staff can spend more of their time doing what matters most: serving customers on the floor. --- ## 1,800 Articles Serving 150 Branches: How Hokuriku Bank Built an AI Internal Helpdesk to Break Knowledge Silos: Hokuriku Bank, Ltd. Source: https://en.helpfeel.com/case-studies/1-800-articles-serving-150-branches-how-hokuriku-bank-built-an-ai-internal # 1,800 Articles Serving 150 Branches: How Hokuriku Bank Built an AI Internal Helpdesk to Break Knowledge Silos - 1,800 articles published to support 150 branches - 30+ minute inquiry response times eliminated through self-service - Full visibility into inquiry patterns driving continuous article prioritization - Zero dependency on specific knowledge experts for routine questions ## The Challenge Hokuriku Bank, a regional financial institution headquartered in Toyama, supports community development across three prefectures plus Hokkaido and the major metropolitan areas. With approximately 150 branches generating a constant flow of inquiries to head office, the bank faced a compounding operational burden. Manuals and reference documents were scattered across different locations, making information difficult to find. Staff relied on a mix of the intranet, an electronic bulletin board, phone calls, and internal "circulation reports" (similar to email). Phone calls and one-to-one messages meant the same questions kept surfacing repeatedly, with no shared knowledge trail. Responses taking 30 minutes or more were routine, and even with two dedicated staff members triaging inquiries, the General Administration Division couldn't keep up. Specialized knowledge concentrated in specific individuals created fragile, person-dependent operations. When staff rotated or took leave, institutional know-how disappeared. For a financial institution where compliance and accuracy are non-negotiable, the bank needed an environment where even newer employees could independently access and learn operational knowledge without compromising service quality. ![A Hokuriku Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/hokuriku-helpdesk_body.jpg) ## Why Helpfeel When evaluating internal helpdesk solutions, Hokuriku Bank prioritized accurate communication of information above all else. In a regulated environment where guidance must align with laws and compliance standards, clarity and precision were essential. The bank chose Helpfeel for three reasons: an intuitive UI that staff could understand immediately, the ability to deliver clear and accurate answers, and the strength of the support structure. The Customer Success team became a critical factor, offering not just technology but operational know-how tailored to the bank's specific situation. ## What They Did - Used Helpfeel's generative AI draft creation feature to convert existing electronic bulletin board content into articles, making a launch with almost no pre-existing content feasible. - Categorized the bank's large volume of internal information into several priority areas, building articles in stages with guidance from the Customer Success team. - Rewrote bulletin board posts (originally one-to-one answers to specific inquiries) into generalized helpdesk articles useful across a wide range of scenarios, spending three months refining content before launch. - Promoted adoption through monthly circulars notifying staff of updated articles, management-level training sessions, and branch manager advocacy to drive habit change among busy frontline teams. - Leveraged Helpfeel's feedback feature and search history analytics to identify high-demand topics and prioritize new article creation based on real branch needs. ## Results With 1,800 articles now published, the internal helpdesk has fundamentally changed how Hokuriku Bank's 150 branches access operational knowledge. The 30+ minute inquiry response cycle has been replaced by immediate self-service search. Repeat inquiries have been eliminated because every answer is now visible to all staff, not locked in one-to-one communication channels. The person-dependent bottleneck is gone; routine questions no longer concentrate on specific individuals. The bank gained full visibility into inquiry patterns for the first time. By analyzing phone volume and search data, the General Administration Division now understands which topics generate the most questions and can prioritize articles accordingly. Staff are submitting requests through the feedback feature, including branch managers and external sales staff who need to resolve uncertainties quickly during customer meetings. Helpfeel's intent-prediction search proved essential. With 1,800 articles, having the right article "findable" is what keeps staff coming back. Just as important: the system makes it clear when information isn't available, preventing wasted time on futile keyword searches. ## Looking Ahead The bank will continue enriching helpdesk content to make "check Helpfeel and your question will be answered" an organic part of conversation in every branch. By reducing the inquiry burden on the General Administration Division, the team can dedicate more time to initiatives that improve administrative efficiency and operational soundness. The longer-term goal is to rationalize internal inquiry channels entirely. Right now, staff navigate a complex mix of the internal helpdesk, phone, the electronic bulletin board, and circulation reports, creating uncertainty about which channel to use. The target state: minor questions resolved instantly via the internal helpdesk, with dedicated channels reserved for individual or complex cases requiring proper consultation. --- ## 10,000 Inquiry Records Over 20 Years Transformed into Working Knowledge: KEM Removes the Burden from Veteran Engineers: Kyoto Electronics Manufacturing Inc. Source: https://en.helpfeel.com/case-studies/10-000-inquiry-records-over-20-years-transformed-into-working-knowledge-kem # 10,000 Inquiry Records Over 20 Years Transformed into Working Knowledge: KEM Removes the Burden from Veteran Engineers - 10,000 inquiry records across 20 years transformed into a working knowledge base - Inquiry volume grew 30% to 1,250/year before implementation; response hours exceeded 2,000 hours annually - 200 FAQ answers deployed, organized across six product categories - Built from real inquiry logs, not manuals, targeting the exact questions users actually ask ## The Challenge Kyoto Electronics Manufacturing (KEM) is a specialist manufacturer of analytical instruments, providing proprietary products that integrate cutting-edge technology, including moisture analyzers and other measurement instruments that support industry across multiple sectors. As inquiry volume grew nearly 30% in recent years, reaching approximately 1,250 inquiries annually, the Technology Development Division faced an escalating burden. Total response time over three years more than doubled to approximately 2,000 hours. The nature of KEM's products made the problem worse. Highly specialized and diverse, they serve customers across industries from food manufacturing to pharmaceuticals. Even for the same product, measurement targets differ by user, so technical inquiries from the same customer often recur. Customers spanning this wide range of use cases and contexts required nuanced, context-specific answers. The team facing the greatest burden was not junior staff, but experienced veterans. The more knowledgeable a person, the more likely they were to be pulled into inquiry responses, repeatedly interrupting development work in progress. When a question required investigation and verification rather than an instant answer, the team had no choice but to stop development work and redirect to inquiry handling. Since multiple staff collaborate on development projects, halting and restarting required coordination. With nearly 90 staff in the Technology Development Division, several simultaneous inquiry-driven work stoppages could affect the efficiency of the entire department. Compounding the challenge, expert managers with deep knowledge in specific areas personally handled certain inquiries to ensure consistent answers on behalf of the company. Because managerial staff time was not included in the calculated response hours, the actual burden was likely even higher than the figures showed. Reducing this invisible burden was a core motivation for the initiative. ![Kyoto Electronics staff with measurement instruments](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/kem_body.jpg) ## Why Helpfeel The team's search for a solution centered on AI. AI chat tools that generate response text were considered, but many of these tools have hallucination risks. Since the AI generates the answer itself, accuracy is hard to guarantee, and humans still have to verify every response, meaning the workload doesn't actually change. Helpfeel uses AI for its intent-prediction search, expanding the language understanding of search queries, but the actual response content is not generated by AI. This eliminates the risk of AI-generated misinformation and ensures accuracy. The availability of post-implementation operational support and trend analysis was also appealing. The team considered building FAQs from product instruction manuals. Many inquiries ultimately lead to the same answer that could be found in a manual, basic operating procedures and so on. But in practice, users' measurement purposes, sample types, and operating environments vary greatly. Even if the manufacturer publishes correct, general guidance, a user facing a specific problem will hesitate ("maybe my case is special") and want to confirm by phone or email unless they find an answer that matches their exact situation. Building from product manuals would cover most cases, but would never hit 100%. The key insight came from learning that if actual inquiry records existed, Helpfeel Analytics could analyze inquiry patterns to identify high-priority topics and generate FAQ article drafts based on historical responses. Every product has its own characteristic questions, so by analyzing the most frequently asked topics from the accumulated data and prioritizing articles accordingly, the team could surgically target the specific inquiries they most wanted to reduce. When Technology Development Division staff tested the FAQ, deliberately trying tricky questions across a wide variety of patterns, it consistently returned accurate, appropriate answers. This gave the team frontline confidence that "this will work." The calculation that the cost savings from reduced inquiry handling time exceeded implementation costs also supported the final decision to proceed. ## What They Did - Hosted the FAQ on the company's corporate website "Frequently Asked Questions" page using Helpfeel's interface, organized into six product-category sections with approximately 200 total answers - Used Helpfeel Analytics to cluster-analyze approximately 10,000 inquiry response records accumulated over 20 years, identifying frequently occurring topics and generating FAQ article drafts - Focused on the most recent 10 years of inquiry data to ensure consistency with current product specifications - Built answers that include examples of commonly encountered usage conditions, with key points reinforced by repeating them with varied phrasing to ensure users can resolve questions with confidence - Prioritized articles based on the most frequently occurring topics per product category, making inquiry trends that each staff member had only an intuitive sense of quantitatively visible ## Results The FAQ went live in October 2025 with 200 answers built from 10,000 inquiry records spanning 20 years. After the launch announcement was posted on the company's groupware platform, it immediately received a large number of "likes" from across the company, not just from the Technology Development Division. Staff in all customer-facing departments had been waiting for exactly this kind of solution. Beyond end users looking up answers directly, customer support staff consult the FAQ when formulating responses, and sales staff reference it in real time during customer visits. The team plans to continue refining content and entry points into the FAQ based on usage patterns. The concrete effects, how much the FAQ will reduce development interruptions from inquiry handling and whether faster self-resolution improves customer satisfaction, will be verified going forward. ## Looking Ahead Implementing Helpfeel and establishing a systematic approach to handling high-frequency inquiries marks the starting line for the company's goal of "DX that converts tacit knowledge into explicit knowledge." The team continues to refine content based on usage patterns. The company's administration division is also showing enthusiasm about applying a similar system for employee inquiries, HR policy questions and similar internal queries that currently come in by phone. From a data utilization perspective, internal systems contain coordination records and department-specific documentation that haven't yet been fully leveraged. Building on the success of this initiative with Helpfeel Analytics, the company plans to continue expanding DX efforts, including unearthing those hidden "treasures" of accumulated organizational data. --- ## 10% Call Answer Rate Increase at Hokuyo Bank: Service Expansion Without Adding Headcount: Hokuyo Bank, Ltd. Source: https://en.helpfeel.com/case-studies/10-call-answer-rate-increase-at-hokuyo-bank-service-expansion-without-adding # 10% Call Answer Rate Increase at Hokuyo Bank: Service Expansion Without Adding Headcount - Call answer rate improved 10% to over 90% - Service expansion managed without increasing labor costs - Search logs now drive product and service improvements ## The Challenge Hokuyo Bank, a financial institution headquartered in Sapporo with approximately 170 branches across Hokkaido, had long prioritized phone support to keep customers from waiting. The strategy created heavy dependence on the call center, and staffing became a persistent problem. Regular retirements, personnel transfers, and Japan's worker dispatch law limiting contract staff to three years in the same department meant frequent turnover. Training new operators consumed supervisor time and reduced call-handling capacity, creating a vicious cycle. The in-house team of five operators was fielding over 2,000 inquiries per month, well beyond sustainable capacity. The bank's target answer rate of 90% routinely fell below 80%, leaving customers frustrated and unable to reach support. The existing FAQ lacked search functionality and sat largely unused, so customers called instead. Without regular content updates, the FAQ couldn't keep pace with service changes or customer needs. The Digital Marketing Division, responsible for internet banking and the mobile app, needed a solution that a small team with other responsibilities could actually sustain. ![Hokuyo Bank](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/hokuyo-bank_body.jpg) ## Why Helpfeel The bank explored multiple customer support solutions, including chatbots, but three factors led them to Helpfeel. First, the management interface was intuitive. When Deputy Division Manager Nishimoto tested the demo environment, he was confident his team could get operations running smoothly. Second, and most decisive, was operational sustainability. Only three people in the division would manage the FAQ, all with other primary responsibilities. Implementing a tool that would sit unused due to operational neglect would be pointless. With limited time available for FAQ management, regular Customer Success support was essential. Third, the bank learned that a partner bank in their alliance had already adopted Helpfeel successfully. The proof point from a peer institution gave the team confidence that the solution would deliver in a similar banking context. ## What They Did - Reviewed and cleaned existing FAQ articles, removing outdated content before migrating to Helpfeel in November 2023 - Worked with Helpfeel's Customer Success team to set operational KPIs: tracking FAQ access volume to measure self-service growth and monitoring the no-hit rate to ensure information reliably reached customers - Established a monthly review cycle with the Customer Success team, using data-centered meetings as a pacemaker for improvement work - Added more entry points to the FAQ when access growth stalled, then watched access numbers climb again - Pinned high-volume topics (address changes, notification phone number changes, transfer limit inquiries) to the "Frequently Asked Questions" section for greater visibility - Used search and inquiry data to identify service and app development improvements, leading to system revamps for address changes and transfer limits ## Results Call volume decreased even as the number of customers using the bank's app and digital services grew. The phone inquiry answer rate improved to over 90%, and the team handled customer inquiries without backlog and without increasing call center headcount. FAQ access volume trended upward as customer awareness expanded. Monthly data analysis revealed which keywords repeatedly appeared at the top of searches, allowing the team to identify customer demand patterns and high-interest products and services. What they had previously sensed intuitively was now confirmed and visualized in data, giving them confidence in deciding next actions. The data also surfaced specific friction points. When the app launched, searches and call center inquiries surged around notification phone numbers. Many customers had registered landline numbers but needed to switch to mobile numbers to receive authentication codes. By identifying this pattern early, the team could address it through both FAQ content and system improvements. Helpfeel's insights increasingly drive product and service development. The bank has been strengthening app-related FAQ articles to keep pace with strategic investments, and expanding the FAQ to cover information managed by other departments. Senior leadership gave positive feedback: "The PDCA cycle is running well. Let's keep expanding the FAQ." ## Looking Ahead The bank plans to continue improving the FAQ based on data analysis, expanding the article library, and increasingly connecting FAQ insights to product and service improvements. The team aims to gain visibility into inquiry patterns from call centers outside the Digital Marketing Division and use that data to further enrich the FAQ, reducing inquiry volume across the entire bank. The long-term vision is an environment where customers can naturally self-resolve their questions to such a degree that displaying a phone number on the website is no longer necessary. The team also plans to expand corporate-facing FAQ articles so that more customer segments can benefit. By incorporating AI tools like Helpfeel into daily work, the bank is raising the quality of customer support while improving operational efficiency and continuing to evolve to meet customer needs. --- ## 10% Higher Article Reach at 16 Million Accounts: Rakuten Bank's Online-First Support Strategy: Rakuten Bank, Ltd. Source: https://en.helpfeel.com/case-studies/10-higher-article-reach-at-16-million-accounts-rakuten-banks-online-first # 10% Higher Article Reach at 16 Million Accounts: Rakuten Bank's Online-First Support Strategy - Article reach rate up 10%+ - No-hit searches reduced to ~3% - 16 million accounts served with online-first self-service - Data-driven operations model with monthly CS-led reviews ## The Challenge Rakuten Bank's Content Planning Division manages every customer touchpoint for over 16 million accounts, yet they couldn't tell whether customers were actually resolving their questions in the FAQ. Inquiry content from phone, chat, and email centered on basic internet banking procedures like opening an account. As an internet bank, customers expect to resolve questions immediately, right where they are. The team repeatedly added FAQ articles and registered expected keywords, but whether these efforts were solving the root problem remained unclear. The real concern: customers who fail to find answers and don't bother contacting support simply risk disengaging from the bank altogether. The FAQ was clearly critical to customer satisfaction, but incremental fixes didn't feel like meaningful progress. ![A Rakuten Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/rakuten-bank_body.jpg) ## Why Helpfeel The team learned about Helpfeel through a referral. The intent-prediction search feature, which handles variations in how customers type the same keyword, seemed like it could address the visibility gap they faced. Equally important: they hadn't been running data-driven improvements to the FAQ system. Introducing Helpfeel with Customer Success support would help them build a better operational structure. They launched a proof of concept with clear KPIs: no-hit rate, article reach rate, search utilization rate, bounce rate, and re-search count. ## What They Did - Migrated existing FAQ articles directly into Helpfeel and launched the PoC, measuring article reach and no-hit rate from day one - After PoC KPIs cleared (article reach rate improved by 10%+, no-hit rate dropped to ~3%), moved to full deployment - Established monthly review meetings with Helpfeel Customer Success, Content Planning, and call center teams to analyze data and plan initiatives - Built an internal FAQ operations guideline based on insights from iterative article improvements, with CS team input - Enabled the entire team to operate the system through detailed onboarding, eliminating reliance on specific individuals ## Results Metrics including article reach rate and no-hit rate have stayed consistently strong across the board. The bank genuinely feels that a significant number of customers are now resolving their questions through Helpfeel. Article reach rate improved by 10%+, and the no-hit rate reduced to approximately 3%. The Customer Success team flags unusual patterns, such as a sudden spike in searches for "NISA" when the new NISA scheme received major media coverage. The team investigated, identified keyword gaps, and added articles to address growing demand before inquiries piled up. Beyond the metrics, Helpfeel solved the knowledge-lock problem. Previously, only specific individuals could update content or configure the admin panel. Now everyone involved can operate the system, and the monthly review process gives the team a quantifiable sense that they're enhancing the customer experience. ## Looking Ahead The FAQ is one of the division's most important customer touchpoints. The team will continue improving individual articles and overall readability so customers can smoothly resolve their questions. They also want the FAQ to handle time-sensitive information needs, like emergency maintenance notices, so customers searching during incidents can find answers immediately. The ideal is a seamlessly designed service where customers never need to contact support with a question. Customer situations will always vary, so questions will arise, but the team will keep driving improvements with the goal of providing an ever-better FAQ experience. --- ## 10x Growth in Product Page Traffic: How 2nd Street Connected Customer Support to Sales: GEO Holdings Inc. (2nd Street) Source: https://en.helpfeel.com/case-studies/10x-growth-in-product-page-traffic-how-2nd-street-connected-customer-support-to # 10x Growth in Product Page Traffic: How 2nd Street Connected Customer Support to Sales - Inquiry volume held flat despite 50+ new store openings and customer growth - Traffic to trade-in pages increased 10x after strategic FAQ link placement - FAQ articles began ranking in organic search within months of launch ## The Challenge GEO Holdings' 2nd Street is one of Japan's fastest-growing resale brands, and that growth came with a tax. As stores, customers, and revenue climbed, inquiries climbed right alongside them at almost the same rate. A lot of those questions already had answers sitting on the website. The old self-service experience just didn't help anyone find them: a static wall of text with no search, and a chatbot that pushed customers through rigid menus and dropped them at dead ends. So customers did what people do when a website fails them. They picked up the phone and called the nearest store. Floor staff ended up fielding questions that had nothing to do with the floor. The team needed two things at once: a way for customers to answer their own questions, and the operational data to keep getting better at it. ![Inside a 2nd Street resale store, racks of curated secondhand apparel](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/2ndstreet_body.jpg) ## Why Helpfeel The team evaluated several help center platforms, and search accuracy decided it. Helpfeel reads past typos and phrasing differences, so a customer reaches the right answer no matter how they word the question. That mattered, because they already knew their existing articles weren't matching the words customers actually typed. Trade-in questions were the clearest example, and trade-ins are one of the most common things customers ask about. Helpfeel's embedded form went a step further. As a customer types their question, relevant articles surface in real time, often answering them before they ever hit submit. When the project went to senior leadership, that was the capability that sold it. ## What They Did - Stood up Helpfeel with a five-person cross-functional team. Uchimura led article creation and configuration, and subject-matter experts from each relevant team verified the content - Ran monthly reviews with Helpfeel's Customer Success team to build out articles on high-frequency search topics, spot anomalies in the data, and turn what they learned into changes - Read the traffic data to find their highest-volume articles, then built navigation paths from those articles to pages that convert (for example, linking "items we can't accept" to "how to trade in items" and "what items are accepted") - Embedded Helpfeel directly into the inquiry form so answers surface as customers type ## Results A year and change after launch, the numbers tell the story. Roughly 50 new stores opened. Customer counts went up. And total inquiry volume stayed about flat. The new questions customers can resolve themselves are getting absorbed by the knowledge base instead of hitting the support queue. Within a few months, organic search traffic started arriving straight on their knowledge articles, especially for trade-in and payment topics. That's the payoff from improving articles one at a time: a customer running a plain internet search now arrives right on the answer they need. The biggest win came out of the data. The article explaining which items can't be accepted for trade-in was pulling heavy traffic. So the team added links from it to the pages explaining how to trade in items and what items are accepted. After that single change, traffic to the trade-in pages jumped 10x. A knowledge article, wired straight to revenue. ## Looking Ahead Next, the team wants to carry the same results into physical stores, where awareness of self-service is still low. They mentioned it during the website redesign, but never gave customers clear guidance on how to use it. Raising in-store awareness is the priority, and it should pull even more calls away from store staff. Further out, they want to figure out what information on conversion pages would lift trade-in conversion rates, then take what works across the wider GEO Group. --- ## 20,000 FAQ Sessions Per Month: Major Airline Avoids Inquiry Surge During Company-Wide Migration: Major Airline Group Source: https://en.helpfeel.com/case-studies/20-000-faq-sessions-per-month-major-airline-avoids-inquiry-surge-during-company # 20,000 FAQ Sessions Per Month: Major Airline Avoids Inquiry Surge During Company-Wide Migration - 20,000 FAQ sessions per month immediately following migration - Phone inquiries contained during company-wide Google Workspace rollout affecting 35,000 employees - General operational issues eliminated from phone queue ## The Challenge Before the Google Workspace migration, support was overwhelmed. The airline group's internal helpdesk fielded 50 phone calls a day with just 3 people. A FAQ existed, but it lived in a spreadsheet with basic search. Employees told the team directly they couldn't find information and it was too hard to read, so they just called instead. With a company-wide migration to Google Workspace looming, affecting all 35,000 employees across the group, the team faced a difficult reality: call volume could surge far beyond an already-challenging baseline. They needed a way to genuinely build employee capacity for self-service, not just add more support channels. ## Why Helpfeel N, who came to IT from a flight attendant background, understood the instinct to "just call" when something is unclear. Helpfeel was introduced through a business partner referral and evaluated against approximately 20 criteria, including admin usability, Customer Success support structure, and reporting. Three factors made Helpfeel stand out: **Expression variation handling**: The system matched searches whether users typed in Japanese katakana (メール) or English (mail), essential in a Japanese workplace where employees switch between scripts. **Low cognitive burden**: Employees could type approximate, intuitive keywords and reach the right answer without careful phrasing. Even vague queries led to the correct content. **Ease of administration**: The management interface was accessible even to non-technical administrators, with low operational stress. ## What They Did - **Drove awareness**: Communicated to all employees to "try the FAQ first" and placed Helpfeel prominently on the internal portal site - **Embedded FAQ links in responses**: When answering email inquiries, operators included links to relevant FAQ articles, shortening response time and ensuring consistency - **Built a custom dictionary**: Helpfeel's Customer Success team created a dictionary for company-internal terminology so employees searching with internal jargon would find the right articles - **Kept content concise**: CS guidance recommended short explanations with images and links to detailed manuals, ensuring quick answers without burdening users who needed only basics ## Results In the month immediately following the Google Workspace migration, Helpfeel recorded 20,000 sessions per month. The feared surge in telephone contacts did not materialize. Phone inquiries were contained, and contacts related to general operational issues essentially disappeared from the phone queue. The capacity freed up allowed operators to focus on genuinely complex cases requiring individualized handling. Support desk operators noted that users had developed a genuine understanding of common operations through FAQ adoption. Beyond end-user self-service, operators actively used the FAQ themselves as a response manual, providing consistent, accurate answers and eliminating variance across staff members. The monthly reporting from Helpfeel's Customer Success team proved continuously valuable. Coming from a non-technical background, N had initially assumed FAQ management would be complicated. The CS team's guidance on article language, content structure, and keyword tagging enabled continuous improvement that would not have been achievable independently. ## Looking Ahead Helpfeel was deployed specifically to support the Google Workspace migration transition. Following the completion of the migration, the dedicated support desk wound down and Helpfeel operation ended in March 2024. The case demonstrates Helpfeel's applicability not only as an ongoing customer-facing FAQ, but as a time-bounded internal tool deployed to manage high-stakes system transitions. --- ## 20% Fewer Calls as Membership Grows: How a Major Credit Card Company Scaled Support Through Self-Service: Major Credit Card Company Source: https://en.helpfeel.com/case-studies/20-fewer-calls-as-membership-grows-how-a-major-credit-card-company-scaled # 20% Fewer Calls as Membership Grows: How a Major Credit Card Company Scaled Support Through Self-Service - Call volume reduced 20% despite membership and usage growth - FAQ page views nearly doubled after implementation - FAQ content expanded 8x through data-driven call analysis - Search data now drives marketing campaigns and website improvements ## The Challenge As cashless payments expanded and contactless transactions surged during the pandemic, credit card adoption accelerated. A growing user base meant a proportional rise in customer inquiries, and the contact center bore the strain. The company had built its reputation on quality phone support and was unwilling to compromise that standard. But as call volume grew beyond manageable capacity, it became clear that additional channels were needed. An FAQ existed, but it had never been prioritized. The company's focus had always been phone service, so the FAQ was underutilized and sparse compared to industry peers. Worse, no clear ownership or operating rules existed. Individual teams posted articles as they saw fit, leaving the FAQ incomplete and often outdated. Customers who needed answers had no reliable way to find them on their own. ## Why Helpfeel The company approached the problem in two phases. First, they analyzed call data, categorized inquiry types, and identified what could be resolved through self-service. They expanded the FAQ to roughly 8 times its original size based on real customer voice data. Then came phase two: adopting a new FAQ system. Superior search performance was the primary functional criterion. The pricing was reasonable and made internal approval straightforward. But those were just the initial hook. What made the deeper impression was the comprehensive support Helpfeel provided from the start. The sales team, Customer Success representatives, and technical writers all contributed to optimizing the FAQ content in ways the internal team could not have achieved alone. The experience felt less like using a passive tool and more like engaging with a human-centered service, a partner rather than a vendor. ## What They Did - Analyzed call data and categorized inquiry types to identify what could be resolved through self-service - Expanded FAQ content to approximately 8x the original volume based on customer voice data - Implemented Helpfeel with guidance on making article titles clearer and more intuitive - Improved internal consistency between articles with ongoing Customer Success support - Established monthly review meetings to generate new insights and optimize content structure ## Results After Helpfeel was introduced and the content revised, FAQ page access volume nearly doubled compared to before. Despite continued growth in both membership and per-member card usage, incoming call volume to the contact center declined by approximately 20%. While multiple factors contributed, Helpfeel's role in helping customers resolve questions independently was a key driver. A significant additional outcome was the ability to surface customer needs through Helpfeel's search data. The company now uses this intelligence in several ways: identifying sudden spikes in search keywords to diagnose communication problems and create targeted FAQ articles, using search trend data to inform website improvements and direct mail campaigns, and prioritizing which new FAQ content to create based on what customers are actually looking for. Helpfeel evolved beyond being purely an FAQ tool. It now functions as a marketing intelligence resource, surfacing hidden customer needs that would otherwise go undetected. ## Looking Ahead Introducing Helpfeel enabled the company to use the FAQ as a lens for identifying communication gaps with customers, the outcome they describe as most significant. The tool exceeded the scope of an FAQ system to serve as a marketing resource as well. Going forward, the company intends to use Helpfeel not only to continue reducing call volume, but to invest more deeply in improving overall communication design and service design, using the insights the platform surfaces as a foundation for broader improvement. --- ## 20% Fewer Inquiries Despite Growth: How a 200,000-Policy Insurer Scaled Support Without Adding Seats: Pet & Family Non-Life Insurance Inc. Source: https://en.helpfeel.com/case-studies/20-fewer-inquiries-despite-growth-how-a-200-000-policy-insurer-scaled-support # 20% Fewer Inquiries Despite Growth: How a 200,000-Policy Insurer Scaled Support Without Adding Seats - Inquiry volume reduced 20% despite growing policyholder base - Inquiry rate improved from 4.5% to approaching 3% range - Call center costs reduced ~10% (2 seats) - FAQ utilization ratio trending upward, accelerating adoption ## The Challenge Pet & Family Non-Life Insurance Inc., a specialist pet insurer in the T&D Insurance Group, had surpassed 200,000 policies in force with premium income exceeding 10 billion yen. As their policyholder base grew, so did inquiry volume. Managing phone and email demand, and the appropriate number of communicator seats at their outsourced call center, was an ongoing challenge. The company had launched a customer portal ("My Page") in 2020 as part of a broader digital transformation strategy, migrating paper-based procedures like address changes to the web. Approximately 60% of policyholders had registered, with more than half using web-based insurance claims submission. One in three policyholders had registered for the official LINE account. But their traditional list-based FAQ was difficult to navigate, and customers frequently reported being unable to find answers during evenings or weekends when the call center was closed. The company wanted not only to reduce phone inquiry volumes for operational efficiency, but also to improve satisfaction from the user's perspective. A conversational chatbot had been considered first, but the burden of analyzing past inquiries and building response scenarios was heavy, and questions outside those scenarios still ultimately required a phone call or live chat. The plan was put on hold. ![A Pet & Family Non-Life Insurance team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/petfamilyins_body.jpg) ## Why Helpfeel Helpfeel was introduced by the parent company, T&D Holdings, as a tool with proven adoption across the group. The fact that it was created by the developer of the Japanese flick input system for iPhone (now a Technical Fellow at Helpfeel) suggested strong usability credentials and contributed to the decision. What struck the team as Helpfeel's real strength was the ability for a single article to cover a wide variety of questions through keyword associations. For example, regardless of whether a customer types "insurance premium," "premium payment," "contribution," "bank account," or "credit card" into the search box, the top result displayed is the frequently asked question "I want to change my insurance payment bank account or credit card," leading to one shared explanation article. Additionally, Helpfeel's intent-prediction search had been upgraded to handle full sentences (not just keywords) through a generative AI-powered enhancement, making the FAQ more accessible to senior users in particular, who tend to type full sentences into search boxes. ## What They Did - Migrated content from the old FAQ system and launched Helpfeel in March 2023, adding two to three articles per month while consolidating duplicate content to reach approximately 120 published articles. - Assigned FAQ operations to a cross-departmental team of four (one representative from each department within the Customer Service Division), gathering ideas and inputs, then reviewing and updating content following monthly meetings with the Helpfeel team. - Used tactics such as linking to related topics within articles, tagging content for discoverability, and promoting seasonally relevant topics like "flea/tick prevention" to the top of results to steadily increase the rate at which customers find what they're looking for and self-resolve. - Promoted FAQ usage actively via the official LINE account, with a FAQ link permanently displayed in the rich menu at the bottom of the LINE screen and FAQ usage promoted visually through content pieces mixed in with pet-related lifestyle articles. - Implemented creative engagement initiatives, including a limited-time "hidden feature" where typing "summer vacation" into the search box triggered a full-screen fireworks animation, combined with seasonal messaging including a summer greeting and awareness reminders about pet risks such as heatstroke and accidental ingestion. ## Results Despite steady growth in the number of policyholders, inquiry volume has been reduced by approximately 20% compared to before Helpfeel was introduced. The inquiry rate improved from the mid-4% range at implementation to the high 3% range, and the low 3% range is now in sight. The FAQ utilization ratio is also trending upward, with FAQ usage accelerating. With Helpfeel usage firmly established, call center phone lines, despite having fewer seats, have actually become more accessible to callers. The maximum seat reduction achieved was approximately 10%, or 2 seats. The way the call center handles inquiries has also evolved. For example, in the case of My Page-related questions, communicators previously tended to walk customers through every step of the procedure. Now they can first direct customers to search the FAQ and guide them with "please follow the steps shown there." Customers are also increasingly calling in after already having searched Helpfeel, arriving at the call with more focused and specific questions. This combined effect has shortened average resolution time per case and enabled a higher volume of calls to be handled overall. During the year-end period when the call center is closed, a targeted FAQ usage appeal was sent via LINE, generating more traffic through that channel than through the website banner, confirming the LINE channel's effectiveness as a promotional vehicle. Since introducing Helpfeel two years ago, the company has experienced multiple events that would normally trigger a significant spike in inquiries, including the launch of web-based insurance claims and product revisions. Throughout all of this, by continuously enriching the FAQ, the call center has consistently maintained a high response rate without significantly increasing seat count. ## Looking Ahead The news that the voice input feature is now supported by the full-sentence intent-prediction search upgrade was very welcome. Customers who aren't experienced searchers can now find what they need in a way that feels close to natural conversation, a further usability improvement. The company plans to promote this capability through LINE and other channels. The team also wants to invest further in refining the FAQ through closer collaboration with the call center. For example, for complex escalated inquiries that get transferred from a communicator to headquarters, currently leaving customers on hold, they'd like to explore whether Helpfeel could be incorporated to enable faster resolution. They want to continue growing Helpfeel not just as an information delivery platform, but as one of their most important customer touchpoints. --- ## 20% More Customers Finding Answers: How a Regional Bank Put AI Search at the Center of Digital Service: Iyo Bank, Ltd. Source: https://en.helpfeel.com/case-studies/20-more-customers-finding-answers-how-a-regional-bank-put-ai-search-at-the # 20% More Customers Finding Answers: How a Regional Bank Put AI Search at the Center of Digital Service - Page reach rate improved 20 percentage points (50% to 70%) - 15 hours per month of operational time saved - FAQ maintenance staff reduced from 3 to 2 - FAQ search hit rate reached ~95% ## The Challenge Iyo Bank, headquartered in Matsuyama City, Ehime Prefecture, has been aggressively pursuing digitalization under the slogan "Iyo Bank is becoming a D-H-D Bank (Digital, Human, Digital Bank)." With branch traffic declining industry-wide, particularly after COVID-19, expanding and enhancing non-branch, web-based customer touchpoints became a strategic priority. The bank had long relied on an FAQ system and chatbot to help customers resolve questions on their own. Both systems showed limitations in search performance, working on a one-to-one relationship between questions and answers that made it difficult to surface the right answer when customers typed in a wide variety of expressions. Meanwhile, the FAQ site remained one of the most viewed pieces of content the bank published, making it clear that customer demand was significant. Improving the FAQ site was critical to improving non-branch customer touchpoints. Operationally, maintaining approximately 800 FAQ pages created substantial workload for the staff responsible for creating, editing, and analyzing content. The system itself was cumbersome to configure and operate. Customer support from the vendor was limited to email-only inquiries, with responses coming back in English. ## Why Helpfeel Ryuhei Takahashi, Section Manager in the Business Marketing Division, had been continuously researching FAQ systems but couldn't find anything that gave him a genuine "that's the one!" reaction. Everything felt like more of the same. When a work acquaintance told him about "a really interesting product," Takahashi looked at the proposal materials. His reaction: pure surprise. "This is incredible." The approach was completely different from conventional chatbots and FAQ systems that link a question to a single answer. The idea of first inferring the user's "question" from a search keyword felt genuinely customer-centric and original. Concretely, Helpfeel displays predicted question candidates based on the search keyword. For example, typing "seal" (hanko) surfaces the question "I've lost my registered seal, who should I contact?" while the user is still typing. The mechanism of predicting the "question" in the user's mind and surfacing search results accordingly struck Takahashi as truly innovative. He decided to move forward immediately. ## What They Did - Migrated approximately 800 FAQ pages from the legacy system to Helpfeel in about one month - Reviewed and revised existing FAQ pages during the migration based on concrete improvement suggestions from Helpfeel's Customer Success team - Established monthly online review meetings with Customer Success to analyze usage data and guide FAQ content additions and revisions - Reallocated resources freed up by reduced maintenance workload to other tasks ## Results Iyo Bank launched a new service at approximately the same time as the Helpfeel rollout and anticipated a temporary surge in inquiries. In reality, they were able to suppress that increase. The team believes this was the direct result of Helpfeel enabling customers to resolve their own questions more effectively. Looking at actual usage data, the FAQ search hit rate (the percentage of searches that return at least one result) reached approximately 95%, meaning virtually any combination of words a customer uses will return search results. This materially increased the reliability of the FAQ. Under the previous FAQ system, the page reach rate (the percentage of users who reach an actual FAQ page after searching) was approximately 50%, meaning many customers who searched were unable to find an answer and left the FAQ. After implementing Helpfeel, the page reach rate improved to approximately 70%, an increase of around 20 percentage points. In other words, significantly more customers are now discovering the FAQ page that resolves their question through keyword search. Thanks to ongoing support from Helpfeel's Customer Success team, the bank dramatically reduced the burden of FAQ site maintenance, cutting approximately 15 hours per month of operational work. As a result, they were able to reallocate resources to other tasks and reduce the number of dedicated FAQ maintenance staff from 3 to 2. The monthly review meetings built a strong internal reputation. "The proposals are packed with know-how and genuinely educational!" one team member said. One staffer who was reassigned to a different department due to an organizational restructuring even requested to keep attending the Helpfeel monthly review meetings after moving on. ## Looking Ahead Iyo Bank will continue advancing the DX of its services and operations in line with the D-H-D Bank model, with the goal of delivering ever-greater value to customers. The bank sees Helpfeel as a solution that plays a truly central role in the DX of non-branch customer touchpoints in particular, and plans to continue leveraging Helpfeel fully, with an eye on a wide range of potential applications going forward. --- ## 25% Fewer Inquiries in One Month: How Persol Business Process Design Scaled Internal Support for 4,600 Employees: Persol Business Process Design Inc. Source: https://en.helpfeel.com/case-studies/25-fewer-inquiries-in-one-month-how-persol-business-process-design-scaled # 25% Fewer Inquiries in One Month: How Persol Business Process Design Scaled Internal Support for 4,600 Employees - Inquiries cut 25% within one month - FAQ usage rate increased ~10% within two months - 3-person Concierge Desk team serving ~4,600 employees - 95% first-response-within-15-minutes SLA maintained ## The Challenge Persol Business Process Design's internal helpdesk, the Concierge Desk, serves approximately 4,600 employees across a diverse set of needs: IT systems, HR, accounting, and general operations. The three-person team had successfully built a strong self-service culture, about 60% of employees turned to the FAQ when they encountered a problem. But the existing FAQ system relied on exact-match keyword search, and that foundation was cracking. Employees who didn't type the precise right keyword either got zero results or were buried in irrelevant articles. The team tried embedding multiple search terms in each FAQ to compensate, but that just made the noise worse. Internal surveys confirmed the frustration: people wanted to help themselves, but the tool kept getting in the way. The opportunity was clear, if search worked better, the already-strong culture of self-service could become a real competitive advantage. ![Inside the Persol Business Process Design office](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/persol-pt_body.jpg) ## Why Helpfeel Kenta Ito, who leads the Concierge Desk, was drawn first to Helpfeel's stated 98% search hit rate. Coming from a system that routinely returned zero results, the promise that employees could phrase questions naturally, colloquially, with typos, or intuitively, and still reliably reach the right answer, stood out immediately. When he tested Helpfeel on a live customer site, it performed exactly as advertised: variations in phrasing, expression, and even mistakes all returned the correct article. The conclusion was straightforward: with Helpfeel, the FAQ could become a genuinely reliable first stop for employees, freeing the small Concierge Desk team to focus on the inquiries that needed human judgment, and helping them consistently hit their 15-minute first-response SLA. Ito also appreciated the clean, minimal user interface, something he described as a feature he "loved." ## What They Did - **Parallel launch with promotion**: Helpfeel was introduced in June 2021 alongside the existing FAQ system, with approximately 500 internal articles published. The team promoted the new FAQ at a company-wide all-hands meeting, explicitly encouraging employees to use it. - **Streamlined article maintenance**: With Helpfeel's search working reliably, the old practice of stuffing articles with diverse keywords to compensate for poor search was abandoned. Articles became cleaner, more readable, and written for humans rather than for gaming a search engine. - **Focus on usage analytics**: The team planned monthly review meetings to systematically examine search keyword data, article view patterns, and re-search rates, translating insights into ongoing FAQ improvements. ## Results Within one month of Helpfeel's introduction, human-handled inquiries to the Concierge Desk fell approximately 25% compared to the pre-launch baseline. Both recurring seasonal questions, like health check inquiries, and general IT system questions decreased noticeably. Within two months, the proportion of employees actively using the FAQ to search for information increased by approximately 10%, a direct reflection of the improved, more reliable search experience driving greater engagement. The improvement was visible enough that other departments began approaching the Concierge Desk to request Helpfeel adoption in their own areas. Employees even proposed offering it as a recommendation to clients. The quality of the experience had generated genuine internal advocacy. With keyword-stuffing no longer required, FAQ articles became cleaner and more useful, written from the reader's perspective rather than the search engine's perspective. ## Looking Ahead Ito and his team plan to place greater emphasis on usage analytics, particularly search keyword data, article view patterns, and re-search rates, to identify where the FAQ can be further improved. Monthly review meetings are planned to systematically translate analysis into concrete improvement actions. The broader goal is to use Helpfeel to deepen the culture of self-service across all approximately 4,600 staff, establishing a workplace norm where looking up the answer yourself is the default first response. The Concierge Desk's role as a critical enabler of company-wide productivity is clear, and Helpfeel is the tool that makes the team's work both scalable and high-quality. --- ## 25% More Traffic with Job Seeker FAQs: How OpenUp IT Engineer Built a Data-Driven Recruitment Strategy: OpenUp IT Engineer Inc. Source: https://en.helpfeel.com/case-studies/25-more-traffic-with-job-seeker-faqs-how-openup-it-engineer-built-a-data-driven # 25% More Traffic with Job Seeker FAQs: How OpenUp IT Engineer Built a Data-Driven Recruitment Strategy - Site traffic increased 25% following Helpfeel introduction - FAQ pages now drive 20% of all recruitment site visits - Candidate pool expanded 25% through organic search discovery - Casual interview applications trended upward since launch ## The Challenge OpenUp IT Engineer trains individuals with no prior IT experience and matches them with client companies as skilled IT engineers. But recruiting candidates for this pipeline had become expensive. The company relied heavily on external recruitment agencies to source applicants, and with Japan's IT talent shortage projected to reach 790,000 workers by 2030, per-hire costs through those agencies kept climbing. The team built a dedicated recruitment site and started publishing content to attract candidates directly. But application rates for casual pre-screening interviews stayed stubbornly low. The root cause: limited direct contact with job seekers meant content was built on assumptions rather than real insight. Takahiro Ota, Section Manager of Media Produce, suspected candidates had unresolved questions or anxieties holding them back from applying, but the team had no way to see what those concerns actually were. ![OpenUp IT Engineer](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/openup_body.jpg) ## Why Helpfeel Mr. Ota learned about Helpfeel through a former colleague. The feature that stood out immediately was the free-keyword search interface and the ability to collect the actual search terms job seekers entered. This was the mechanism to surface the raw voice of candidates: a direct window into what they were actually uncertain about, without needing to go through a recruitment agency intermediary. For a team with limited direct contact with job seekers, this data collection capability was uniquely valuable. It was as important a reason for adopting Helpfeel as the FAQ function itself. ## What They Did - **Launched a two-person operation with dedicated Customer Success support.** Regular Helpfeel analytics reports and operational recommendations made two-person FAQ management sustainable. The core operating loop: review analytics, add or edit FAQ articles based on findings, monitor impact on casual interview applications. - **Uncovered an unexpected SEO benefit.** Approximately one year after launch, the FAQ pages began ranking in search engines, driving job seekers directly to the FAQ without any prior awareness of OpenUp IT Engineer. The FAQ became a top-of-funnel acquisition tool, not just a conversion tool for existing visitors. - **Built data-driven content around real anxieties.** Search keyword data revealed "standby" (the gap period between project assignments) was being searched frequently. At OpenUp IT Engineer, employees on standby continue to receive a salary, but whether such protection exists varies between staffing companies, making it a major concern. The team created a dedicated FAQ article explaining the policy. It attracted strong organic traffic, confirming the hypothesis. Prior to Helpfeel, this concern would never have appeared in internally generated content ideas. - **Split FAQ articles based on Customer Success recommendations.** When search data showed a spike in "hair color" queries, the CS team proposed splitting one article into two: one covering rules during the selection and training period (where freedom is permitted) and one covering rules when deployed to a client site (where client-appropriate presentation is required). The rationale: give candidates this information before they even come in for an interview, pre-resolving anxiety and improving participation rates. ## Results Overall recruitment site traffic increased approximately 25% compared to before Helpfeel introduction. Roughly 20% of all recruitment site visits now arrive via FAQ pages, representing a substantial share of the total audience coming through a channel that didn't exist before. Because one-quarter of overall site traffic is attributable to the FAQ's SEO effect, the addressable candidate pool has effectively grown by approximately 25%. Given the cost of acquiring candidates through external agencies, Mr. Ota described this expansion as highly cost-efficient. Casual interview application numbers have trended upward since Helpfeel's introduction, consistent with the original hypothesis that pre-resolving candidate anxieties reduces the barrier to applying. The team now has a reliable mechanism for collecting and analyzing real, unsolicited candidate questions, enabling genuinely demand-driven content creation rather than assumption-based content creation. Mr. Ota noted that the gap between what he had assumed candidates cared about and what the data showed they actually cared about was significant. Helpfeel closed that gap. ## Looking Ahead OpenUp IT Engineer defines its business as "engineer career support," not simply staffing or contracting, but meaningfully supporting the career formation of IT engineers as a mission. Mr. Ota views Helpfeel's ability to listen to users and rapidly reflect their real needs in content as a strong fit with this mission. The team intends to extend Helpfeel's use beyond the recruitment site to support engineers throughout their careers in additional contexts. As SEO and AIO (AI Optimization, the practice of optimizing content for AI-powered search and discovery) become increasingly important acquisition channels, the ability to continuously collect and analyze real user search behavior will grow in strategic value. --- ## 30% Fewer Inquiries at Peak: How Lawson Ticket Scaled Support for 500,000 Customers a Year: Lawson Entertainment Inc. Source: https://en.helpfeel.com/case-studies/30-fewer-inquiries-at-peak-how-lawson-ticket-scaled-support-for-500-000 # 30% Fewer Inquiries at Peak: How Lawson Ticket Scaled Support for 500,000 Customers a Year - Email inquiries reduced 9,000 per month year-over-year - Peak monthly contacts dropped from 30,000 to 22,000 - Response times improved to same-day for afternoon event-day inquiries - 340 live FAQ articles covering music, sports, theater, and travel tickets ## The Challenge Lawson Entertainment's Customer Service Department was running a marathon at 100-meter sprint pace every day. The team of roughly 30 people faced 500,000 annual inquiries, including 400,000 emails and 40,000 phone calls, fueled by a post-pandemic entertainment boom driving 105, 110% year-over-year ticket sales growth. Response times stretched to one and a half months during peak periods. The backlog was so relentless that staff placed bowls of salt at the office entrance, a traditional Japanese charm against hardship. Some team members became physically unwell from sustained pressure. The root cause wasn't effort. The legacy FAQ relied on a hierarchical category structure that forced customers to navigate through layers to find answers. Creating detailed articles made the hierarchy too deep to use; consolidating made articles too long to read. A chatbot layered on top to guide users only doubled the maintenance burden. High-volume events needed dedicated FAQ pages but took days to publish, too slow for a time-sensitive ticketing business. The conclusion was unavoidable: service quality depended on reducing absolute contact volume at the source. ![A Lawson Entertainment team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/lawson_body.jpg) ## Why Helpfeel Lawson Ticket serves a broad age range, with a large proportion of customers in their 40s, 50s, and older. The team's requirements were clear: **Simple, accessible interface for all age groups.** Helpfeel's clean, uncluttered design met the bar for intuitive use across demographics. **Instant search response.** Results appear as users type, without needing to press a button and wait. The team confirmed during trials that the experience was genuinely superior. **Direct article access.** Helpfeel eliminated the need for a separate chatbot to guide users through hierarchy. Users reach relevant articles directly from the search bar, removing the double-maintenance burden entirely. **Flexible top-page link placement.** For major events generating concentrated inquiry spikes, the ability to feature prominent links to event-specific FAQ articles on the top page was essential. **Article-count-based pricing.** Unlike page-view-based pricing, Helpfeel's structure doesn't penalize teams for adding content, which mattered for long-term sustainability. ## What They Did - **Migrated 270 existing articles** at launch, expanding to 340 live FAQ articles covering music, sports, theater, and travel (including airline tickets). - **Made FAQ a mandatory touchpoint** before customers could reach the inquiry form, with a policy of resolving in the FAQ any contacts that didn't require more than what the FAQ already stated. - **Created event-specific dedicated FAQ articles**, a new content format with production responsibilities distributed across the department. - **Used Helpfeel's dashboard to surface gaps** in real time, tracking searched keywords and no-hit cases (searches returning no results) to create actionable intelligence for rapid response. - **Monitored KPIs** including inquiry volume reduction, no-hit rate reduction, and article-level satisfaction via a "Did this article resolve your question?" button at the end of each FAQ. - **Assigned dedicated staff on high-volume event dates** (such as priority pre-sale days for popular artists) to handle concentrated inquiry spikes efficiently while using the FAQ to absorb predictable volume. ## Results Monthly email inquiries dropped by approximately 9,000 compared to the same month the prior year. The previous baseline of 30,000+ monthly emails was pushed toward the final target of 20,000 per month. In the best months, volume reached as low as 22,000, a reduction of approximately 30%. Post-launch average settled at approximately 25,000 per month. The team considers itself at the halfway point toward its final reduction target in just over six months of operation. With total inquiry volume down, response times improved dramatically. Inquiries submitted in the afternoon on the day of an event, which previously couldn't be answered before the show, are now responded to in time. Cases requiring confirmation from event organizers are increasingly resolved same-day. The department-wide commitment to a "within 24 hours" response target, supported by cooperation from the sales side of the business, has become achievable thanks to capacity freed up by reduced contact volume. The workplace atmosphere transformed. Team members who previously had no bandwidth to fix outdated FAQ content now pause, improve, and move forward. Staff have started proactively suggesting improvements to the user experience and site design. Cross-departmental collaboration on improvements has advanced, benefiting both customer satisfaction and operator wellbeing. The morning atmosphere has noticeably brightened. The bowls of salt, once a sincere plea for relief, are now a source of laughter. ## Looking Ahead The team plans to improve search coverage for artist name queries, ensuring that informal spellings, katakana versions, and common abbreviations all successfully surface the correct content. They're trialing Helpfeel Analytics to surface improvement insights from historical inquiry data and experimenting with table-of-contents structures and in-page anchor links to improve readability of long-form articles. Beyond the FAQ, the Customer Service Department is evolving into a bridge connecting customer voices to upstream service improvement. Frequently asked questions about purchasing tickets for children's smartphones aren't just FAQ material, they're product communication opportunities. The team advocates to the sales side to include clear guidance in event announcement notices, preventing the question from arising in the first place. The highest aspiration is an experience where customers don't need to contact support at all because the service itself is intuitive enough not to require it. --- ## 30% Fewer Inquiries: How LUSH Japan Built Internal Knowledge for 1,000 Staff: LUSH Japan LLC Source: https://en.helpfeel.com/case-studies/30-fewer-inquiries-how-lush-japan-built-internal-knowledge-for-1-000-staff # 30% Fewer Inquiries: How LUSH Japan Built Internal Knowledge for 1,000 Staff - 77 stores and ~1,000 shop staff using the internal FAQ - No-hit rate reduced to 8%, below the 10% target - Register inquiries cut 30% (from ~200/month to ~80/month) ## The Challenge LUSH Japan operates 77 stores and spa locations across Japan with around 1,000 staff. In May 2022, the company rolled out a new point-of-sale system called Lushpay across all stores. Inquiries to the 3-person IT service desk doubled overnight, jumping from ~150 per month to over 300 at peak. The company had been managing documentation in spreadsheets, scattered and hard to search. Store staff in the middle of serving customers often resorted to digging through email history to find old responses. With no FAQ structure in place, every question landed on the service desk, regardless of complexity. ## Why Helpfeel Multiple tools were evaluated, including chatbots. Helpfeel won on two fronts: **Low implementation burden.** The IT team wanted to extend the FAQ beyond their own department to Finance and Customer Care, which also maintained manuals in spreadsheets. Helpfeel's technical writers handle content creation based on materials provided by the client, making it feasible to roll out across multiple departments simultaneously. **Proactive, proposal-style build process.** Niiyama was struck by the Helpfeel technical writers' careful reading of the submitted Q&A scripts. They asked clarifying questions like "does 'register' mean the Lushpay system?" and offered specific content structuring suggestions, demonstrating genuine understanding of LUSH's internal context. Inconsistent internal terminology was a concern going in, but Helpfeel's approach to resolving those ambiguities through dialogue made the process far smoother than expected. ## What They Did - **Extended FAQ coverage across IT, Finance, and Customer Care** departments, consolidating knowledge previously trapped in spreadsheets and email history - **Pinned frequently asked topics** at the top of the FAQ for maximum visibility - **Handled naming variation automatically**: the register system is called "Lushpay," "ラッシュペイ" (katakana), and "レジ" (register), all three expressions now correctly route to the same FAQ content - **Leveraged ongoing Customer Success support** for periodic recommendations to remove or consolidate FAQ articles that are no longer being accessed or that duplicate other content ## Results The FAQ achieved an 8% no-hit rate, below the initial 10% target, indicating it successfully covers the majority of what staff search for. Register-related inquiries dropped from a peak of ~200 per month to ~80 per month, a reduction of approximately 30%. Store staff with iPhones can already access the FAQ on mobile. The next improvement opportunity is reducing login friction on that flow to increase mobile adoption among floor staff. Niiyama highlighted that the ongoing support has been more valuable than anticipated, particularly the external perspective prompting "spring cleaning" of the FAQ library. ## Looking Ahead The broader vision is to evolve from team-level knowledge sharing to company-wide knowledge infrastructure. The 3-person service desk has no plans to expand headcount, so improving per-person efficiency through FAQ-based self-service is a strategic priority. Current FAQ coverage spans IT, Finance, and Customer Care. Niiyama wants to extend this to HR, General Affairs, and any other departments where knowledge retention is critical. With approximately 2,000 total employees across the company, the potential scale of impact is significant. A key motivation is knowledge preservation across staff turnover. Knowledge that currently lives in individuals' memories or email histories is lost when people leave. Systematically organizing and making that knowledge searchable in Helpfeel ensures it remains accessible to future staff, becoming part of the company's institutional memory. --- ## 30% Fewer Inquiries: How S.RIDE Scaled 1.7x with Just One New Hire: S.RIDE Inc. Source: https://en.helpfeel.com/case-studies/30-fewer-inquiries-how-s-ride-scaled-1-7x-with-just-one-new-hire # 30% Fewer Inquiries: How S.RIDE Scaled 1.7x with Just One New Hire - Inquiry rate cut 30% relative to active users and orders - Self-resolvable inquiries reduced 20% of total contact volume - Downloads grew 1.7x in one year with just one new support hire (4 to 5 people) - ~200 FAQ articles cover 4,900 question variations via Intent Prediction Search ## The Challenge S.RIDE runs a taxi dispatch app backed by the Sony Group. Roughly 19,000 taxis, 10 major Japanese cities. When mobility demand snapped back after the pandemic and accelerated from March 2023, cumulative downloads grew 1.7x over the next year. That growth hit a five-person support team led by Senior Manager Masashi Adachi. The team fields inquiries from passengers and taxi company partners, answers app store reviews, and turns user feedback into product improvements. As downloads climbed, the inquiries piled up faster than the team could keep pace. Cutting raw inquiry volume was never the goal. Volume climbs with the user base. That's just math. So Adachi's team set a sharper target: lower the inquiry rate relative to active users and dispatch orders. The plan was simple. Route the questions people can answer themselves to self-service, fast, and reserve human support for what actually needs a person. Accidents. Complaints. The cases that demand real, personal attention. ![A S.RIDE team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/sride_body.jpg) ## Why Helpfeel The team was already running a different help center tool with nearly 200 published articles. When they weighed their options, including bolting a chatbot on top of what they had, the answer was to consolidate. One stronger self-service tool beat maintaining chatbot scripts alongside existing content. Less to operate. Less to break. Helpfeel won on four counts. **Content moved over without a rewrite.** Existing articles transferred and worked right away, with nothing to rebuild from scratch. **Intent Prediction Search.** This was the differentiator. Helpfeel surfaces the right answer even when a customer's wording doesn't match the article text, so a wide range of natural phrasing connects to the right content. **Speed.** The response to keyword input was fast and fluid, and the team noticed it immediately. **Inquiry form integration.** Most of S.RIDE's contacts come through the inquiry form. Helpfeel applies to that form directly, surfacing relevant articles right as someone is about to hit submit. That's self-resolution at the exact moment it counts, with no waiting on a reply. ## What They Did - **Migrated, then started iterating.** After launching in December 2023, the team moved existing articles into Helpfeel with minimal changes, then improved them using daily monitoring of user behavior and monthly reviews with the Helpfeel team. - **Killed no-hit searches.** The team tracked keywords that returned nothing and added or adjusted content to cover them. - **Tamed broad keywords.** Terms like "taxi" and "app" returned too many results to help anyone. The team used Helpfeel's word hints to guide people toward specifics. Type "taxi," and the word hint panel surfaces "dispatch surcharge" as a related term, so users home in on what they actually need. - **Built a curated list of commonly viewed words.** Frequent terms like "dispatch surcharge" and "receipt" sit right on the help center screen, each linking straight to the relevant topic. - **Worked across teams.** Adachi talks with the head of S.RIDE's app service planning almost daily. Watching Helpfeel usage trends shows what people search for but current content doesn't explain well, which lets the two teams coordinate fixes, like adding a short in-app note that links out to a fuller article. - **Let the data write content.** When "labor pains" started showing up in search history, the team added an article with contact details for each taxi company's priority maternity taxi service, so pregnant users can find the right number fast. - **Used it inside and out.** The help center was built for customers, but its constantly updated content and easy operation made it useful internally too. Employees are told to search it first with questions about S.RIDE's own services, which holds down internal inquiries and keeps reference material in one place. More eyes on the content also means typos, outdated entries, and inconsistent terms get caught and reported faster. ## Results Monthly active users grew. Dispatch orders grew. The inquiry rate, measured against MAU and order numbers, still fell by up to 30% after Helpfeel went in. The share of total inquiries about self-resolvable topics dropped by up to roughly 20%, which means more customers are finding their own answers before they ever contact support. As demand recovered and downloads grew 1.7x in a year, the support team added exactly one person, going from four to five, and still finishes all its work inside regular business hours. Adachi credits Helpfeel directly with containing what could have been a brutal jump in workload. Roughly 200 articles plus Intent Prediction Search now cover about 4,900 distinct question variations, and the wider company has taken note. With self-service handling more of the resolvable questions, the support team gets to focus on the inquiries that genuinely need a human: accidents, complaints, and the situations that call for nuanced, personal handling. ## Looking Ahead Next, the team plans to extend Helpfeel's Contact Sense integration to the in-app inquiry form, not just the web version. Since most contacts start there, guiding self-resolution at that point should push the self-resolution rate higher still. In April 2024, Japan opened taxi operators to ride-share. S.RIDE folded ride-share dispatch into the app and added content on how the new service works. Regulations around mobility keep shifting, and Adachi sees Helpfeel's job, getting new features and usage guidance to customers quickly, growing more important as they do. The core taxi dispatch business is already profitable. Looking further out, the Sony Group has asked S.RIDE to make the service usable by anyone through accessibility improvements, and to build experiences where mobility itself becomes the destination, autonomous vehicles included. As the service grows, keeping customer-facing information current will sit at the center of the experience, and Helpfeel is the tool the team counts on for that work. --- ## 30% Fewer Web Inquiries: How Noritz Earned Top Support Ratings Two Years Running: Noritz Corporation Source: https://en.helpfeel.com/case-studies/30-fewer-web-inquiries-how-noritz-earned-top-support-ratings-two-years-running # 30% Fewer Web Inquiries: How Noritz Earned Top Support Ratings Two Years Running - Web inquiries cut 30% through the inquiry form - FAQ access via inquiry form doubled within 4 months - Online repair booking rate increased 7 percentage points year-on-year - Top three-star HDI-Japan rating in Contact Center and Web Support, 2 consecutive years ## The Challenge Noritz Corporation manufactures essential home infrastructure products like gas water heaters, radiant floor heating systems, and gas stoves, serving customers nationwide through five contact centers. During peak season (October through February), inquiry volume roughly doubles, placing enormous strain on contact center staff. The Business Support Center Team 1, responsible for operations support and web systems, faced a fundamental challenge: their existing FAQ couldn't interpret the nuances behind customer searches. A query for "no hot water" could mean very different things, whether the tap wasn't producing hot water or the automatic bath-filling function wasn't working, and the resolution was different in each case. The previous FAQ tool lacked the ability to make that kind of nuanced interpretation of customer intent. With sparse analytics data, the team struggled to even identify the problems clearly. Operating under the mission of "understanding customer needs and delivering value," the team set out to enable self-resolution through a smarter FAQ while strengthening web-based customer relationships. ![Noritz team members in the office](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/noritz_body.jpg) ## Why Helpfeel After evaluating multiple FAQ systems, Helpfeel ranked highest across functionality, admin interface usability, and cost. Three factors proved decisive: **Search capability.** Noritz's products serve all age groups, and the terms people search vary widely, often mismatching product manual language. Even the water heater "remote control" is called "operation panel" by some customers and "screen" by others. The team concluded that flexible handling of these variations would be key to connecting customers with the information they needed. **Inquiry form function.** The ability to surface relevant FAQ articles at the point of inquiry, helping customers self-resolve before submitting a formal request. **Admin interface and analytics.** Usability and richness of the admin interface made it practical to run structured improvement cycles based on quantified, visualized data. ## What They Did - **Upgraded to AI search algorithm version 2,** adding support for sentence-form searches that predict customer intent from full sentences and always surface appropriate answer candidates. - **Reorganized the FAQ top page** based on Helpfeel search data, adjusting category structure and button layouts to improve navigation pathways. - **Added header links to "Online Repair Request" and "Product Manuals"** to improve customer access to high-priority tasks. - **Introduced Inquiry Form AI** in May 2024, surfacing relevant FAQ articles at the point of inquiry to enable self-resolution before submission. - **Embedded repair request form links** in every relevant FAQ article to encourage online booking over phone and email. - **Set a clear KPI** (FAQ article reach rate, from 77% to 80%) and worked systematically through each gap in navigation pathways and search keyword coverage identified at monthly review meetings. ## Results The FAQ article reach rate KPI reached the 80% target. Iterative improvements to the top page enabled many more customers to reach the information they were looking for. The Inquiry Form AI feature introduced in May 2024 contributed to resolving customer questions right at the point of inquiry. FAQ access via the inquiry form entry doubled within four months of launch, indicating a growing number of customers self-resolving before submitting a formal inquiry. Web-based inquiries through the inquiry form were reduced by 30%. Online repair booking rates also improved. By embedding links to the repair request form in relevant FAQ articles and adding header navigation, the team encouraged more customers to use online booking, which enables automated processing and faster response compared to phone and email. The online repair booking rate rose 7 percentage points year-on-year. In the 2024 HDI-Japan Public Ratings Survey, organized by the world's largest membership organization for the support services industry, Noritz received the top three-star rating in both the "Contact Center" and "Web Support" categories for the second consecutive year. Beyond the numbers, the team found a good balance between human effort and digital capability. Previously there was a sense of resignation that "FAQ can't really solve fundamental problems." The team now sees Helpfeel as contributing to their vision of "an organization that uses digital tools to maximize human value." ## Looking Ahead The team continues examining what KPIs to set, working to bridge the gap between quantitative data and what they perceive from customer interactions. Helpfeel's strength is delivering information to customers in a compact, fast format. The official website contains FAQ, repair requests, product information, and much more, and it can be hard for customers to find what's relevant to them. Through Helpfeel, Noritz can promote online handling of repairs and other procedures while assisting customers when they're deciding which products and services to choose. The Customer Service Division is constantly thinking about how to help customers use products with confidence every day. Gas water heaters have a typical lifespan of about 10 years, but as combustion appliances, age-related deterioration can lead to unexpected accidents. Hot water is also a lifeline. The team works to proactively encourage inspections and timely replacements, and rapid coordination with service shops across the country when repairs are needed is essential to maintaining the customer relationship. The aspiration is to build an organization that is indispensable to customers, one that can flexibly respond to their needs, so that customers continue to love and rely on Noritz products for years to come. Digital tools including Helpfeel will be central to making that possible. --- ## 40% Fewer Inquiries Despite Rising Orders: SOLIA Frees Time for Customer-Focused Work: SOLIA Inc. Source: https://en.helpfeel.com/case-studies/40-fewer-inquiries-despite-rising-orders-solia-frees-time-for-customer-focused # 40% Fewer Inquiries Despite Rising Orders: SOLIA Frees Time for Customer-Focused Work - Inquiries cut 40% while order volume increased, over 1,200 fewer inquiries in absolute terms - Inquiry rate dropped from 20% to 12% within 9 months - FAQ bounce rate decreased, customers finding answers faster - Escalation volume dropped, freeing time for higher-value customer work ## The Challenge SOLIA runs an e-commerce store for cosmetics and health foods, built around ALOBABY, a skincare brand made for babies. Most of the questions coming in were the easy kind. What's in this product. How do I use it. Can I change my address. Can I update how I pay. Parents could have answered these in seconds, and yet the volume never dropped. A lot of that came down to the tooling. Every change to the help center or the automated chat meant filing an edit request with another department, so there was always a gap between a customer asking something and the answer going live. The chat tool made it worse: each update had to be made in three separate places, which tripled the work and left plenty of room for a fix to go in incomplete. Launch a new product or change a policy, and the team watched unanswered questions pile up faster than they could clear them. ![Inside the SOLIA office](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/solia_body.jpg) ## Why Helpfeel The team had already tried tightening the help center and adjusting the chat tool. It helped, but not enough. So before looking at anything new, they set two requirements. First, the product had to show real inquiry reduction at other companies. Second, Customer Support had to be able to run the help center on its own, without routing every edit through another department. Helpfeel fit both. It had the implementation record they wanted, and its intent-prediction search let customers find answers in their own words instead of guessing at the right keyword. Before any commitment, the pre-sales team built a mockup of what SOLIA SHOP's help center would look like after launch, so the team could see the result rather than picture it. The Customer Success support that came after launch is what made the decision easy. ## What They Did - Set three KPIs: inquiry rate (inquiries divided by orders, which controls for campaign swings), raw inquiry volume, and the share of inquiries customers could resolve on their own - Met weekly to review inquiry and search data, adding or updating articles whenever a topic spiked - Used Helpfeel analytics to watch access trends, the keywords customers searched most, and no-hit searches (terms that returned nothing), so they could answer unmet needs quickly - Worked with Customer Success to move past simple session counts toward user-centric analysis: did people reach the information they came for, did articles match what they meant, did they give up mid-search - Edited articles directly in the Helpfeel Cosense interface, seeing every article at a glance, emphasizing content or adding images, and reordering what shows first, all without technical knowledge ## Results Inquiries fell about 40% even as orders went up, more than 1,200 fewer questions in absolute terms. The inquiry rate, inquiries divided by orders, dropped from over 20% before launch to 18% right after, then kept falling to 12% by nine months in. Repeat questions on the same topic dropped sharply. Bounce rate on the help center fell too, a sign that customers were finding what they needed. And as escalations dropped, the team got back the hours they used to spend handling them. That time now goes to managing the outsourced call center, analyzing what customers need, sharing what they learn with other departments, and making the help center better. Manager Yuri Suzuki put it plainly: better analysis has surfaced new problems, so the work isn't any easier. But the team spends its energy on customer satisfaction now instead of firefighting. ## Looking Ahead The team wants to push inquiries lower still and open up more time for customer-facing work. Some self-resolvable questions are still coming through, which points to room to improve how customers are guided and how the member page works. More people need to know the help center is there, too. Further out, the goal is to use the analytics to improve the whole store, not just support, and build a shopping experience customers actually enjoy. Suzuki is clear about where Helpfeel's value comes from. The tool is easy to use, yes. But the depth of the Customer Success support behind it is what sets it apart. For any company facing the same challenges, ease of use and quality of support are what matter most when choosing a help center service. --- ## 40% Fewer Inquiries in One Month: How FiNC Technologies Scaled Health App Support: FiNC Technologies Source: https://en.helpfeel.com/case-studies/40-fewer-inquiries-in-one-month-how-finc-technologies-scaled-health-app-support # 40% Fewer Inquiries in One Month: How FiNC Technologies Scaled Health App Support - Inquiries cut 40% in the first month - Response times improved from 3-4 days to same-day - 176 FAQ pages unified across 3 services in 6 weeks ## The Challenge FiNC Technologies operates three health and wellness services: FiNC (Japan's #1 healthcare/fitness app by downloads), FiNC Plus (a paid premium version), and FiNC MALL (a wellness marketplace). Each service had its own FAQ system, and none were successfully driving user self-resolution. Basic questions about app operations and settings kept inquiry volumes consistently high. When FiNC received media coverage and new user numbers spiked, inquiry volumes surged in step, putting significant strain on customer support teams. With limited headcount, the team needed to reduce inquiry volumes by 30% to maintain and improve support quality. The three services used completely different systems: FiNC used a third-party FAQ platform with UI usability issues and exact keyword matching that often failed to surface relevant content. FiNC MALL used an open-source CMS. FiNC Plus had FAQ content written directly into the service site's code. This fragmentation made management cumbersome and knowledge accumulation difficult. The FiNC system was also restricted to a single designated editor due to licensing constraints, creating a bottleneck. ## Why Helpfeel The team evaluated two other FAQ systems before discovering Helpfeel. One was rejected due to its English-only management interface, which raised concerns about implementation and ongoing operation. The other was eliminated when commercial discussions revealed unexpectedly high costs. The team needed a system that could migrate all three services' FAQ pages quickly, serve as both an FAQ system and a knowledge-sharing tool, and keep ongoing running costs low. Helpfeel matched these criteria. FiNC Technologies was already using Cosense (Helpfeel's knowledge-sharing service) internally for meeting minutes, notes, and documentation. The familiar interface meant FAQ pages could be created and knowledge shared with minimal friction. A contact at Helpfeel introduced the team to the FAQ product through that existing relationship. When the team tested Helpfeel from a user perspective, the search performance was a genuine surprise. Despite having assumed that all FAQ systems would perform roughly the same functionally, the difference with Helpfeel was immediately apparent in hands-on testing. ## What They Did - **Migrated 176 FAQ pages across three services in six weeks.** Kickoff took place in early July 2020, with the new site going live in mid-August. The internal workload was minimal: text adjustments, word dictionary creation, and detailed setup work were handled by Helpfeel's team. - **Established monthly customer success meetings.** Standing meetings with Helpfeel's customer success team deliver detailed reports covering metrics such as search hit rates and bounce rates. These data points form the basis for identifying next actions. The search data also serves as a valuable source of insight into user needs. - **Unified the FAQ platform to create cross-service awareness.** Consolidating the three FAQs on a single platform created a pathway for free FiNC users to discover the paid FiNC Plus and FiNC MALL services. The team expects this cross-service visibility to accelerate conversion from free to paid membership over time. - **Enabled simultaneous editing by multiple team members.** Removing the previous bottleneck of a single designated editor enabled rapid incorporation of user feedback into content. ## Results Within just one month of deployment, inquiry volume fell by approximately 40% month-over-month, exceeding the team's target of a 30% reduction for the full year of 2020 almost immediately after launch. The improvement was attributed to Helpfeel's superior keyword search performance (handling vague expressions, intuitive language, and spelling errors) combined with a clean, simple design that dramatically improved FAQ usability and user self-resolution rates. With inquiry volumes reduced, the customer support teams gained sufficient capacity to respond to remaining email and chat inquiries both more thoroughly and more quickly. Response times that had previously stretched to 3-4 days for initial replies improved to same-day responses after Helpfeel was deployed. ## Looking Ahead A growing trend among modern users is a preference for self-service experiences, resolving issues independently rather than contacting a human representative. Helpfeel is the right tool for realizing self-service customer support. When evaluating a system migration, functionality and cost are important considerations, but Helpfeel's customer success support was also exceptional. Even with limited internal resources, the migration can be completed smoothly. --- ## 40% Fewer Self-Resolvable Inquiries: How KOMEHYO Freed Support Teams to Focus on High-Value Customer Conversations: KOMEHYO Inc. Source: https://en.helpfeel.com/case-studies/40-fewer-self-resolvable-inquiries-how-komehyo-freed-support-teams-to-focus-on # 40% Fewer Self-Resolvable Inquiries: How KOMEHYO Freed Support Teams to Focus on High-Value Customer Conversations - EC procedure inquiries down 30-40% year-over-year - 8,300+ support interactions per month (phone, email, chat) with leaner staffing - 115 articles live after one year, up from 95 at launch ## The Challenge KOMEHYO operates more than 160 luxury resale stores across Japan, with an EC site listing approximately 50,000 unique items. As a resale business, every piece is one-of-a-kind, and many customer conversations require detailed human attention: discussing condition, appraisal nuances, or arranging in-store transfers for inspection before purchase. As the company accelerated store openings and trade-in acquisition, the support team proactively built capacity to handle growing inquiry volumes. But a significant portion of phone and email inquiries, even during business hours, covered straightforward topics customers could resolve themselves, like accepted payment methods or basic EC procedures. These self-resolvable inquiries mixed in with high-value conversations, causing phone congestion and requiring frequent callbacks. At night, when no human support was available, the EC site received substantial traffic with no fast self-service path for simple questions. The existing FAQ was hard to find, relied on external vendors for content updates (causing delays), and had poor search functionality, making it largely impractical. For a company dealing in high-value pre-owned goods, the goal was to concentrate human resources on interactions that truly require them. Streamlining self-resolvable inquiries was not a cost-cutting measure, but a way to deliver a better overall customer experience. ![KOMEHYO team members during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/komehyo-ec_body.jpg) ## Why Helpfeel KOMEHYO was simultaneously evaluating a customer-facing FAQ and a new internal FAQ to strengthen information sharing between stores and headquarters. Given the dual rollout, the quality of ongoing support was a top priority. The team needed a vendor who could provide concrete, regular improvement recommendations for both use cases. Helpfeel's search accuracy and customizability were standout factors. The combination of strong search performance and the ability to customize the system addressed the existing pain points comprehensively. ## What They Did - Migrated 95 articles from the previous FAQ and grew to 115 published articles over approximately one year of operation. - Invested heavily in search customization: trade-in related content was configured to surface reliably from multiple spelling variants and synonyms ("kaitori," "kai-tori," "satei," "baikaku"), routing customers to the same relevant FAQ content. - Added input placeholder text cycling through example prompts ("certification," "warranty") to guide customers toward the right keywords. - Promoted trade-in topics directly below the search box as a business priority. - Solved the discoverability problem using Helpfeel's popup feature, accessible directly from the EC site top page, with banner size and position adjustable in fine detail to remain prominent without causing accidental taps on narrow smartphone screens. - Ran monthly improvement cycles with Helpfeel's Customer Success team: expanding target search keywords, enriching FAQ content to match, and continuously refining the system based on user behavior analysis. ## Results Phone and email inquiries each run at approximately 4,000 per month, with an additional approximately 300 chat inquiries per month. Of these, inquiries related to EC procedures and on-site operations have decreased by 30-40% compared to the same month in the prior year before Helpfeel was introduced. These gains are attributed to the cumulative effect of monthly improvement cycles: expanding target search keywords, enriching FAQ content to match those keywords, and continuously refining the system based on user behavior analysis. Both the customer-facing and internal FAQ improvements benefit from regular, specific, actionable proposals from the Helpfeel team. ## Looking Ahead Web traffic data shows that the overwhelming majority of visitors, regardless of whether they intend to buy, sell, shop online, or visit a store, enter through the EC site's top page. The FAQ is now prominently positioned there, and the team has observed that customers are searching for specific product and service topics far more than expected, not just EC usage guidance. FAQ content has been expanded accordingly. The next phase will focus less on simply adding more articles and more on deepening integration between the FAQ and other existing site content, guiding customers fluidly toward resolution across all touchpoints. For example, the trade-in service spans three distinct channels (in-store, courier, and home-visit), and the FAQ could play a role in explaining what is common across all three before routing customers to the specific channel most relevant to them. For KOMEHYO, the customer-facing FAQ is a tool for achieving the kind of careful, attentive service that creates loyal customers. When a customer calls about an item listed on the EC site, the staff member typically doesn't have the physical item in hand. Explaining what can be determined remotely and offering options like in-store transfer takes time. Preserving that human capacity for situations that genuinely require it is exactly why the team continues to invest in self-service support. --- ## 42% Drop in Unnecessary Inquiries Despite 80% User Growth at MarmGarden Resort: Mom's Inc. (MarmGarden Resort Hayama) Source: https://en.helpfeel.com/case-studies/42-drop-in-unnecessary-inquiries-despite-80-user-growth-at-marmgarden-resort # 42% Drop in Unnecessary Inquiries Despite 80% User Growth at MarmGarden Resort - Unnecessary inquiries cut 42% despite doubling total inquiry volume - User base grew 1.8x in first year of operation - 8,800 LINE users engaging with self-service FAQ - Extended multi-turn support conversations significantly reduced ## The Challenge MarmGarden Resort Hayama opened in December 2021 as a 24-hour postnatal care hotel, a service category still relatively new in Japan. Their customers faced a unique problem: they were "not knowing what they don't know," uncertain even about what questions to ask about a service they'd never experienced before. Despite publishing information on their website, approximately 10 inquiries arrived daily, many asking about details already documented online. The information wasn't reaching customers effectively. With 90%+ of inquiries coming through LINE (Japan's dominant messaging platform), casual back-and-forth conversations often stretched across many messages. For a small team handling support alongside other duties, extended exchanges created real risk of disrupting other work. Previous solutions fell short. LINE's built-in AI auto-response couldn't handle variations in how customers phrased searches. A customer support tool they tested couldn't be sustained independently. With limited staff, the team couldn't dedicate resources purely to content creation and had no clear starting point for improvement. As the service expanded and new offerings launched, continuing in reactive inquiry mode felt unsustainable. ## Why Helpfeel The path to Helpfeel came through proof within their own corporate group. Karaoke Pasela, another business in the same group, had already adopted Helpfeel with concrete results. Hearing the internal success story gave President Saito a clear picture of what adoption could deliver. Four features made Helpfeel the right fit: **Intent-prediction search.** Previous tools couldn't match different phrasings of the same question. Helpfeel's ability to handle search expression variation addressed a significant gap directly. **Rapid content updates.** As a hospitality business tied to government tourism campaigns and subsidy programs where policies change frequently, the ability to publish and revise content quickly was essential. **Popup functionality.** A "Need help? Click here" button triggering a popup search window lowered the barrier for users to try the FAQ before contacting support. **Ongoing Customer Success support.** With a lean team that couldn't dedicate a specialist to FAQ management, having active CS team expertise was essential for sustaining quality. ## What They Did - **Used FAQ pages as campaign landing pages.** For time-sensitive tourism subsidy campaigns where government policies change rapidly, Helpfeel's rapid publication capability allowed the team to reflect current information quickly and create proper search navigation. - **Integrated FAQ across all channels.** The popup FAQ button was placed prominently on the website. During phone calls, staff directed callers to the FAQ. For LINE exchanges, the team shared links to specific FAQ pages, encouraging customers to explore at their own pace. - **Enriched content with Customer Success guidance.** One frequently searched article, "How old can my baby be?", originally provided only "We accept infants under 3 months of age." At the CS team's suggestion, it was expanded to include persuasive details about the hotel's specific care offering and key differentiators, transforming a bare answer into content that actively supports booking decisions. - **Used search data cross-departmentally.** Customer search data accumulated through Helpfeel informed not only FAQ improvements but also on-the-ground hospitality quality across departments. ## Results Total inquiry volume grew roughly in proportion to the 1.8x user growth, estimated to have approximately doubled based on promotional activity and service expansion. Despite this, the share of FAQ-resolvable inquiries fell from approximately 70% to 40%, a 42% reduction in unnecessary contacts. Because total volume doubled during the same period, the felt impact on the team is estimated to be at least half of what the burden would otherwise have been. Extended multi-turn LINE conversations became significantly less frequent. Staff reported that guests' anxieties were being resolved through the FAQ, with the majority of questions now answered independently rather than requiring personal exchanges. Guest feedback was positive. One comment: "I was planning a long stay, so it was great to be able to find out so much detail in advance." The FAQ also helped guests who had already decided to book, who naturally start thinking of follow-up questions, to find answers independently without diminishing their sense of being cared for. ## Looking Ahead For a hotel offering a service category that guests have never experienced before, investing in a rich, well-designed FAQ directly supports booking conversion, guest satisfaction, and operational efficiency simultaneously. Saito's view: the FAQ entry point is a customer's first window into the service. Enriching that content converts anxious uncertainty into confident, informed interest, ultimately supporting bookings and allowing the team to scale without being overwhelmed by inquiry volume. --- ## 50% Better Search Results in One Week: How Raksul Scaled Customer Support Through Peak Season: Raksul Inc. Source: https://en.helpfeel.com/case-studies/50-better-search-results-in-one-week-how-raksul-scaled-customer-support-through # 50% Better Search Results in One Week: How Raksul Scaled Customer Support Through Peak Season - Search hit rate improved 50% within one week of deployment - No-hit rate decreased 32%, directing customers to the right answers - Inquiry reduction of ~20% during cross-departmental FAQ improvement efforts - Handled TV commercial surges and COVID-19 inquiry spikes without overwhelming the support team ## The Challenge Raksul Inc. runs a lot of businesses under one roof: online printing and marketing support, logistics, TV commercial services, and corporate IT platforms. On the flagship Raksul service, most of the people contacting support are first-timers trying to order print online for the first time. And the learning curve is steep. They have to choose between dozens of paper types that vary by thickness and finish, figure out the data file requirements, and follow steps that traditional offline print shops never asked of them. So Raksul built a help center. Roughly 700 articles, written to let customers find their own answers. But the search underneath it had a problem. Printing is full of words that mean the same thing. A "flyer" might be a "pamphlet," a "leaflet," or a "pera," a Japanese printing term. A "sticker," a "label," and a "seal" are all the same thing to the person typing. The old keyword-matching search only returned a result when the customer used the exact word Raksul had used. Anything else came back empty. Four out of every ten searches found nothing. Here's what that does. Customers can't find the answer, so they contact a human. Human-handled inquiries stay high. The support team works under pressure that never lets up. And the whole time, TV commercial campaigns were pulling in waves of new customers, each peak threatening to bury support all over again. ![A Raksul team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/raksul_body.jpg) ## Why Helpfeel Raksul didn't sit still. The team ran a cross-departmental project to overhaul the help center and cut inquiry rates by about 20%. Good result. But they hit a wall. Better content alone couldn't push the number any further, and hiring more agents every time the business grew was never going to hold. Plenty of customers would rather solve the problem themselves than wait on support, so whatever came next had to make self-service actually work. Raksul's COO first ran into Helpfeel at an industry event. He tried Helpfeel-powered sites himself and the search response surprised him: fast, and flexible about how people phrased things. It helped that a long list of well-known companies in the same e-commerce space were already running it. The last piece was the customer success support that comes with Helpfeel. The team already knew that analyzing and improving a help center is real, ongoing work, so having hands-on help with that was a deciding factor. ## What They Did - Deployed Helpfeel on the Raksul website's public FAQ - Integrated Helpfeel with both the existing FAQ system and chatbot UI features - Used Helpfeel's customer success support for ongoing analysis and content optimization ## Results One week in, the search hit rate was up 50% and the no-hit rate had dropped by about 32%. Customers could find the right answer and get on with it. The team had braced for slow gains over six months of iterating. Getting there in a week was a genuine surprise. Then the campaigns showed what it was really worth. New customers spike hard when the commercials run, and inquiries climb right alongside them. During the rollout it got harder still: COVID-19 hit, state-of-emergency declarations and their lifting set off waves of new orders and cancellations, and inquiry volume ran past every projection. Because Helpfeel was already live, customers kept resolving things on their own, and the inquiries that needed a human stayed in check. ## Looking Ahead Helpfeel moved the customer experience at Raksul forward in a real way. When customers can answer their own questions and use the service without friction, they come back, and Raksul expects that to show up in repeat purchase rates. The team sees Helpfeel as a tool for companies that treat customer support as a driver of operational quality and business value rather than a cost to contain. From here, Raksul wants to push both the quality and the efficiency of support, and to open up new support channels as customer needs change. All of it in service of a better experience for the people on the other end. --- ## 50% Search Adoption and 37% Lower Bounce Rates: How NICE TV Cut Support Load While Scaling Fiber Migration: Shinkawa Information Center Inc. (NICE TV) Source: https://en.helpfeel.com/case-studies/50-search-adoption-and-37-lower-bounce-rates-how-nice-tv-cut-support-load-while # 50% Search Adoption and 37% Lower Bounce Rates: How NICE TV Cut Support Load While Scaling Fiber Migration - Search utilization rate up from 43% to 50% in four months - Bounce rate down from 43% to 37% in the same period - Approximately 300 monthly phone inquiries, each requiring 30-60 minutes, targeted for self-service migration ## The Challenge Shinkawa Information Center (NICE TV) provides cable television and telecommunications services centered on Uozu City in Toyama Prefecture. In early 2022, the team faced a critical resource conflict: their highest operational priority was migrating all contracted customers to fiber optic services, a project demanding significant staff time. But phone support was consuming that time. Roughly 300 inquiries per month were coming in by phone, with each call requiring 30 minutes to an hour to resolve. Many of these calls came from elderly customers asking about issues like "the TV isn't displaying a picture" or "the internet isn't connecting." In many cases, the fix was simple. For example, after equipment maintenance, modems often just needed to be unplugged and replugged. But customers called rather than troubleshooting themselves. The team estimated that roughly one-third of all inquiries could be handled via an FAQ site instead of tying up the phone lines. At the time, the working assumption was to strengthen phone support itself by pulling in staff from other departments or hiring part-time help. But those were stopgap measures, and the fiber migration couldn't wait. ## Why Helpfeel When the team first heard about Helpfeel, the reaction was immediate: "There is nothing else out there like this." Two aspects stood out: the technology and the support. On the technology side, Helpfeel's web directors and technical writers handled the initial construction of the FAQ site, meaning minimal effort required from the client. They also anticipated the kinds of words customers might search for and linked those terms to FAQ content to improve hit rates. On the support side, having a dedicated customer success specialist and a team of experts provide ongoing guidance was seen as highly valuable. In particular, the monthly access analysis reports and continuous improvement proposals were a major draw. The company had maintained an FAQ site previously, but it had been left largely untouched after initial creation. Even when looking at analytics data, staff had no idea what it meant or how to act on it. Knowing that an expert would walk them through the data and provide actionable guidance made the decision easy. The team was ready to adopt Helpfeel immediately. ## What They Did - **Built FAQ content around photos and diagrams for intuitive understanding.** For issues like the internet dropping after maintenance, the team created FAQ pages with actual photos of the equipment. Previously, staff would explain over the phone "find the modem and..." but customers often didn't know which device was the modem, leading to on-site visits. By using photos and diagrams, the team made information easy to understand at a glance. - **Minimized content creation burden through collaboration.** When there was something to add, staff simply prepared the body text along with any photos or diagrams. The customer success team provided recommendations based on access analysis data, making the PDCA cycle move much faster than when the team managed everything alone. - **Maintained ongoing communication via Slack.** Monthly meetings with Helpfeel's customer success team provided access analysis results and specific improvement proposals. In between meetings, Slack allowed flexible and timely responses to questions or consultations. ## Results The biggest change was gaining visibility into the customer's situation. Beyond operating and improving content based on identified search keywords, the team began tracking satisfaction-related metrics for the FAQ site and started to see results: - Search utilization rate increased from 43% in April to 50% by August - Bounce rate decreased from 43% in April to 37% by August While inquiry volume reduction had not yet reached the target figure given the system had only recently been deployed, the team expressed confidence this would be achievable through continued incremental improvements. One particular insight stood out: the analytics revealed not only which search terms returned zero results, but also what words customers searched for on their second attempt after a failed first search. This uncovered keywords the team had never anticipated customers would use, enabling new content additions to address previously unmet needs. ## Looking Ahead The current goal is to increase the proportion of middle-aged and elderly customers who are able to resolve issues using the FAQ site. Because IT literacy varies significantly among this demographic, simply placing a link to the FAQ on the website will only reach a limited audience. The team is exploring a range of awareness-building tactics, such as including QR codes and URLs on physical mailings, to gradually increase FAQ utilization rates among this user group. --- ## 60% Fewer Inquiries in One Month: How Karaoke Pasela Centralized National Support with Helpfeel: Newton Inc. Source: https://en.helpfeel.com/case-studies/60-fewer-inquiries-in-one-month-how-karaoke-pasela-centralized-national-support # 60% Fewer Inquiries in One Month: How Karaoke Pasela Centralized National Support with Helpfeel - Inquiries cut 60% within one month of launch - Support centralized to head office with significantly reduced headcount - Company-wide consistency in responses achieved for the first time - Contact Center Award 2024 People Division Prize winner ## The Challenge Newton Inc. operates Karaoke Pasela, a national karaoke chain offering multi-purpose room rental, and GRACE BALI, a private event space for parties and banquets. For years, each branch handled its own customer support: phone lines, FAQ pages, the works. The fragmentation created three problems. First, cost. Every location absorbed its own inquiry overhead, and the most common questions (how to connect a smartphone to the room's TV, whether outside food was allowed) were high-frequency and recurring. These questions arrived during the venue-selection stage, meaning slow or inconsistent responses risked losing customers before they even booked. Second, inconsistency. Customers received different answers depending on which branch they called, creating the impression that "each location does things differently." This made the problem self-perpetuating: customers who experienced inconsistent responses were more likely to call again rather than trust the FAQ the next time they visited a new location. Third, search failure. The existing FAQ systems couldn't accommodate different ways of expressing the same question. If a customer's keyword didn't precisely match article text, they got no results and ended up calling. Newton decided to consolidate all customer support at head office. A centralized team with reduced headcount would only be viable if the FAQ could absorb the bulk of inquiries that had previously been handled individually by each branch. ## Why Helpfeel After evaluating multiple FAQ systems, Head of Sales Promotion Tomoyuki Kitami's overall assessment was that virtually all competing products failed to meaningfully satisfy customer needs. The baseline expectation had become that FAQ tools simply wouldn't perform well, and that was just accepted. Helpfeel was the only system that delivered the search experience customers actually expected. His first reaction to seeing Helpfeel's search experience was genuine excitement. While the initial appearance resembles any other FAQ (a search box, a magnifying glass button, commonly searched articles below), the behavior after typing is unmistakably different. As keywords are entered, article candidates update in real time, anticipating the user's intent and surfacing relevant content before the search is even completed. This "getting ahead" of the user's needs was something he had not seen in any other FAQ system. Before selecting Helpfeel, Kitami reviewed live FAQ sites of other companies already using the product. He was impressed by how cleanly the Helpfeel interface integrated with each company's own site design, maintaining visual consistency with the parent brand without jarring the user. ## What They Did - **Deployed FAQ entry points** on the Karaoke Pasela top page ("Frequently Asked Questions / Contact Us") and GRACE BALI's banquet and party page, both powered by Helpfeel. - **Integrated Helpfeel's inquiry form feature**, which predicts and displays relevant FAQ articles based on what a user is typing into the contact submission form. This creates a second self-service intervention point: even at the moment a customer is about to submit a contact request, relevant content is surfaced, giving them a final opportunity to resolve their question without completing the submission. - **Migrated per-branch FAQs into a single unified Helpfeel FAQ**, with Helpfeel's Customer Success team reviewing not just the technical transfer but also the content itself, proposing improvements to individual articles and the overall structure. The consolidation became an opportunity to upgrade content quality at the same time. - **Established monthly review sessions** with the Customer Success team, receiving data-driven improvement recommendations covering FAQ usage patterns, article-level analytics, and keyword search maintenance. The team also receives feedback on the content of inquiries submitted through the contact form, providing intelligence on customer needs that informs new product ideas and business development planning. ## Results Within one month of launch, human-handled inquiries dropped by approximately 60%. The centralized head office contact team, operating with significantly fewer staff than the previous distributed model, was able to manage the reduced volume without the window becoming overwhelmed. Company-wide alignment on how to answer customer questions was achieved for the first time. Kitami noted this had meaningful brand value as well as operational benefit: customers now receive the same answer regardless of which channel or location they interact with. The team is also planning to use the FAQ as an internal training and reference tool for new staff education, leveraging the consistency and accessibility of the centralized knowledge base. At the Contact Center Award 2024 (September 2024), Newton Inc., presenting as part of the NS Group, received the People Division Prize for the Karaoke Pasela inquiry consolidation and DX initiative. The award, which focuses on human capital aspects of contact center excellence, recognized the achievement of launching a contact center from scratch in a short timeframe with a lean team while meaningfully improving customer satisfaction scores. ## Looking Ahead Among companies that operate real-world physical spaces or sell physical products, the instinct to leverage online tools for marketing and customer support has historically been slow to develop. Yet investing in online customer support directly drives customer satisfaction, operational efficiency, cost reduction, and staff workload relief. Beyond the core function of guiding customers to self-resolution through superior search, Helpfeel's search data provides a quantitative window into customer needs, surfacing signals that would otherwise be invisible in an offline-only operation and generating new business opportunity insights in the process. --- ## 7,000 Daily Inquiries Resolved: How Nagoya Scaled COVID-19 Vaccine Support Without Adding Headcount: City of Nagoya Source: https://en.helpfeel.com/case-studies/7-000-daily-inquiries-resolved-how-nagoya-scaled-covid-19-vaccine-support # 7,000 Daily Inquiries Resolved: How Nagoya Scaled COVID-19 Vaccine Support Without Adding Headcount - 7,000 daily FAQ accesses at peak during vaccine rollout - Equivalent impact of adding ~300 operators to call center capacity - 85%+ answer rate maintained even during peak periods - 190 FAQ articles covering reservations, precautions, and vaccine information ## The Challenge Nagoya, Japan's third-largest government-designated city with over 2.3 million residents, faced an unprecedented surge in citizen inquiries when COVID-19 vaccine rollout began in April 2021. Even after expanding to 125 phone lines, the city's call center was overwhelmed: 80,000 calls per day flooded in, and most could not be answered. Citizens who couldn't get through by phone began forming queues at ward offices, requiring the city to deploy staff on-site to help residents make smartphone reservations. As eligibility was set to expand to younger age groups, nearly doubling the eligible population, the city recognized the need for new information channels that would enable citizens to find answers before resorting to a phone call. ## Why Helpfeel The city consulted with the outsourced operator managing the vaccine call center, which had prior implementation experience with Helpfeel and proposed it as a tool that could function as a 24-hour, 365-day online support window. Compared to printed flyers enclosed with vaccination tickets, Helpfeel offered the advantage of reflecting the latest information at all times with no print lead time. While the city's website constantly published updated information, the need to cover every edge case created a dilemma: the more thorough the writing, the harder it became for any individual reader to find what was relevant to them. Helpfeel's ability to let users search for exactly what they wanted to know and receive a clean, simple answer addressed this directly. The intent-prediction search function was particularly highly rated. For example, surfacing "reissuance of vaccination ticket" as a result when users searched for colloquial terms like "lost it" or "misplaced it" bridged the gap between everyday conversational language and the official terminology used in public health communications. The city also used a conversational chatbot for COVID-19 communications, but reliably interpreting user intent and delivering resolution through a dialogue format proved difficult. Helpfeel's consistent ability to understand citizen intent was seen as a key advantage. ## What They Did - Launched a Vaccine FAQ Site in August 2021 with approximately 190 articles covering topics such as how to make vaccination reservations and precautions before and after vaccination - Printed the FAQ URL as a QR code on the informational flyer enclosed with every vaccination ticket mailing sent to residents to maximize awareness - Entrusted day-to-day FAQ operations to an experienced vendor partner, with monthly reporting and ongoing email communication to review results and identify issues - Conducted a focused effort to continuously improve cases where user searches returned zero results (no-hit instances), ultimately achieving near-comprehensive coverage of actual questions being asked - Adopted the finalized FAQ content as an internal reference manual for call center operators, enabling consistent and unified responses across all channels ## Results The FAQ site received up to 7,000 accesses per day at peak. Given that the vast majority of visits resulted in self-resolution, the city treated each FAQ access as equivalent to one phone call answered. Calculated against the call center's actual response performance, handling 7,000 inquiries per day via FAQ was assessed as delivering the equivalent impact of adding approximately 300 operators. This contributed to substantially alleviating congestion at the call center, which had scaled to a maximum of 400 lines, and maintaining an answer rate of 85% or higher even during peak periods. One notable trend observed in the FAQ data: searches related to vaccination certificates (vaccine passports) for overseas travel were disproportionately common in the FAQ, a pattern that differed from the call center, where more immediate vaccination scheduling questions dominated. This suggested that FAQ is particularly well-suited for topics where citizens need to reliably retrieve important information at any time of day, rather than for urgent, time-sensitive matters. ## Looking Ahead Government information must be communicated in ways that are accessible to everyone. An FAQ tool like Helpfeel can be used by anyone capable of internet searching on a smartphone, regardless of age. Citizens can access it at whatever time is convenient, and can take screenshots to retain the information they find. The proportion of citizens who find an FAQ more convenient than calling a phone helpline is likely to continue growing. For self-resolution of questions with single, straightforward answers, a well-designed FAQ tool like Helpfeel is an effective and proven approach. --- ## 70% Fewer Inquiries: How NTT Data MSE Doubled Employee Adoption and Cut Helpdesk Workload with AI: NTT Data MSE Corporation Source: https://en.helpfeel.com/case-studies/70-fewer-inquiries-how-ntt-data-mse-doubled-employee-adoption-and-cut-helpdesk # 70% Fewer Inquiries: How NTT Data MSE Doubled Employee Adoption and Cut Helpdesk Workload with AI - Employee utilization doubled from 5-25% to over 60% in six months - Inquiry-handling workload reduced by up to 70% across business divisions - Cost savings equivalent to 2 full-time employees per year - 870 articles prepared within 4 months of launch ## The Challenge NTT Data MSE Corporation employs approximately 2,000 staff and provides IT solutions across IoT, automotive, and mobile domains. As part of a company-wide DX initiative launched in 2022, the company identified a critical bottleneck: information was siloed across 14 departments under 5 business units, each managing knowledge independently with wildly different methods. Some ran intranet helpdesk pages, others used chatbots, and some simply posted PDF files. Veteran employees who knew where each department's information lived could navigate the system by asking the right person directly. But feedback from employees on secondment and mid-career hires made the problem impossible to ignore. The company needed an internal helpdesk that anyone could use intuitively, without relying on institutional knowledge that took years to build. ![Inside the NTT Data MSE office](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/nttd-mse_body.jpg) ## Why Helpfeel When evaluating solutions, NTT Data MSE prioritized two factors: usability and the quality of operational support. Helpfeel stood out for its polished search experience, including natural language AI-powered search, and an interface so intuitive that employees could use it without training. Since this was a cross-departmental project involving internal coordination, formal approvals, and many organizational hurdles, the availability of hands-on implementation support was critical. The team had no prior experience running a helpdesk operation internally. Looking ahead to post-launch operations, continuous support from Helpfeel's Customer Success team gave the team confidence. Tools like this require ongoing refinement, and Helpfeel offered a framework to reliably run that PDCA cycle. ## What They Did - **Phased rollout by department.** Rather than overhauling everything at once, the team took a department-by-department approach. They extracted information from each department's intranet pages and shared folders, compiled article candidates, had staff review them, and converted them into articles. Errors were corrected and duplicate content was consolidated. - **Intentional launch strategy.** While article creation was phased, all employees were targeted as users from day one. The goal was twofold: let employees experience how easy it was to find information, and deliberately surface gaps to motivate departments to contribute articles. A one-month parallel operation period alongside old FAQs and chatbots allowed employees to acclimate gradually. Within 4 months of launch, approximately 870 articles had been prepared. - **Branding for approachability.** To make the helpdesk feel welcoming, the team named it "MSE Concierge" instead of calling it a helpdesk. A link was placed on the company intranet homepage, with announcements in the information section, executive meetings, and weekly business promotion meetings to drive awareness. - **KPI tracking and iteration.** Early on, the team focused on exit rate and tracked "no hit" keywords to identify content gaps. As the article library grew, they shifted focus to re-search rate and search depth as indicators of whether articles were genuinely helping employees. - **Customer Success partnership.** Monthly review meetings with Helpfeel's Customer Success team provided data-driven improvement recommendations. Over time, the internal team developed enough understanding to use analytics reports to feed back usage status to each department's helpdesk owner, enabling a continuous improvement cycle. ## Results Within six months of launch, over 60% of all employees were actively using MSE Concierge, compared to just 5-25% who had used the previous chatbot. The number of active users has continued to grow month over month. Helpfeel reduced employee time spent searching for information and cut the workload of business divisions (both inquiry response time and the time previously spent managing legacy FAQs and chatbots) by up to 70%. Based on internal estimates, the workload reduction achieved in fiscal year 2024 is equivalent to the effort of approximately 2 full-time employees per year in cost terms. Employees who had originally raised the problem, including those on secondment and mid-career hires, reported that information is now found overwhelmingly faster compared to using the old chatbot or static files. Employees have also begun proactively requesting new articles, reflecting the growing adoption of the platform. These achievements were recognized internally. MSE Concierge won a company-wide Project Award, honoring not only the Information Systems Department but also helpdesk representatives from each contributing business unit, validating the project's value as a company-wide DX initiative. ## Looking Ahead The team plans to use data analytics to identify and fill content gaps. A recent internal system change caused a temporary spike in inquiries, highlighting the need for a proactive operational framework to prepare content ahead of known change events in coordination with other departments. The team is evaluating Helpfeel's PDF search functionality. Much of the company's policy documentation and manuals exist as PDF files. Currently the helpdesk links to those PDFs. The ability to surface and directly navigate to the relevant section within a PDF would significantly improve the employee experience. From a company-wide information dissemination perspective, the team also wants to explore push-style notification mechanisms to proactively deliver important information to employees. Currently, employees must actively go find information. The team is exploring ways to design notification integrations and timing to ensure critical updates reliably reach employees. --- ## 70+ Monthly Inquiries Prevented: How DAC's Internal Helpdesk Scaled Without Adding Headcount: Digital Advertising Consortium Inc. Source: https://en.helpfeel.com/case-studies/70-monthly-inquiries-prevented-how-dacs-internal-helpdesk-scaled-without-adding # 70+ Monthly Inquiries Prevented: How DAC's Internal Helpdesk Scaled Without Adding Headcount - Prevented 70+ inquiries per month on average despite user base growing by 100+ - ~70% of users rated the system 'easy to use' in internal survey - 300+ FAQ pages consolidated to 230 well-maintained pages - Search hit rate improved further within 5 months of go-live ## The Challenge Digital Advertising Consortium (DAC), a leader in digital advertising and marketing solutions since 1996, runs a 5-person internal helpdesk serving more than 2,000 users across the organization and group companies. The team fielded over 1,000 inquiries per month covering PC malfunctions, software updates, and hardware replacements. Many questions were repetitive, and volumes spiked during new employee onboarding, personnel transfers, and company-wide hardware rollouts. The company had run an internal FAQ using a third-party service since 2013, but it never functioned effectively. Weak search performance made it hard for employees to find answers. On the management side, creating and updating articles was cumbersome and slow, causing new information to be published late and outdated content to go unmaintained. There were no clear standards for what content to create or who owned updates. The result was a vicious cycle: users found it faster to ask the helpdesk directly rather than search the FAQ, and the helpdesk staff found it faster to answer directly rather than maintain the FAQ. The FAQ was never updated, and the burden on the team remained unchanged. ## Why Helpfeel The helpdesk team first explored introducing an AI chatbot to automate operations, but that approach required significant effort to develop and refine scenario scripts. They shifted focus to replacing the existing FAQ with a system that offered better usability for both end users and administrators while allowing existing content to be migrated and reused. Helpfeel's search performance was the standout quality. The first time the team tested a demo, they were struck by how instantly and accurately the right pages appeared across a wide variety of search terms. The speed was genuinely surprising. The quality and proactiveness of Helpfeel's content improvement proposals and hands-on support provided strong reassurance, and together these factors led to the decision to replace the existing system. DAC also requested a page design customization. For system-related inquiries, FAQ pages often function as step-by-step operational manuals. When related administrative procedures on separate pages also needed to be referenced, links buried at the bottom of a page were easy to miss. Helpfeel adjusted the design so that related links appear in the first visible area of the screen (above the fold), ensuring users notice them immediately. ## What They Did - Consolidated more than 300 FAQ pages down to approximately 230 during migration, with Helpfeel's web directors and technical writers identifying and merging overlapping content - Customized page design to display related links above the fold for better visibility on procedure-heavy pages - Partnered with Helpfeel's customer success team on data-driven content additions, with concrete metrics driving each improvement and collaborative co-creation and review ## Results Since deployment, the number of users including group company employees has grown by more than 100 people, yet inquiry volumes have not increased. In proportional terms, the monthly inquiry count has been held down by approximately 70 inquiries relative to what would be expected given the user growth. Comparing month-over-month against the same months in the prior year, the reduction ranges from a minimum of 30 inquiries in lower months to more than 100 inquiries in higher months. Meaningful results are being achieved even as the user base expands. An internal survey conducted 3 months after deployment found that approximately 70% of respondents rated Helpfeel as "easy to use." When asked what specifically made it easy, comments included "even with vague search terms, I quickly find the answer I need" and "query suggestions appear while I'm still typing." The feedback was consistently positive. Within 5 months of migrating to Helpfeel, the search hit rate improved further beyond the already-strong baseline at launch. The helpdesk staff's operational burden has been substantially reduced. Recent company-wide changes, including a mass smartphone replacement rollout and significant shifts in working styles and infrastructure, have been navigated smoothly, with employees resolving issues themselves and reducing the load on the support team. ## Looking Ahead Currently the FAQ content is centered around information systems topics. Going forward, the team wants to expand coverage to include other departments that handle internal inquiries, ultimately making Helpfeel the single destination where any question in the organization can be answered. In alignment with DAC's corporate mission, "Empowering the digital future," the long-term goal is to reach a state where 100% of users turn to Helpfeel first whenever a work-related question arises, creating an environment where issues are resolved quickly and efficiently. --- ## 70% Self-Resolution: How Ashikaga Bank Transformed Web Support with AI FAQ: Ashikaga Bank, Ltd. Source: https://en.helpfeel.com/case-studies/70-self-resolution-how-ashikaga-bank-transformed-web-support-with-ai-faq # 70% Self-Resolution: How Ashikaga Bank Transformed Web Support with AI FAQ - Over 20,000 FAQ sessions per month - 70% of inquiry form visitors self-resolve - 300 fewer phone calls per month - Support now available 24/7/365 ## The Challenge Ashikaga Bank's Direct Sales Division was handling over 4,000 phone inquiries per month, mostly basic questions about internet banking and mobile apps. The call center could only operate during business hours on weekdays, leaving customers without support evenings and weekends. With limited operator capacity, customers often couldn't get through during peak times. The bank's website had roughly 670 FAQs scattered across individual product pages as plain text. Customers had to know which page to visit before they could find an answer. If they couldn't navigate there, they had no path to resolution. The team recognized customers were in a mindset of "it's faster to just call than try to search." Previous chatbot implementations had fallen into disrepair after staff transferred to other roles. The bank needed a solution that could scale support without indiscriminately adding headcount, while maintaining service quality through inevitable personnel rotations. ![A Ashikaga Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/ashikaga_body.jpg) ## Why Helpfeel The team evaluated FAQ systems alongside call center staffing improvements. What set Helpfeel apart was the operational consulting support. With no in-house FAQ expertise, having ongoing support felt reassuring as staff rotated every few years. The intent-prediction search was equally compelling. Customers could search using keywords like they would on the internet, and the right article would surface. While a category-organized FAQ was easier for the bank to manage internally, customers didn't know which category their question belonged to. Helpfeel would let them solve their problem smoothly, thinking from the customer's perspective. ## What They Did - Built a cross-functional FAQ operations team combining website staff and call center operators, bringing front-line customer perspectives directly into content creation - Focused on minimizing no-hit search results, treating any search that returned no information as unacceptable - Added an "FAQ" field to the stakeholder review format for homepage updates, ensuring teams considered FAQ changes alongside any content updates - Deployed Helpfeel's inquiry form AI to surface relevant articles as customers typed their questions - Made the FAQ accessible from the banking app and LINE in addition to the website ## Results FAQ access exceeded 20,000 sessions per month, with customers self-resolving even outside call center hours. Monthly phone inquiries dropped by approximately 300 cases compared to the previous year. Inquiries about internet banking and the app, which had been trending upward, reliably decreased. The inquiry form AI delivered the headline result: 70% of customers who accessed the inquiry form transitioned to the FAQ and self-resolved without submitting a request. The intent-prediction search displays relevant articles as customers type, naturally guiding them to answers. Beyond inquiry reduction, the team gained data-driven insight into what customers wanted to know. When FAQ data showed many customers searching for how to check their account number, the bank built a system allowing customers to look it up through an online application. The data-backed proposal convinced other departments quickly. Coordination strengthened between the website team, call center, and product management. Previously, customer feedback from the call center would get delayed reaching the product team. With the FAQ team as intermediary, call center staff report things move more easily. Call center operators now proactively anticipate topics likely to generate inquiry surges and suggest adding FAQ articles in advance. ## Looking Ahead The team is exploring ways to connect call center data with FAQ data for unified operational use, with the goal of eventually generating FAQ articles directly from inquiry history. They've joined Helpfeel's Regional Bank AI Community to share common challenges and solutions with other banks already using the latest AI technologies. With ongoing consulting support, the FAQ operates stably even as personnel rotate every few years. The bank has solved financial services' continuity challenge: staff transfers can't be stopped, but operations won't stop either. --- ## 700+ Pages Created: How Persol Tempstaff Built a Living BPO Knowledge Base with Helpfeel: Persol Tempstaff Inc. Source: https://en.helpfeel.com/case-studies/700-pages-created-how-persol-tempstaff-built-a-living-bpo-knowledge-base-with # 700+ Pages Created: How Persol Tempstaff Built a Living BPO Knowledge Base with Helpfeel - Over 700 knowledge pages created and growing - Usage approaching levels of existing internal portal site - Staff reporting 'information I couldn't find before is now findable' ## The Challenge Persol Tempstaff's BPO division had reached a turning point. In the early years, the team was small, projects were few, and knowledge lived at the project level. When someone had a question, they could walk over and ask a colleague working on a similar account. Informal peer-to-peer knowledge sharing was enough. But as the BPO market expanded, so did the division. More projects, more staff, more diverse client needs. The team recognized that continuing to rely on individual project silos and ad-hoc hallway conversations was no longer viable. To keep delivering high-quality value to clients, the accumulated knowledge across all projects needed to be managed systematically at the division level, not left scattered across individual contributors. ## Why Helpfeel The team surveyed a wide range of knowledge management systems, but most didn't align with the experience they wanted to create. When they encountered Helpfeel, the reaction was immediate: "This is it." The defining feature was Helpfeel's exploration-oriented search interface. Users could type freely using the words they naturally use in their work, and follow links from one piece of content to related content in an organic, non-hierarchical way. This "follow the thread" browsing style, where discovering one piece of knowledge leads naturally to discovering another, aligned precisely with the team's vision for how knowledge should live and be used in a dynamic BPO environment. Helpfeel felt less like a manual to look up answers in and more like a web of connected knowledge to explore. ## What They Did - **Built a submission-driven content model.** A three-person secretariat leads content creation and structure, but the source material comes from the field. Anyone in the division can submit knowledge they believe could be useful to others through a dedicated form. - **Incentivized contribution with a points-based reward system.** Staff who submit knowledge that becomes the basis for a new page receive points redeemable for rewards, making participation tangible and building the "use it and grow it together" culture. - **Shifted the mindset through internal education.** Multiple online orientation sessions communicated a core message: knowledge sharing is not about uploading perfect, finalized answers to a manual. It's about each person bringing their real-world experience into a shared pool, where it can be refined and improved over time by the collective. - **Leaned on Helpfeel's Customer Success team.** When the secretariat was unsure how best to structure or present submitted knowledge, Helpfeel's CS team provided guidance on both the technical and creative dimensions of content design. ## Results Over 700 pages, including drafts, had been created at the time of the interview, and the library continues to grow. Access volume has been increasing steadily, approaching the usage levels of the division's existing internal portal site. Staff have commented that "information I couldn't find before is now findable," and a core group of regular, enthusiastic users has begun to emerge. The initiative successfully built momentum toward the "use it and grow it together" culture the team envisioned, with more staff participating in knowledge contribution than before. ## Looking Ahead Accumulating and making knowledge visible is important, but the more critical challenge is turning that knowledge into tangible value. Simply storing knowledge is not enough. It must be actively used in ways that improve client outcomes. The goal is to transform the individual know-how scattered across the division into lasting organizational assets for the BPO business. Helpfeel is the platform through which the team's collective expertise can be brought together, refined, and made to shine. The ambition going forward is to continue raising the quality of the BPO knowledge base so that it directly elevates the level of value delivered to clients. --- ## 8,500 Hours Saved: How Life Card Built Smartphone-First Support for Gen Z: Life Card Inc. Source: https://en.helpfeel.com/case-studies/8-500-hours-saved-how-life-card-built-smartphone-first-support-for-gen-z # 8,500 Hours Saved: How Life Card Built Smartphone-First Support for Gen Z - Support workload reduced 8,500 hours/year - FAQ view rate achieved 17× baseline (1.7× the 10× target) - Inquiries cut 5,000 cases/month despite growing cardholders - FAQ operations streamlined from 2 staff to 1 ## The Challenge Life Card Inc. offers credit cards including the no-annual-fee "Life Card," frequently chosen by Gen Z customers as their first credit card. The company was handling 55,000 inquiries per month, with phone lines severely congested on the 27th of each month (direct debit date) and after long holidays. The majority of these questions covered basic topics already documented in the FAQ: payment confirmation, transaction history, and cancellation. The existing FAQ had limited analytics (only page views and a simple Good/Bad rating), and customers simply weren't using it. No matter how comprehensive the information was, if customers couldn't find what they needed, the FAQ served no purpose. The team needed a system that could provide detailed data on customer behavior and fundamentally improve the self-service experience. ![A Life Card team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/lifecard_body.jpg) ## Why Helpfeel Life Card discovered Helpfeel through their group company, Aiful, which had already adopted it. Three capabilities stood out: **Search performance.** When the team tested Aiful's FAQ, they were impressed by how search results appeared as soon as you started typing, before you'd even finished entering your keyword. They felt confident this would satisfy customers who prioritize "time performance" (getting value out of every minute). **Data analytics.** Helpfeel's built-in analytics dashboard would consolidate all data in one place, eliminating the need to download and manually analyze data by hand. **Customer Success support.** For the FAQ operations lead, having an expert walk alongside the team and offer advice and proposals they would never think of on their own sealed the decision. ## What They Did - **Set clear KPIs focused on self-service.** The team launched with a target of 10× improvement in FAQ view rate (FAQ views ÷ inquiry volume), and continuously verified they were providing the right content by monitoring article reach rate and tracking no-hit keywords. - **Built a smartphone-first experience.** With 90% of customers accessing the FAQ from smartphones, the team focused on creating articles that were easy to read on a smartphone screen, with images optimized for mobile. Helpfeel's drag-and-drop image editing made this dramatically easier, allowing FAQ operations to shift from a two-person team to one person handling it entirely. - **Used analytics to understand the full customer journey.** The team prioritized understanding the full picture of customers' search behavior. Helpfeel's analytics dashboard visualized, step by step, what kinds of questions customers were looking for, what they searched for, and whether they successfully self-resolved, making bottlenecks immediately clear. - **Leveraged Customer Success insights.** The CS team provided third-party perspective the internal team wouldn't notice themselves. For example, they alerted the team that a basic article on "card reissuance" was missing, a topic the team had taken for granted as requiring human contact. They also provided advice on tuning articles to appear in the right searches and making them more readable for customers. ## Results Life Card initially set a target of 10× improvement in FAQ view rate. They are now achieving 16-17× the view rate, reaching 1.7× their original target. The habit of customers checking the FAQ before reaching out is becoming established. Even as the number of cardholders continues to grow, inquiry volume has decreased by approximately 5,000 cases per month. Assuming 7 inquiries handled per hour, this translates to 714 hours per month, or approximately 8,500 hours per year of reduced support workload. Because Helpfeel makes it easy to update FAQ articles, the team can quickly add or update content ahead of campaigns and feature releases. In the past, these events would cause inquiry spikes. Now, by preparing the FAQ in advance, many customers view those articles proactively, and the team has been able to suppress sudden surges in inquiries. The nature of inquiries has also changed. Support agents handling live chat and email have noticed an increase in customers who have already read the FAQ before reaching out. Some even include the URL of the FAQ article they read in their inquiry form submission. Being able to understand "what the customer already knows" means less back-and-forth, shorter handling time per inquiry, and improved customer satisfaction. ## Looking Ahead Life Card's ultimate goal is to achieve zero inquiries by providing a customer experience where searching feels intuitive and optimal answers are immediately at hand. The team plans to use search logs and inquiry content commercially as well, suggesting the most relevant services (like credit limit increases or installment payment options) at the right moment for each customer. Looking further ahead, they want to integrate customer account data with FAQ data to enable more personalized information delivery, where the information you want to know appears naturally, exactly when you need it. This kind of new customer experience will contribute to improved satisfaction and business growth. --- ## 80% Fewer Inquiries: How Kokusai Kogyo Scaled Solar Support Without Adding Headcount: Kokusai Kogyo Inc. Source: https://en.helpfeel.com/case-studies/80-fewer-inquiries-how-kokusai-kogyo-scaled-solar-support-without-adding # 80% Fewer Inquiries: How Kokusai Kogyo Scaled Solar Support Without Adding Headcount - Inquiries cut 80% with a lean 3-person team - 43% of FAQ visitors use search; 80%+ reach an answer - Reduced churn rate and increased revenue ## The Challenge Kokusai Kogyo's Enegaeru service was facing a perfect storm. As solar energy demand surged in Japan following electricity price hikes and energy supply concerns, their cloud-based solar panel simulation platform was attracting an influx of new users: retailers, installation contractors, and trading companies entering the renewable energy market. The growth was welcome, but it came with a critical operational challenge. A single sales staff member was handling the entire inquiry load. During peak periods, product engineers were being pulled away from development to answer customer questions. The existing FAQ had been in place for some time, but its search functionality was poor. Customers couldn't find what they needed, and with new market entrants flooding in with basic questions about renewable energy fundamentals, the team recognized they needed to act proactively before the situation became unsustainable. ![A Kokusai Kogyo team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/kokusai-kogyo_body.jpg) ## Why Helpfeel While researching solutions, the team discovered Helpfeel and found exactly what they needed: high-precision search that could surface relevant articles even from vague or imprecise search terms. The intent-prediction search would enable customers to find answers and resolve questions independently, no matter how they phrased their query. The UI was superior to alternatives, and the team judged the investment worthwhile based on the expected improvement in customer experience. ## What They Did - **Built a lean FAQ operation with three sales staff**, prioritizing content on the most frequently asked topics while avoiding highly case-specific information that would bloat the FAQ and make navigation difficult. - **Added sales-support content like talking point guides** for their retailers and distributors, turning the FAQ into a resource that went beyond troubleshooting to actively support their users' business success. - **Used search data as voice-of-customer intelligence**, sharing keyword trends regularly with the product development team. Rising searches for "gasoline" signaled interest in EV simulation; increased views of subsidy articles indicated demand for policy-related information. - **Made FAQ onboarding a standard part of sales meetings**, walking new customers through the FAQ screen directly and establishing it as the primary resource for resolving questions, with phone, email, or chatbot reserved for issues that couldn't be self-served. - **Held monthly customer success meetings with Helpfeel's team**, reviewing search activity and keyword trends to identify what improvements were needed. These structured check-ins kept the FAQ from stagnating after initial setup. ## Results Since launching the Helpfeel-powered FAQ, the team estimates inquiry volume has decreased by approximately 80% on a perceived basis. The combination of comprehensive, well-curated content and continuous data-driven improvement delivered this result. Even with a small team, FAQ operations run efficiently and team members are now able to focus on their core responsibilities. Among visitors to the FAQ, 43% conduct a search, and more than 80% of those users successfully reach an answer. User feedback from interviews confirms the smooth experience: many companies reported that "when we have a question, we just search and it's resolved immediately, we're very satisfied." This seamless self-service experience translated into measurable business outcomes: reduced churn rate and increased revenue. High-traffic FAQ articles have been republished in the Enegaeru service blog, and the team plans to expand this content repurposing into social media channels to drive broader service awareness. ## Looking Ahead Helpfeel's search data revealed a near-term opportunity: the FAQ currently lacks enough introductory content about solar power and storage batteries, the foundational information that newly entering retailers and installation contractors are seeking. The team is building out a keyword library to serve these users better. Looking further ahead, the team envisions using AI agents to cover the full customer journey, from resolving questions and providing support to sales and lead generation. The concept is to automate the process from initial inquiry through proposal and closing, creating the equivalent of a skilled, always-available "online salesperson" operating entirely through AI on the Enegaeru platform. --- ## 80% Less Inquiry Workload: How Fujimak Unified 218 Engineers Across 70 Locations with AI Search: Fujimak Corporation Source: https://en.helpfeel.com/case-studies/80-less-inquiry-workload-how-fujimak-unified-218-engineers-across-70-locations # 80% Less Inquiry Workload: How Fujimak Unified 218 Engineers Across 70 Locations with AI Search - Inquiry handling workload reduced 80% - Access volume increased 5x from launch - 370 pages of PDF manuals made searchable - 102 articles added in first year ## The Challenge Fujimak Corporation manufactures commercial kitchen equipment for hotels, restaurants, and hospitals, where kitchens must never stop. Their 218 service engineers, spread across roughly 70 locations nationwide, operate 365 days a year on rotating shifts. For years, specialized knowledge lived in the heads of a few product experts. When service engineers in the field encountered a question about Combi Ovens or Vario products, everyone called Keita Aikawa in the East Japan Purchasing Group. When he was in a meeting, on a day off, or simply overwhelmed, engineers were stuck. Customers waited. Sales opportunities slipped away. Aikawa felt the weight of being a bottleneck. On working days, constant calls interrupted his concentration and pushed his own work aside. On days off, he knew colleagues in the field couldn't get answers they needed. Knowledge sharing relied on multi-hundred-page PDF manuals and email announcements, both easily missed or impossible to search efficiently. Remote single-engineer offices had no nearby colleagues to consult. The knowledge gap between urban and regional locations widened. The company had a large product lineup with frequent spec updates. Busy engineers, especially those less comfortable with IT, struggled to keep up. Previous system implementations had gone unused. The team needed a solution that would actually be adopted, and that could reproduce the skill Aikawa had developed over years: inferring what an engineer was really asking from a vague description. ![A Fujimak team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/fujimak_body.jpg) ## Why Helpfeel The team initially explored a chatbot, but Fujimak's president challenged them: "Isn't there something more cutting-edge? Have you thought through the operational burden?" That sent them back to evaluate from scratch. Two factors made Helpfeel the answer. First, its intent-prediction search could replicate what the team called "Aikawa-style intent prediction search," the craftsman's skill of interpreting varied expressions and guiding engineers to the right information. Second, Helpfeel's accompaniment-style Customer Success support structure addressed past failures. The company had experienced multiple system rollouts that staff never used. This time, they wanted professional expertise to ensure adoption happened reliably. ## What They Did - Launched with PDF Search to make 370 pages of existing product manuals instantly searchable, then prioritized frequently asked historical questions and rare critical-emergency edge cases not covered in the manuals for article creation. - After launch, the Customer Success team delivered monthly improvement proposals based on usage data. Aikawa added articles in real time: when responding to an inquiry, he searched Helpfeel first, and if an article didn't exist, he created it on the spot. This approach yielded 102 new articles in the first year. - Built adoption habits by promoting Helpfeel in department meetings, briefing sessions, and in-person training. The team leveraged digitally-native younger employees as early adopters, then shared their use cases to bring senior staff on board. Aikawa quietly promoted usage by advising engineers on search keywords during inquiries or including article links in email replies. - Measured success by tracking access volume, with a target of exceeding the number of traditional phone and email inquiries, proving more engineers were finding answers independently. - Maintained momentum through monthly Customer Success review meetings that created a productive cycle: data-driven proposals from the CS team, tasks completed before the next meeting. The team credited this external support for preventing Helpfeel from becoming another unused system. ## Results Monthly access volume climbed steadily from approximately 200 at launch to 1,000 currently, a more than 5x increase. The growing number of engineers independently resolving questions through Helpfeel became clearly evident. For service engineers in the field, "just search Helpfeel first" became habitual. Kenya Karita, a service engineer in Tokyo, now searches Helpfeel in advance and prints relevant sections to bring to customer sites. For critical issues that could shut down equipment completely, preparing thoroughly via Helpfeel before the visit means even unexpected events can be handled calmly. In one recent example, a product had a parts change and Karita didn't know how to wire it correctly during a job. He inquired with the responsible department and was told "it's posted on Helpfeel." He found an article with diagrams, far easier to understand than a verbal explanation over the phone. For Aikawa, the efficiency improvement was dramatic. Previously, answering a repeated question meant searching through email history to compose a response. Now he often pastes a link to the relevant Helpfeel article and replies immediately. His inquiry handling burden dropped approximately 80%. The benefits extended beyond individual efficiency. Knowledge silos began to dissolve, and the gap between urban and regional engineers started to close. Freeing up Aikawa's time from constant inquiry calls gave him capacity to focus on product development and improvement work. ## Looking Ahead Service engineers submitted survey feedback requesting coverage of more products. In response, Fujimak extended Helpfeel to products from Fujimak Neo, a group company handling development and manufacturing. Using Helpfeel Analytics, the team generated 200 articles from 1,000 inquiry logs, making the expansion efficient. The plan is to continue adding supported products so more engineers can respond smoothly in the field. The Service Promotion Department's goal is to eliminate knowledge silos entirely and equalize service engineer knowledge levels, so any engineer visiting a site can resolve malfunctions quickly. From a company-wide perspective, reducing the inquiry burden on product specialists frees their time for development work, contributing to business growth along two axes: improving service quality and strengthening development capability. As Fujimak expands internationally, Helpfeel addresses a critical challenge: overseas service engineers encountering questions in the field often can't get immediate answers from Japan due to time zone differences. With Helpfeel available, they can access the information they need regardless of time or location. As the company grows its overseas sales footprint, Helpfeel will be a powerful enabler. --- ## 80% of Users Found What They Needed: How Aiful Turned FAQ Reach into Customer Resolution: Aiful Corporation Source: https://en.helpfeel.com/case-studies/80-of-users-found-what-they-needed-how-aiful-turned-faq-reach-into-customer # 80% of Users Found What They Needed: How Aiful Turned FAQ Reach into Customer Resolution - FAQ reach rate of 80% (target was 40%) - No-hit rate of 3% (target was below 10%) - FAQ page views doubled to 400,000 in peak months ## The Challenge Aiful Corporation, a consumer finance provider with over 1,000 employees, faced a support volume challenge that came with growth: on busy days, phone inquiries alone reached 2,000 calls, with normal days averaging 1,000 to 1,500. On top of that, the team handled approximately 500 live chat inquiries and 100 emails per day, all with fewer than 30 operators. The root cause was clear. Over 90% of new customers completed their loan applications online, but when it came time to actually use the service, questions arose. The existing FAQ had been carefully written and regularly updated, but customers couldn't find the answers they needed. When self-service failed, they picked up the phone. Aiful had focused primarily on how FAQ articles appeared to customers, but that approach alone hadn't reduced inquiry volume. The team knew that if customers could search the FAQ more effectively and find answers on their own, it would reduce inquiry volume and lead to smoother transactions. The challenge was getting customers to the right information. ## Why Helpfeel What attracted Aiful was the approach to search itself. The ability to handle word variations and vague or imprecise search terms offered a fresh pathway they'd never seen before. The Helpfeel structure prepares multiple question pathways to help customers find the route that actually leads to resolution. The speed of search in the trial demo was genuinely impressive. Watching it return results for virtually any keyword gave the team a real sense that it would work. The support structure was also a deciding factor. Aiful needed a partner who could guide them beyond article content and into the entire customer journey to those articles. ## What They Did - Implemented Helpfeel's search-based FAQ system with a focus on mobile (97% of their FAQ users access from mobile devices) - Thoroughly reviewed not just article content, but the entire path customers take to reach the information they need - Built pathways within the FAQ that link from the smartphone app to relevant web pages where customers can find clear, visual information - Operators began using Helpfeel's search capabilities to quickly pull up the latest information while listening to customer questions, giving them confidence to respond with certainty - Participated in regular Customer Success review meetings where data analysis methods and keyword refinement were refined based on the psychological mindset of users behind their searches - Introduced Helpfeel's generative AI FAQ Draft Generation feature to improve the efficiency of article creation - Expanded FAQ operations to a cross-departmental effort, with other business units requesting articles to help customers self-resolve questions ## Results All KPIs significantly exceeded their targets. The article reach rate surpassed 80% from the very first month after launch in March 2023, doubling the 40% target. The no-hit rate came in at approximately 3%, well below the 10% target. FAQ article monthly page views increased from 200,000 to up to 400,000 in peak months, a 1.5x to 2x increase. Previously, customer satisfaction surveys showed only 3% of FAQ visitors found the content helpful. After Helpfeel, the team now has a real sense that articles are actually reaching customers. Operators handling inquiries also benefited. Being able to quickly pull up the latest information while listening to a customer's question gave operators the confidence to respond with certainty and direct customers to specific resources. The Customer Success partnership delivered concrete, data-based advice that fed directly into article improvements. Review meetings pinpointed pain points and made them visible in a concrete way, with proposals that were always persuasive and grounded in patterns observed across many implementations. ## Looking Ahead Aiful is working to turn FAQ reach into a measurable reduction in phone inquiry volume, with a near-term goal of a 2% reduction. The next phase is improving the quality and accuracy of the content itself. Beyond the FAQ, the team wants to identify confusing words, ambiguous terminology, and unclear expressions within the app, and surface those observations to the development team as recommendations. As that data accumulates, they plan to collaborate with departments capable of large-scale data analysis for further validation. The development team is also exploring whether Helpfeel could be applied not just to the FAQ but also to on-site search, with the hope that as the system becomes more robust, that kind of integration becomes feasible. --- ## 84% Fewer Inquiries in 10 Months: How SOKUYAKU Cut Support Costs in Half: SOKUYAKU / Jay Frontier Inc. Source: https://en.helpfeel.com/case-studies/84-fewer-inquiries-in-10-months-how-sokuyaku-cut-support-costs-in-half # 84% Fewer Inquiries in 10 Months: How SOKUYAKU Cut Support Costs in Half - Inquiries cut 84% in 10 months - Inquiry rate dropped from 40% to 8% of bookings - Call center costs reduced 70% through insourcing (7-8 outsourced staff to 4 in-house) - Overall CS costs down ~50% net of Helpfeel fees ## The Challenge SOKUYAKU is a one-stop platform for online medical consultations and prescription delivery, serving busy families across all 47 Japanese prefectures. As the user base grew, so did inquiries from patients and partner pharmacies. Call center labor costs ballooned in step with business growth, becoming a significant financial burden just as the company was transitioning from heavy launch investment to disciplined growth. The FAQ page already covered most common topics, but the search functionality was limited to keyword matching on article titles only. Users searched, couldn't find answers, and called anyway. Many inquiries were about topics already documented in the FAQ. Even after thoroughly revising the content, the team hit a hard ceiling: title-only search prevented meaningful improvement. The team needed a new FAQ system that could actually connect users to the answers already written. ## Why Helpfeel Helpfeel's search experience was fundamentally different. Instead of returning a single result or requiring an exact title match, Helpfeel displays multiple relevant questions when a keyword is entered. The "word hints" displayed beneath the search box surface commonly searched keywords proactively, preventing excessive fragmentation of search terms and guiding users to resolutions more quickly. The customer success support model stood out as a key differentiator: rather than leaving clients on their own after onboarding, Helpfeel provides data-driven, objective analysis and recommendations each month, grounded in cross-industry benchmarks. ## What They Did - Reviewed all existing FAQ articles with Helpfeel's customer success team, consolidating duplicates and merging related content - Restructured all inquiry pathways on the SOKUYAKU service site so every route to submitting an inquiry first passes through the Helpfeel FAQ page - Launched the FAQ within approximately one month, prioritizing speed to begin reducing basic operational inquiries immediately - Used the monthly no-hit list (queries that returned zero results) to iteratively add user search keywords directly into article titles, improving discoverability - Shared precision search data cross-departmentally to identify UI improvement opportunities and product feature gaps ## Results Inquiry volumes began declining immediately after launch. After 10 months, total inquiries had fallen by 84% compared to pre-deployment levels, driven by both Helpfeel's FAQ effectiveness and the redesigned inquiry pathway on the service site. The inquiry rate relative to consultation booking volume dropped dramatically: from 40% before deployment to just 8% at the time of the interview. With inquiry volumes sufficiently reduced, the company eliminated the outsourced call center for weekday operations entirely, replacing it with an in-house team. The outsourced call center had operated with 7-8 staff members; the insourced team now operates with 4 internal staff. This restructuring reduced call center operating costs by 70%. Even accounting for Helpfeel's licensing fees, overall CS operating costs were reduced by approximately 50%. The timing was strategically significant: SOKUYAKU had just moved past its initial investment-heavy launch phase and was entering a stage of cost-disciplined growth. Being able to right-size the previously inflated call center costs at this transition point was a major operational and financial achievement. The FAQ page launched in November 2023 and the insourcing was completed by July 2024, a transformation accomplished in under one year. ## Looking Ahead For a service like SOKUYAKU, the first-use experience is the decisive factor in whether a user continues. The team's goal is to make that first experience as smooth and satisfying as possible through continued UI improvements and ensuring that any questions arising during or after initial use can be resolved quickly through Helpfeel. Further reducing the proportion of inquiries that could be self-served, improving user satisfaction, and feeding search data into future product development are all priorities going forward. --- ## 85,000 New Users, Zero Increase in Inquiries: How Hokkaido Bank Scaled Digital Banking Without Scaling Support Calls: Hokkaido Bank, Ltd. Source: https://en.helpfeel.com/case-studies/85-000-new-users-zero-increase-in-inquiries-how-hokkaido-bank-scaled-digital # 85,000 New Users, Zero Increase in Inquiries: How Hokkaido Bank Scaled Digital Banking Without Scaling Support Calls - App user base grew by 85,000 in one year while inquiry volume held flat - Inquiry channel shifted from 100% phone toward web self-service - PDCA improvement cycle now driven from the contact center ## The Challenge Hokkaido Bank's digital banking success created a support challenge. With over 550,000 app users and growing, inquiries to the call center were rising in direct proportion. Customers often called simply because they couldn't find answers, saying "I don't know what to do, so I had no choice but to call." FAQs were scattered across individual product pages, and if customers couldn't navigate to the right page, they couldn't resolve their question. On peak days like Mondays and paydays, the phones were temporarily unreachable. The team knew many inquiries covered topics already documented in the FAQ, but information wasn't reaching customers. Despite this, they had no structure or expertise for improving the FAQ. Meanwhile, the app's features and services continued to expand, and inquiries grew more complex. ![A Hokkaido Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/hokkaido-bank_body.webp) ## Why Helpfeel When a group-wide website redesign presented an opportunity to standardize the FAQ across Hokkaido Bank and Hokuriku Bank, three factors drove the decision to choose Helpfeel: **High customer convenience.** Helpfeel's simple design and intent-prediction search could return the right information even when customers searched using a wide variety of words and phrases. **Migration support.** The team was uncertain how to migrate approximately 400 existing FAQ articles and had no internal expertise. Receiving support through the migration process provided essential peace of mind. **Ongoing operational support.** With no structure for FAQ operations, the team worried they might "implement it and then let it go stale." Continuous Customer Success support was essential to prevent that outcome. ## What They Did - **Built an operational structure around the launch.** Nishihigashi handles article planning and creation. Katsumi, who manages the call center, joined the effort to ensure front-line customer voices informed the FAQ. The entire call center team participated in migrating and improving articles, drawing on operators' deep knowledge of what customers struggle with. - **Focused on the no-hit rate as a priority KPI.** Whenever search log data revealed a gap where customers couldn't resolve their questions, the team treated it as a priority and addressed it immediately. - **Ran monthly PDCA improvement cycles.** Monthly review meetings with Helpfeel's Customer Success team provided data analysis that made improvement priorities clear. One memorable insight: "balance inquiry" was being searched frequently. After years in banking, the team assumed everyone could check their balance without trouble, but the data revealed meaningful customer confusion. - **Set up navigation from the app to Helpfeel early.** This decision ensured customers could self-serve directly from the context where questions arose. - **Shared insights across the organization.** The team circulated monthly review meeting materials to other departments, raising awareness of how the FAQ was being used and what challenges remained. This generated requests from administrative departments to add articles and improvements. ## Results Inquiries about basic functions, such as address changes, are trending downward, indicating customers are successfully self-serving through the FAQ. The ability to publish article updates and edits immediately has been transformative. With the old FAQ, even minor corrections could take around two weeks through an external production company. Now the team can publish changes the same day if needed. The operational benefits extend beyond metrics. Running the PDCA cycle gives the team a tangible sense that they're helping more customers resolve issues through the FAQ. Call center operators are contributing advice on how to improve articles, and seeing their input reflected in the FAQ has become a source of motivation. ## Looking Ahead Hokkaido Bank is redesigning its website with the goal of strengthening customer touchpoints across non-face-to-face channels. In an era where digital has become the norm, the website and app serve as a "non-face-to-face branch." The FAQ plays a central role in this vision, not just as a means of handling inquiries, but as a navigation function that provides detailed product and service information and guides customers through procedures. The team plans to design seamless connections from the FAQ to related functions such as account opening, transfers, and various update procedures. They're also exploring how to leverage AI and other technologies to improve operational efficiency and the quality of customer touchpoints, building what they call "evolving digital banking capabilities" with the FAQ at the center. --- ## 90% Fewer Human Inquiries: How Kirin Beer's Frontline Built an Internal Knowledge Hub with Helpfeel: Kirin Beer, Ltd. Source: https://en.helpfeel.com/case-studies/90-fewer-human-inquiries-how-kirin-beers-frontline-built-an-internal-knowledge # 90% Fewer Human Inquiries: How Kirin Beer's Frontline Built an Internal Knowledge Hub with Helpfeel - Human inquiries cut 90%, from ~240/month to ~30/month - 20,000 monthly searches, scaling support ~80x - Role transformed from inquiry handler to digital operations specialist ## The Challenge Kirin Beer's Business Support team was drowning in email. As employment styles diversified and internal systems grew more complex, the volume and difficulty of questions steadily increased. The COVID-19 pandemic accelerated the shift: casual hallway help evaporated, and all questions converged on a centralized "Concierge Desk." A 2.5-person team fielded approximately 240 to 250 emails per month, covering everything from HR policies to PC troubleshooting. The information already existed on the company intranet, but employees couldn't find it. A culture of "it's faster to just ask" took root. For Yuko Aiura, who staffed the desk, the psychological toll was severe. The constant standby mode, the compulsion to reply immediately, and the unpredictable arrival of each next question created mental strain. She proposed a direct solution: replace human effort with a system, so everyone could be happier. ![A Kirin Beer team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/kirin_body.jpg) ## Why Helpfeel Aiura evaluated FAQ systems and chatbots at industry exhibitions. Her top priorities were search speed and ease of use. Frontline salespeople work in a world where every second counts. A chatbot that requires multiple conversational exchanges doesn't match their pace. Helpfeel's instant search, where answers appear the moment you type, was exactly what the team needed. Given Kirin's scale, IT literacy varied widely across the workforce. The team deliberately kept less tech-confident employees in mind. A tool that only enthusiasts would adopt was not viable. The system had to be intuitive enough to operate without instruction, simple enough to be accepted by anyone. Helpfeel's single search box, combined with intent-prediction that handles vague phrasing and typos, met that non-negotiable condition. The ability to answer extremely rare, long-tail questions was equally critical. If an employee searches once and finds nothing, they won't try again. Confident that Helpfeel could win in that one-shot encounter, the team committed to implementation. ## What They Did - **Frontline-led ownership**: The team, rather than IT, drove the implementation, ensuring the system reflected deep firsthand knowledge of the work. - **Pre-built comprehensive content**: Multiple years of human inquiry logs, including one-time-only questions, were converted into articles before launch, prioritizing first-time success experiences. - **Visual identity**: Banners, icons, and color choices were customized to evoke the feeling of working at Kirin, creating an internal hub employees would feel affectionately toward. - **Pre-release with administrative staff**: A one-month pilot with several dozen admin staff across offices built the system together, incorporating feedback and creating internal advocates. - **Real-time content updates**: If information wasn't in the system, it was added no later than the following day. Corrections were sometimes made on the spot during meetings, demonstrating commitment through action. ## Results Monthly page views now average 20,000 searches. Before Helpfeel, human capacity capped responses at approximately 240 per month. The new system handles 20,000 searches per month, scaling support roughly 80-fold. Human-handled inquiries dropped approximately 90%, from ~240/month to ~30/month. The human-staffed Concierge Desk was closed, and all remaining inquiries were consolidated into a single form. Employees described Helpfeel as a "mental stabilizer." Sensitive topics like compensation and leave policies are difficult to ask a manager directly. Self-service search means employees who previously left uncertainties unresolved can now find answers themselves, contributing to an improved workplace environment. For Aiura, the change was transformative. Freed from the constant psychological burden of standby inquiry handling, she now spends time on creative work. Her motivation increased, and she went on to lead multiple other system implementations, earning internal recognition as "the person who knows systems." She transformed from inquiry responder to someone who solves problems through systems. ## Looking Ahead Helpfeel has embedded itself across the company. Initially targeting the sales division, usage has spread to approximately 2,000 employees, essentially all staff outside manufacturing. Expansion has been driven by other departments requesting that their own information be included. The good experience spreads internally, generating new content requests and use case consultations, which increases comprehensiveness and drives further usage, a virtuous cycle. The team is committed to continuously sharpening the freshness and accuracy of information, ensuring Helpfeel remains the natural first instinct. The goal is to establish Helpfeel as a robust information foundation where everyone is confident that "if you check here, you'll be fine," even through organizational changes and policy revisions. --- ## 90%+ Search Success Rate: How Aderans Turned FAQ Into a Psychological Safe Space: Aderans Inc. Source: https://en.helpfeel.com/case-studies/90-search-success-rate-how-aderans-turned-faq-into-a-psychological-safe-space # 90%+ Search Success Rate: How Aderans Turned FAQ Into a Psychological Safe Space - No-hit rate reduced to below 10%, clearing initial KPI - Routine inquiries decreased significantly, freeing capacity for complex consultations - Chatbot discontinued entirely, all effort redirected to FAQ operations - Search insights now feed LINE campaigns and marketing content creation ## The Challenge Aderans' Web Service Center existed to answer customer questions and help people feel comfortable enough to book in-store counseling. But routine inquiries about store locations and parking consumed the team's capacity, leaving little room for deeper consultations. A chatbot had been installed and rebuilt three times, each iteration requiring painful manual scenario mapping with no vendor support and no meaningful improvement. The deeper challenge was psychological. Many customers were too anxious to make contact at all. They worried about discussing hair loss topics on the phone, feared email contact would expose their address or invite sales pressure, and hesitated to ask about pricing in case they felt obligated to buy. For this segment, an FAQ allowing anonymous, frictionless self-service was not a convenience feature. It was the only viable touchpoint. The existing FAQ had high bounce rates. The team's energy was being consumed by system maintenance rather than customer-facing improvement. ![A Aderans team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/aderans_body.jpg) ## Why Helpfeel The company's president surfaced Helpfeel as a potential solution. The team saw it directly addressed their pain points: **No scenario scripting required**, in sharp contrast to the chatbot experience that had consumed months of effort across three rebuilds. **Strong search functionality** that could surface answers even from vague or colloquial queries, meeting customers where they actually were. **Ongoing Customer Success support**, seen as the most critical differentiating factor after years of struggling to improve without guidance. From inquiry to live operation took approximately 2 months, far faster than the team had expected based on prior system implementations. ## What They Did - Shared all existing FAQ and chatbot content with Helpfeel's Customer Success team at launch, who organized and structured it into usable articles - Built comprehensive foundational content first, after learning that conversion-focused articles failed when fundamental questions remained unanswered - Linked to relevant FAQ articles when responding to email inquiries, simultaneously resolving the immediate question and making customers aware the FAQ exists for future use - Used FAQ articles with diagrams and screenshots to answer complex questions the email system couldn't support with image attachments (e.g., cooling-off process illustrated with figures) - Analyzed Helpfeel search data and discovered customers searched for hair growth topics far more frequently than product topics, revealing a content gap that would never have been identified otherwise - Extracted high-frequency search keywords and shared them with the advertising department to create related content for distribution via LINE - Linked campaign emails to dedicated FAQ articles containing campaign details, allowing customers to access further information in a self-service format ## Results No-hit rate fell to below 10%, clearing the initial target KPI. Routine, low-complexity inquiries decreased significantly. Incoming contacts now more likely involve individual circumstances that genuinely require human handling or post-purchase consultations, indicating that basic information is being successfully delivered through the FAQ. Bounce rate on FAQ articles improved through data-driven content improvements in collaboration with Customer Success. The chatbot, which had been maintained in parallel, was discontinued entirely, with all effort redirected to FAQ operations. Marketing use of FAQ data expanded: search log keywords now feed into content creation for LINE distribution campaigns. Customer Success support enables improvements that would not be identified internally, including content gap analysis and article-level optimization. ## Looking Ahead The team's philosophy is that anxiety stems from unanswered questions. For Aderans' customers, many dealing with sensitive personal concerns around hair loss, the psychological barrier to reaching out is real. Some customers contact the company once, then reconnect 5 or 10 years later before finally deciding to visit a store. For these customers, having access to reassuring, comprehensive information on their own terms, without any pressure, is what makes the eventual decision to seek counseling possible. The FAQ is not framed as a tool for reducing operational costs, but as a way to gently lower the psychological threshold for a population of customers who need time, information, and privacy before they are ready to take the next step. Future plans include continuing to expand article coverage, reducing bounce rates on underperforming articles, extending FAQ data to inform more campaigns and content channels, and exploring broader applications for Helpfeel data across other parts of the business. --- ## Call Abandonment Cut to 1% While Serving 60,000 Staff: Bears' Internal Helpdesk Success: Bears Inc. Source: https://en.helpfeel.com/case-studies/call-abandonment-cut-to-1-while-serving-60-000-staff-bears-internal-helpdesk # Call Abandonment Cut to 1% While Serving 60,000 Staff: Bears' Internal Helpdesk Success - Call abandonment rate reduced to 1% monthly average - Call-in rate dropped from 8.95% to 7.6%, continuing to decline - 95.3% of staff aware of the FAQ; 73.3% have used it - Regional call center capacity freed to absorb overflow, no headcount added despite growing service volume ## The Challenge Bears Inc. is a home services company offering cleaning, housekeeping, childcare, and elderly care support, handling over 550,000 service sessions per year. These services are delivered by 60,000 directly employed housekeeping staff known as "Bears Ladies." The Staff Support Team was receiving over 300 inquiries per day: more than 100 phone calls and nearly 200 emails. Bears Ladies span a wide age range, with the core group in their 50s and some active members in their 80s. Inquiries varied enormously in urgency, from "I'm feeling ill and need to cancel tomorrow's service" to "I'd like to discuss next month's schedule." Simply sorting and prioritizing this volume created a significant workload on its own. Many inquiries had fairly standard answers, but the team struggled to enable self-service. They tried adding headcount, introducing a basic FAQ within the Bears Ladies app, and even launching a live staffed chat function. The chat failed: with only 3 to 4 staff members managing it simultaneously, response times were too slow, and Bears Ladies complained. As call volumes increased, phone lines became harder to get through. ## Why Helpfeel Knowing that many Bears Ladies, who skew older in age, were less comfortable with smartphone operations, the team understood that the ease and immediacy of a phone call would remain appealing. But phone capacity has a hard ceiling. Helpfeel's intent-prediction search stood out. Even with vague or imprecise input, the system surfaces relevant answers, making it well-suited to users who might struggle to formulate the "right" search keyword. The team felt this was the kind of tool that Bears Ladies would find usable and that would genuinely lead to self-resolution. ## What They Did - **Staff profile articles:** During the early post-launch period, uptake among Bears Ladies was slower than hoped, a common resistance to new digital tools in this demographic. In response, the team began publishing staff profile articles within the FAQ, introducing the internal staff members who interact frequently with Bears Ladies. These articles share each person's character, hire date, birthday, blood type, hometown, career history, reason for joining, and a personal message to the Bears Ladies. Some profiles include photos. The content has since been used to run a "Staff General Election" in which Bears Ladies vote for their favorite colleagues. The goal was to build familiarity and warmth toward the people behind the company, using that emotional connection as a gateway to Helpfeel engagement. In practice, the staff profile articles frequently rank among the most-viewed content on the FAQ. - **Monthly customer success meetings:** With such a large volume of inquiries, detailed keyword-level analysis is extremely time-consuming. The call center team often operates until around 9:00 p.m., making it nearly impossible to conduct thorough data analysis alongside day-to-day responsibilities. Helpfeel's customer success team provides analysis and recommendations at monthly standing meetings, a significant source of operational support the team describes as indispensable. - **AI draft generation:** The team is currently experimenting with Helpfeel's FAQ draft generation feature (which uses ChatGPT to create FAQ content from brief notes or keywords), aiming to further improve content creation efficiency. ## Results The call abandonment rate, calls that disconnect before reaching a staff member, has not yet reached zero, but after Helpfeel's introduction it dropped to a monthly average of approximately 1%. As more Bears Ladies resolve their questions through the FAQ, the call-in rate (calls relative to service volume) decreased from 8.95% in October 2022 (the month of Helpfeel's launch) to 7.6%, 7.7% at the time of the interview, and continues to decline gradually. A notable operational development: the call centers in Tokyo, Saitama, Chiba, and Kanagawa have seen enough reduction in incoming volume that freed capacity is now being used to absorb inquiries from the Kansai-region call center as well. Despite an increase in total service volume year-over-year, the call center headcount has not been expanded, a result the team directly attributes to Helpfeel. An internal survey (N=244) found that 95.3% of Bears Ladies are aware of the FAQ, and 73.3% have used it. Analysis of inquiry topics showed no heavy concentration in any single category, the FAQ is being used broadly across a wide variety of subjects. Selected comments from the internal survey: - "When I type a keyword I'm curious about, several related questions come up. It's very easy to understand." - "It's easy to ask questions, and it's handy even when I have a question at a customer's home. Very easy to use." - "Just typing one character brings up related articles. Very user-friendly." ## Looking Ahead With usage among Bears Ladies now growing, the next goal is to raise the self-resolution rate further. As inquiry data continues to accumulate, the team wants to analyze it deeply and use those insights to create a more stable, supportive working environment for Bears Ladies, and to proactively address needs before they turn into inquiries. Bears adopted Helpfeel from the perspective of improving employee satisfaction (ES). As labor shortages become an increasingly serious social challenge, ES will only grow in importance. The company particularly recommends Helpfeel to businesses in service industries and other sectors where securing and retaining human talent is critical. --- ## Call Center Abandonment Rate Cut from 15% to 2.5%: How Hokuriku Bank Built a Self-Service FAQ That Works Like Digital Customer Service: Hokuriku Bank, Ltd. Source: https://en.helpfeel.com/case-studies/call-center-abandonment-rate-cut-from-15-to-2-5-how-hokuriku-bank-built-a-self # Call Center Abandonment Rate Cut from 15% to 2.5%: How Hokuriku Bank Built a Self-Service FAQ That Works Like Digital Customer Service - Call abandonment rate dropped 83%, from 15% to 2.5% - FAQ articles reduced from 800 to 200 with better coverage - Customers now self-serve asset management inquiries in real time, reducing call volume ## The Challenge Hokuriku Bank's call center was overwhelmed. Abandonment rates hit 15% at peak times, meaning one in seven customers calling for help got no answer. Many of those calls were about information already sitting in the bank's 800-article FAQ or chatbot, but customers weren't finding it. The FAQ had no search function, and the chatbot was just a scenario tree. Updates happened twice a year, meaning the content was always behind real-time needs. Kurabayashi, who directs the bank's website and advertising, wanted more than an FAQ refresh. The goal was to turn the entire website into a "non-branch store" where customers could resolve questions instantly, online, without picking up the phone. ![A Hokuriku Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/hokuriku-bank_body.jpg) ## Why Helpfeel Hokuriku Bank was looking for an FAQ that could function like digital customer service, instantly resolving whatever customers needed. When Kurabayashi tried Helpfeel's demo, the intent-prediction search delivered exactly that: results changed in real time with every character typed, converging on the right answer faster than any chatbot the team had tested. Until then, the assumption had been that FAQs and chatbots needed to work together. Helpfeel changed that. The simplicity, speed, and breadth of financial services case studies reassured the team. What sealed the decision was Helpfeel's Customer Success support, responsive proposal process, and clarity around the long-term partnership. ## What They Did - Reviewed the existing 800-article FAQ with Helpfeel's team and migrated approximately 200 articles, eliminating duplicates created to cover different keyword variations - Eliminated the need for exact keyword matches, compressing the FAQ to one-quarter its original size while improving coverage - Focused post-launch operations on closing search gaps, adding and revising articles based on monthly data analysis - Established a monthly PDCA cycle with Helpfeel's Customer Success team to prioritize improvements and track progress toward zero call abandonment ## Results The call center abandonment rate dropped from 15% to 2.5%, and raw call volume decreased alongside the rate, showing customers were successfully self-serving through the FAQ. Increasing staffing contributed, but the data confirmed the FAQ was working. Beyond call reduction, data analysis revealed shifting customer needs. Searches for financial terminology spiked, driven in part by government promotion of NISA (Japan's tax-exempt investment scheme). Customers weren't just looking for NISA overviews, they wanted details: growth investment quotas, growth allocations, and step-by-step guidance for opening NISA accounts through the app. The bank responded by expanding FAQ coverage to match those needs. Operationally, collaboration improved across departments. Kurabayashi now prepares question text and sample answers when requesting FAQ updates from other teams, reducing their workload and speeding turnaround. Communication with the call center improved, too. Staff now suggest FAQ ideas based on inquiry trends, and teams across the bank understand that a good FAQ is worth building for customers. ## Looking Ahead Kurabayashi doesn't want Helpfeel to function purely as an FAQ. The next initiative is improving search depth, getting customers to browse beyond their initial question and use the FAQ as a broader source of information. One approach: featuring important announcements and campaign information directly in the FAQ. The bank's website already sees substantial traffic, and Helpfeel offers an efficient way to deliver timely information to customers. Building toward true digital customer service through the FAQ is the goal at the end of that series of improvements. --- ## Inquiry Rate Below 1% Despite 30% Growth: How Raccoon Financial Scaled Support with Search-First Self-Service: Raccoon Financial, Inc. Source: https://en.helpfeel.com/case-studies/inquiry-rate-below-1-despite-30-growth-how-raccoon-financial-scaled-support # Inquiry Rate Below 1% Despite 30% Growth: How Raccoon Financial Scaled Support with Search-First Self-Service - Inquiry rate cut 20%, from ~1.1% to ~0.9% - Member base grew 30% after launch - Inquiry rate stayed below 1% through rapid growth ## The Challenge Raccoon Financial operates Paid, a B2B payment service that handles invoicing, collections, and payment guarantees for over 4,500 companies. The entire application process is online, and both seller companies and their buyer trading partners have dedicated inquiry channels. Despite having a FAQ site, the team was fielding 2,500 to 3,000 inquiries per month. Inquiry volume grew in direct proportion to the member base, placing increasing strain on operators. The frustrating pattern: simple questions asked repeatedly, questions already answered in the FAQ that customers could easily resolve on their own if they had found the right answer. The root causes were clear. The FAQ site had low awareness, poor navigation pathways, and was genuinely difficult to use. Paper invoices listed a phone number as the contact channel, making it hard to guide customers from physical to digital. The most common question, "When will my invoice be issued?", stemmed from customers not understanding Paid's monthly billing cycle. When Hiroaki Tanaka, Business Development in the Paid Promotion Division, took responsibility for the FAQ in summer 2018, he discovered some content hadn't been updated since October 2016, nearly two years prior. Tanaka revamped the customer-facing portal to surface key information immediately upon login, but inquiry volume didn't change dramatically. The team needed to overhaul the FAQ itself. ## Why Helpfeel A colleague introduced Helpfeel to Tanaka at the end of 2021, just as he was gathering materials from other FAQ vendors. High build costs and ongoing running costs from those vendors had been obstacles. The internal discussion moved quickly, narrowing to a simple "do we adopt Helpfeel or not?" The decision was made to proceed. The core motivation was straightforward: Tanaka wanted more customers to actually use the FAQ site. If customers could resolve their own questions, inquiry volume would decrease and customer satisfaction would rise. That positive cycle was what the team was trying to create. ## What They Did - **Eliminated no-hit search results.** The Helpfeel Customer Success team extracted no-hit keywords and proposed content to address each gap, surfacing related search terms so customers' search needs could be met even when their phrasing wasn't exact. - **Emphasized search as the resolution path.** Tanaka observed from his operator experience that customers asking questions often "don't know what they don't know." The act of searching, pausing to ask "why don't I understand this? what exactly don't I understand?", is itself necessary for arriving at genuine resolution. - **Made the bold decision to hide FAQ categories entirely.** Initially, category navigation was displayed alongside search to accommodate the full spectrum of digital literacy across Paid's member base. But this conflicted with the core purpose of introducing Helpfeel: to drive resolution through search. A comment from Tanaka's manager dissolved all remaining doubt: "Have you ever seen someone get stuck trying to use Google search?" Google's homepage is nothing but the logo and a search box, and it functions as the universal search platform. The fact that Paid maintained robust phone support for customers who needed it allowed the team to feel comfortable with a search-first approach. - **Promoted the FAQ URL proactively.** The team began including the Helpfeel FAQ URL on paper invoices sent to members, bridging the physical-to-digital gap. ## Results The inquiry rate dropped from approximately 1.1% to approximately 0.9%, a roughly 20% reduction. Over the same period, the member base grew approximately 30%. The inquiry rate was held below 1% despite that growth. The nature of incoming inquiries shifted meaningfully. Simple questions, the kind easily answered by checking the FAQ, gradually declined. This reduced the workload on operators. The previous FAQ had inconsistent terminology and only matched exact-keyword searches. If a customer searched using a slight variation in phrasing or script, the search would return no results. The site was genuinely difficult to use, and the team lacked the confidence to actively promote it to customers. After introducing Helpfeel, the improved search capability made it possible to center customer support around the FAQ site. The customer support team described the change as "the shock of going from a feature phone to a smartphone." The UI and UX improvement was that dramatic. Raccoon Commerce, Inc., a group company that observed Raccoon Financial's success, also decided to adopt Helpfeel. Success in one entity spread organically to another within the group. The ongoing support from the Helpfeel Customer Success team, covering detailed adjustments and aspects the internal team doesn't have bandwidth to address in daily operations, was highly valued. ## Looking Ahead The goal going forward is to create an experience where customers feel no friction at all when using Paid. When reviewing inquiries in depth, patterns emerge: "if this is confusing, that probably is too." The aspiration is to have the FAQ site proactively address those connected follow-on questions, so that resolution happens entirely within the FAQ without needing to reach out. Tanaka noted a desire not to be bound purely by numeric metrics. Customer satisfaction surveys will always have a gap from reality, and most customers who have a positive experience with the FAQ will never say so explicitly. What he'd truly like is a tool that captures how customers feel while actually using the FAQ site. The Helpfeel Customer Success team has been doing a good job of picking up on those kinds of needs and blind spots that get missed in the rush of daily operations, and he hopes to continue that partnership in pursuit of genuine customer satisfaction. --- ## Internal FAQ Usage Grew 5.2x in One Year: How PALTAC Reduced Inquiry Load with Self-Service: PALTAC Inc. Source: https://en.helpfeel.com/case-studies/internal-faq-usage-grew-5-2x-in-one-year-how-paltac-reduced-inquiry-load-with # Internal FAQ Usage Grew 5.2x in One Year: How PALTAC Reduced Inquiry Load with Self-Service - Internal FAQ usage grew 5.2x in one year - QA testing after article creation became nearly unnecessary - Routine, repetitive inquiries dropped sharply ## The Challenge PALTAC Inc., a 126-year-old wholesale distributor serving drug stores, home improvement centers, supermarkets, and convenience stores across Japan, was scaling its digital transformation initiatives. As the company strengthened system capabilities and advanced digitalization, internal inquiry volume was trending upward. A team of just 3 to 4 people in the Information Systems Division was handling all phone inquiries about company-issued hardware and software, and the team was perpetually overwhelmed. Until 2018, frequently asked questions were managed in Excel spreadsheets, but the format couldn't capture the full range of inquiry content. Knowledge wasn't being shared effectively, leading to over-reliance on specific individuals, and similar questions were written up differently by different staff members, making it hard to identify duplicate issues. In 2019, the company introduced an AI chatbot. Despite high expectations, internal reception was poor. The chatbot used a single Q&A format with full-text search, requiring users to input exactly the phrasing the system anticipated. If they didn't, they got no answer. Initial usage averaged around 70 sessions per month, but the internal consensus quickly became "you can't find what you need" and "it's faster to just call." The more effort the team put into maintenance, the worse the return on that effort. The most common inquiry types included initial PC setup, internal system usage, and password-related questions. Despite setup manuals being available, most people called anyway saying they couldn't understand them. ![A PALTAC team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/paltac_body.jpg) ## Why Helpfeel While searching for a new approach after the chatbot's failure, the team discovered Helpfeel. With the long-term goal of consolidating inquiries from all departments into a single system, it was important that the tool be easy to use both as an operator and as an end user, even for non-IT departments. The key criteria were ease of article creation and maintenance, and a search function capable of finding information regardless of how a user phrases their query. Helpfeel's intent-prediction search, which surfaces results from a wide variety of input expressions, was the decisive factor. The intuitive usability that the team felt anyone could master sealed the decision. ## What They Did - Built up the article library gradually and added General Affairs content before making a company-wide announcement, reasoning that a spike in usage from novelty-driven curiosity would be meaningless if users weren't satisfied and never returned - Created a dedicated "PC initial setup manual" FAQ article with a link to the always-current manual, included in setup notification emails sent when issuing new PCs - Established an internal rule: when the same inquiry is received by phone twice, a Helpfeel article is created immediately (one occurrence may be rare; two occurrences establishes it as a pattern) - The General Affairs division, which receives inquiries from branch offices across Japan regarding attendance management, employment regulations, and form submissions, began creating their own FAQ articles in Helpfeel - Leveraged monthly check-ins with Helpfeel's Customer Success team to track no-hit rates, utilization rates, engagement rates, and article performance ## Results By July 2024, just over one year after launch, monthly usage (search sessions) reached 365, a 5.2x increase compared to average chatbot usage of approximately 70 sessions per month. Phone inquiries also dropped significantly. With the previous chatbot, verifying that AI-generated responses were correct required significant QA testing and subsequent tuning. With Helpfeel, the intended FAQ articles reliably appear in search results, making QA testing after article creation almost entirely unnecessary. Kikuchi, who joined the team only a few months before the interview, said he was concerned it would be complex to learn, but was editing articles successfully from his first day. The editing interface (Helpfeel Cosense) is intuitive enough that no programming knowledge is needed; learning basic conventions is sufficient. The PC setup manual approach raised organic awareness of Helpfeel across the company, since almost all employees interact with PC setup at some point. Because the FAQ article link remains stable even when the underlying manual is updated, there is no need to resend updated files. Routine inquiries are decreasing as the habit of "checking Helpfeel first" takes hold. Questions about digital seal usage (electronic stamps), which had previously been common, largely disappeared. Staff can now focus on phone inquiries that genuinely require human intervention, such as individual system errors. Feedback from the General Affairs division: "Helpfeel gives us a place to immediately direct people to answers, which has reduced the psychological burden of handling inquiries." ## Looking Ahead The target is to consolidate inquiries from all departments into Helpfeel within 2 years, and to grow internal FAQ usage to 10x the current level. The broader goal is to embed a culture of self-service problem solving across the entire company, which would further reduce inquiry volumes. The team plans to involve additional departments, expand the scope of Helpfeel coverage progressively, and continue improving article clarity by simplifying technical terms, adding images, and refining content to keep pace with growing usage. --- ## Peak Season Phone Surge Cut to One-Eighth: Sompo Himawari Life Insurance Makes FAQ the Core of Non-Voice Support: SOMPO Himawari Life Insurance Inc. Source: https://en.helpfeel.com/case-studies/peak-season-phone-surge-cut-to-one-eighth-sompo-himawari-life-insurance-makes # Peak Season Phone Surge Cut to One-Eighth: Sompo Himawari Life Insurance Makes FAQ the Core of Non-Voice Support - Peak season phone surge reduced to 1/8 of prior year (16% → 2%) - 540+ Q&A items managed with data-driven improvement cycles - 100% satisfaction with Helpfeel support structure - FAQ positioned as the core of non-voice customer service strategy ## The Challenge SOMPO Himawari Life Insurance manages approximately 4.7 million insurance policies. Like many companies facing shifting communication preferences and labor shortages, they were pursuing a "non-voice" strategy to shift customer service away from phone calls. They recognized FAQ as the core of this approach, but lacked the infrastructure to make it work. Despite having FAQ operations staff working on the same floor as the call center, positioned to quickly reflect the latest phone inquiry trends into content, improvements were ad hoc and driven by individual judgment rather than data. The team had no established system for continuously improving the FAQ based on metrics, and wasn't collecting sufficient data to form the foundation for improvement. The vision was clear: achieve stress-free self-resolution by minimizing user input effort and providing the right information proactively. Customers looking for the same information might search using different terms like "insurance premium" versus "contribution amount," and anyone, including those unfamiliar with insurance terminology, needed to easily reach the correct answer regardless of how they phrased it. The team concluded they needed a new FAQ tool with rich measurable indicators and strong support. ## Why Helpfeel The decision came down to usability for policyholders and features that would actually improve self-resolution rates. Key deciding factors were real-time access to metrics indicating whether customers were finding the FAQ after searching (exit rate after search) and whether they were having to search repeatedly (re-search rate), high search capability that returned results even for partial keyword matches and synonyms, and the availability of ongoing support to maintain a consistent improvement cycle after implementation. FAQ content accumulates over time, requiring not only updates to reflect the latest information but also constant verification of whether it's keeping pace with evolving search needs. At the company, FAQ managers rotate every few years due to job rotation policies. To ensure that improvement efforts would continue regardless of who was in the role, they needed to combine the knowledge of customer service veterans with regular data tracking and data-driven technical analysis and proposals. Helpfeel could provide all of this together in one package. ## What They Did - Revamped the policyholder-facing FAQ with approximately 540 Q&A items, combining content revised from the previous system with new additions - Established FAQ operations shared among seven members of the Planning Group plus several staff from other departments - Continuously refined FAQ design and features based on proposals from Helpfeel and requests from FAQ creators - Regularly adjusted "frequently viewed articles" and "frequently searched terms" sections at the top of the FAQ page based on recent access data, actual inquiry data, and mailing schedules - Proactively surfaced topics that policyholders were likely to ask about to encourage smoother self-resolution - Conducted monthly article reviews as a standard agenda item at meetings with the Helpfeel team, grounded in actual performance data ## Results Every year, when SOMPO Himawari sends out "Policy Content Notices" to policyholders so they can confirm coverage details, July sees a sharp spike in phone inquiries compared to the previous month. The prior year saw a 16% increase in call volume in July compared to June. The year after implementing Helpfeel, a similar increase of approximately 20% was anticipated, but the actual increase was only 2%. The peak season phone increase rate was reduced to one-eighth of what it had been the previous year. While the team doesn't track every single call's path prior to contact, the proportion of calls arriving via digital channels such as the website is known to remain relatively constant. Furthermore, the total number of policyholders who accessed the FAQ after receiving the mailed notice was approximately the same as the prior year. The team concluded that the main driver of the phone call reduction was customers reaching the answers they needed through improved FAQ search capability, resulting in higher self-resolution rates. By being an early adopter of Helpfeel's keyword and tap-based intent-prediction search, the company's FAQ is now positioned as offering industry-leading search capability. Category-based list navigation FAQs, where users search manually through organized lists, remain the industry norm across life insurance company websites. Internal FAQ operations have also been dramatically strengthened. Article reviews that previously occurred perhaps once a year at each individual's discretion are now conducted monthly as a standard agenda item, grounded in actual performance data. When Helpfeel proposed countermeasures for FAQ pages with high exit rates, the improvements became visible the following month, a fast improvement cycle that was exactly what the team had hoped for. ## Looking Ahead With Helpfeel, the company has finally been able to invest properly in FAQs, which had always been recognized as critical to non-voice goals but hadn't received sufficient attention. The plan going forward is to combine the FAQ with new initiatives to further accelerate the shift to digital channels, targeting a "non-voice rate" (the share of total customer contacts handled through channels other than phone, including chat and LINE) rising from the current approximately 40% to a sustained level above 50%. In the first year of Helpfeel, addressing the FAQ's newly visible weaknesses systematically produced broadly expected results. In year two, the goal is to move to a higher level, making more sophisticated decisions about what can be learned from user behavior in Helpfeel and what actions those insights suggest. The company is currently promoting adoption of "My Link Cross," a registered web service enabling policyholders to confirm contract details and perform various procedures online, with the goals of strengthening relationships with policyholders, enabling stress-free procedures, and improving the non-voice rate. As part of this, the company plans to use the FAQ to encourage registration for the service and support users as they navigate it. The team plans to continue using insights from call center interactions to refine FAQ content toward better self-resolution, working side by side with Helpfeel and verifying through performance data what content best enables smooth self-resolution. --- ## Phone Inquiries Drop as Online Accounts Grow: How Kirayaka Bank Scaled Digital Banking Without Scaling Support Staff: Kirayaka Bank, Ltd. Source: https://en.helpfeel.com/case-studies/phone-inquiries-drop-as-online-accounts-grow-how-kirayaka-bank-scaled-digital # Phone Inquiries Drop as Online Accounts Grow: How Kirayaka Bank Scaled Digital Banking Without Scaling Support Staff - Added 4,000 online accounts and ¥10B in deposits without increasing support capacity - Lower total phone inquiries in March 2024 vs. March 2023, despite account growth - 24/7 self-service outside business hours, evenings, and weekends for online customers - 200-article FAQ continuously refined through monthly data-driven improvement cycles ## The Challenge Kirayaka Bank, a regional bank based in Yamagata Prefecture, planned a full relaunch of its online branch in August 2023 to serve customers nationwide. The bank would be making first contact online with people who had no prior awareness of the institution, promoting fully online account opening, net-exclusive high-interest products, and 24-hour internet banking primarily via web advertising. The bank's call center and branch offices were handling approximately 2,400 phone inquiries per month from customers. The existing FAQ on the website had no system for tracking usage trends or making continuous improvements. As the industry workforce shrank (roughly 27,000 fewer bank employees nationwide over five years), Kirayaka needed a way to let online customers instantly access the information they needed at any time, without adding staff to handle the expected surge in inquiries from a growing nationwide customer base. ![A Kirayaka Bank team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/kirayaka-bank_body.jpg) ## Why Helpfeel The bank evaluated chatbots and other tools, but chose Helpfeel as their FAQ solution for three key reasons. First, existing article content could be reused immediately to get started without heavy rework. Second, Helpfeel keeps humans responsible for response accuracy, ensuring no AI-generated misinformation reaches customers. Third, the platform delivers monthly improvement proposals based on actual usage trends, making data-driven optimization straightforward. Helpfeel's intent-prediction search was particularly valuable for banking inquiries, where the same concept is often called by different names (for example, "registered seal," "personal seal," and "hanko"), or where terms are used ambiguously. The search could surface relevant topics even when the input keyword didn't exactly match. Confidence in the decision was reinforced by hearing directly from Sendai Bank, which belongs to the same group under Jimoto Holdings, that Helpfeel was being used broadly for deposit and loan inquiries and was delivering measurable results. ## What They Did - Launched Helpfeel in August 2023 alongside the nationwide online branch relaunch, starting with approximately 200 FAQ articles - Spent the first six months intensively addressing no-hit cases, where a user's search keyword returned zero results, refining content quality through ongoing deletions and additions - Implemented monthly improvements based on Helpfeel's data-driven analysis: total access counts, keywords with increased search volume, and specific recommendations - Deployed the "Element" pop-up FAQ feature in strategic locations across the bank's website, including the top page, to intercept inquiries before customers reached the phone - Used search history analysis to prioritize real customer needs, for example reordering one-time password articles to lead with device-change instructions after confirming with the call center that it was a common inquiry - Built cross-departmental collaboration by sharing Helpfeel insights with other divisions, including requesting new articles from the Administrative Operations Section on high-volume topics like magnetic stripe defects in cash cards - Coordinated a seasonal content strategy with Helpfeel, adjusting article prominence during high-inquiry periods such as October (when balance certificates for home loan deductions are mailed) and year-end before the new school year ## Results From the start of nationwide online account availability through May 2024, contracts grew by approximately 4,000 accounts and deposits exceeded ¥10 billion. Yet the weekday-only call center did not experience a backlog, and customers were not left waiting. The bank believes the vast majority of questions about online accounts are being resolved through Helpfeel, which is available at any time including evenings and weekends. Comparing total phone inquiry volumes across all bank locations in March 2023 (before Helpfeel) and March 2024 (after Helpfeel), the 2024 figure was actually lower, despite the net increase in online account contracts. Data-driven improvements yielded particularly strong results for customers using internet banking. For example, after analyzing search history and confirming with the call center, the team reprioritized one-time password articles to lead with device-change instructions rather than a general overview. When an article on recommended browsers was added and set to display prominently after the online branch relaunch, page views rose significantly, confirming the team was meeting a genuine customer need. Helpfeel also deepened cross-departmental collaboration. Other departments began requesting that new content be added to the FAQ, and the Digital Sales Division used Helpfeel as a fast stopgap to publish answers for confusing website content before the main site could be updated through the standard approval process. ## Looking Ahead Kirayaka Bank plans to continue enriching internet banking-related content, which represents the highest share of page views in Helpfeel today. The team will add more "Element" pop-up entry points at strategically effective locations on the bank's website to create an even smoother path to self-service resolution. As the bank invests in infrastructure to strengthen information sharing between branches, the team is exploring the use of Helpfeel as an internal FAQ, drawing on examples from other companies that have deployed it for employee-facing knowledge management. The goal is to build an internal structure capable of providing fast, consistent responses from any touchpoint, using Helpfeel to deliver smooth online information during the transition period as the bank expands its dedicated online channels alongside traditional branch and phone service. --- ## Scaling to 90,000 Investors Without Adding Support Staff: How THEO Serves Investment Beginners: THEO / Ookane no Design, Inc. Source: https://en.helpfeel.com/case-studies/scaling-to-90-000-investors-without-adding-support-staff-how-theo-serves # Scaling to 90,000 Investors Without Adding Support Staff: How THEO Serves Investment Beginners - 90,000+ investors supported by just 8 customer experience operators - 60 billion yen in assets under management with no support team expansion - 70% of users are investment beginners finding answers through self-service ## The Challenge THEO is a robo-advisor investment service built for everyone, particularly people with no investment experience. By March 2020, the service had grown to over 90,000 investors managing approximately 60 billion yen in assets. But the Customer Experience Division's 8-person call center was hitting a wall. The existing FAQ system wasn't working for THEO's audience. With 70% of users completely new to investing, they simply didn't know the financial terminology needed to search effectively. They'd look for answers using everyday language, while the FAQ was organized around industry-specific phrases. Searches came up empty. Users gave up and called in. The team tried a chatbot for six months. Instead of keyword search, it walked users through questions from the system side. In practice, most users dropped off before reaching an answer. Just as frustrating: the FAQ system made it nearly impossible to measure what was working. Usage reports had to be manually extracted into Excel. There was no way to iterate and improve systematically. The team needed a FAQ solution that could deliver high-quality support at scale, without scaling headcount. ## Why Helpfeel A supervisor's recommendation brought Helpfeel to the team's attention. After watching a demonstration video, Customer Experience Division's Masahiro Kato immediately saw the difference: the search algorithm was fundamentally different from conventional FAQ systems. Helpfeel could solve both problems at once. The search technology would help users find answers even when they didn't know the right industry terms. And the analytics platform would make FAQ effectiveness visible and measurable, enabling continuous improvement. The dedicated Customer Success team sealed the decision. Proactive improvement proposals, backed by real data, meant THEO could get better every month. ## What They Did - **Simplified the interface even further.** THEO stripped down Helpfeel's already clean default design, creating a smart, unambiguous UI where users are never left wondering what to do. The insight: users want one thing from a FAQ, to resolve their question quickly. - **Made content editing accessible to the whole team.** Under the old system, only Kato updated articles because the platform required careful technical handling. Helpfeel felt as casual as editing a text document. Other team members began naturally updating content on their own. - **Used monthly analytics reports to drive action.** Detailed usage reports with improvement proposals made it possible to quantify issues and act on them the following month. - **Partnered with a Customer Success rep who became an actual THEO user.** The Helpfeel CS representative registered as a real investor, reviewed customer-facing emails, and provided granular feedback on specific word choices from a user's perspective. ## Results Within a few months of going live, inquiry rate had declined compared to the previous period. Isolating Helpfeel's exact contribution was difficult because other initiatives were running in parallel, but the results were clearly better than what the chatbot had achieved. The design delivered exactly what the team hoped for: users knew what to do immediately. Operation was smooth. No points of dissatisfaction. The ease of editing had a secondary effect the team hadn't anticipated. Opening up content updates to the broader team meant the FAQ stayed fresher and more accurate, without bottlenecking on one person. Monthly reports from the Customer Success team made the FAQ a living tool. Quantitative data clarified where to focus. Qualitative feedback improved the details, down to individual word choices. Having someone so close to the actual user experience providing ongoing support proved invaluable. ## Looking Ahead At the time of the interview, THEO had only been using Helpfeel for a few months. But the long-term potential was already clear. Over time, accumulated data from user search keywords would enable increasingly precise understanding of what users need, driving more sophisticated improvements. The ultimate goal: zero inquiries. The aspiration is to multiply the user base 100-fold without scaling the call center at all, maximizing the customer experience entirely through self-service. --- ## Several Times the Normal User Volume, No Added Headcount: How Pococha's FAQ Survived the COVID-19 Surge: Pococha / DeNA Inc. Source: https://en.helpfeel.com/case-studies/several-times-the-normal-user-volume-no-added-headcount-how-pocochas-faq # Several Times the Normal User Volume, No Added Headcount: How Pococha's FAQ Survived the COVID-19 Surge - Inquiry rate dropped significantly from the first month of launch - Scaled to several times the normal user volume during COVID-19 without adding support staff - 4 distinct use cases across user support, talent agencies, internal knowledge, and creator resources - Users spontaneously shared the FAQ on Twitter, recommending it to other community members ## The Challenge Pococha, DeNA's live-streaming app with over 1.8 million downloads, operates at the speed of real-time community culture. The 10-person support team wasn't dissatisfied with their existing FAQ, but as they planned to scale the service, COVID-19 stay-at-home demand pointed to an imminent surge in new users. The team needed to improve support efficiency fast, enabling more users to resolve questions on their own. The highly active community, while energetic and positive, also meant misinformation could spread quickly. Support had to stay accurate and accessible at scale, without the option to simply hire more staff. ## Why Helpfeel What convinced the team was Helpfeel's ability to respond to the vague, conversational search terms real customers naturally use. Multiple team members, including producers, independently flagged Helpfeel in separate Slack channels after seeing it. During the detailed presentation, Helpfeel provided concrete metrics that FAQ systems should aim for and outlined a clear improvement roadmap. The ease of content editing sealed the decision. ## What They Did - **Migrated in approximately two months** with minimal internal resources. Helpfeel handled content migration and search keyword configuration. Pococha's main task was updating embedded FAQ links throughout the app. - **Refined wording with user-perspective language.** Core FAQ content carried over largely unchanged, but Helpfeel's team made targeted suggestions on phrasing to communicate more clearly from the user's viewpoint. - **Expanded to four distinct use cases** after the initial user-facing FAQ deployment: a knowledge hub for top streamers ("Ask a Senior Liver"), a dedicated FAQ for talent agency staff (replacing PowerPoint documents), and an internal knowledge base (migrating from spreadsheets to Scrapbox, the document service underlying Helpfeel). - **Embedded FAQ links throughout the app** to surface answers at the moment users needed them, reducing friction and improving discoverability. ## Results Results appeared immediately. From the first month, the inquiry rate dropped significantly while FAQ page access increased simultaneously. Internal feedback was consistently positive, highlighting strong search performance and ease of editing. When COVID-19 drove stay-at-home behavior, user numbers grew to several times the previous level, a far larger surge than originally anticipated. Despite this, the number of inquiries did not increase proportionally, and the support team was able to manage the volume without any additional hiring or capacity expansion. Users began voluntarily sharing Helpfeel FAQ links on Twitter, a sign that the FAQ had become genuinely useful enough to pass along. Pococha's highly engaged community of users who want to help spread accurate information about the app's proper use and rules amplified the behavior of turning to the FAQ first when questions arose. ## Looking Ahead Pococha will continue to evolve without fear of change, responding to the ever-shifting needs and communications of its users. Because the app itself operates in real time, support must also function in real time. Helpfeel has enabled the team to quickly collect information about what users are wondering and rapidly produce accurate responses. The team intends to continue using Helpfeel to accelerate the speed of user support and deepen the penetration of Pococha's community culture. --- ## Turnover Improved 2.7%, 60 Hours of Overtime Eliminated: How Nagai Hospital Transformed Operations with AI: Nagai Hospital Source: https://en.helpfeel.com/case-studies/turnover-improved-2-7-60-hours-of-overtime-eliminated-how-nagai-hospital # Turnover Improved 2.7%, 60 Hours of Overtime Eliminated: How Nagai Hospital Transformed Operations with AI - Turnover rate improved 2.7% year-over-year - HR overtime reduced by ~60 hours annually - 87% of staff aware of Helpfeel within one year ## The Challenge Since its founding in 1947, Nagai Hospital has served as a regional secondary emergency care hospital in Tsu City, Mie Prefecture. Over the past decade, the hospital expanded rapidly, growing from 400 to over 500 employees. Annual emergency admissions doubled over five years to approximately 3,000 cases. With both staff and patient numbers growing sharply, frontline and back-office operations were approaching their limits. In HR, three staff members managed labor administration for 500+ employees while fielding constant inquiries about salaries and policies. On the clinical side, frequent questions from new and mid-career nursing staff interrupted workflows. The hospital's many proprietary procedures and internal abbreviations meant even experienced staff needed time to become fully familiar with them. As a private hospital in a city without a municipal facility, Nagai Hospital bears responsibility for regional healthcare. Remaining a cornerstone of regional care required operational reform across both frontline and back-office functions. The hospital launched its medical DX project "NAGAI 100" in 2024, and a cross-departmental sub-committee identified "questions and inquiries" as a significant drag on efficiency. The team needed a tool enabling unified access to operational knowledge and institutional policy information. ![Nagai Hospital](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/nagai-hospital_body.jpg) ## Why Helpfeel The hospital initially considered a chatbot to answer staff questions, but Helpfeel's intent-prediction search, which surfaces the right information even when search terms vary slightly, was seen as ideal for a hospital environment filled with medical terminology and internal shorthand. The team needed a solution that could: - Consolidate knowledge and improve access for 500+ staff across departments - Reduce new employee training workload and the psychological burden of repeatedly asking the same questions - Cut HR inquiry volume related to leave entitlements and employee benefits - Support frontline staff with instant access to procedures and protocols Helpfeel's ability to handle variation in terminology and provide instant answers from smartphones met all these requirements. ## What They Did - **Built content aligned with actual demand.** Initial adoption was slow when the team focused on employment regulations. A staff survey asking "What information would be useful to you?" surfaced concrete frontline needs, including strong demand for infectious disease response protocols. - **Ran a creative awareness campaign.** An employee suggested creating an original mascot character, "Fiiru Gorira" (a play on "Helpfeel" and "gorilla"), with its own chat account. The character regularly posted tips about Helpfeel's features in the company chat tool, building familiarity and interest. - **Timed deployment with company-wide smartphone rollout.** The hospital introduced smartphones and a new chat tool for all staff, creating an opportunity to promote Helpfeel alongside the new infrastructure. ## Results Within one year, 87% of staff knew about Helpfeel, and 75% rated it as "easy to use." In nursing operations, the reduction in training-related time costs was substantial. Head nurses previously fielded repeated questions from junior staff about examination procedures and clinical protocols. Now, staff look up procedures instantly on Helpfeel from their smartphones, and the frequency of such questions has dropped sharply. Helpfeel reduced not only operational inefficiency but also the psychological burden on new staff during onboarding. While it is natural for new employees to ask their seniors for guidance, having to ask the same question multiple times creates discomfort. In terms of retention, the 2024 turnover rate improved by 2.7% year-over-year, a result the team attributes in part to Helpfeel's contribution to a more supportive and less stressful working environment. Helpfeel also contributes to the quality of nursing care itself. With Helpfeel accessible at the bedside, nurses can answer patient questions immediately and confidently, delivering reassurance and value in the moment of care. In the back office, inquiries related to leave entitlements and employee benefits declined. HR department overtime in fiscal 2024 decreased by approximately 60 hours per year compared to the prior year. From a management perspective, a significant mindset shift occurred. Staff who had accepted analog processes as unavoidable began to feel that "existing tasks could actually be easier." This consciousness now extends beyond digital tools to encompass improvements in workplace conditions and institutional systems more broadly. The experience of successfully transforming operations through autonomous digital change became an important motivational foundation for further DX initiatives. Helpfeel's introduction was, in the director's words, "not an exaggeration to call the template for our medical DX." ## Looking Ahead While reducing staff burden through operational efficiency is one goal, the deeper ambition is to enable richer and more attentive support for patients and their families. Routine and inefficient tasks can and should be reduced through digital tools. However, the communication between nurses and patients, the human presence that addresses anxiety, supports treatment, and maximizes the effectiveness of rehabilitation, can never be replaced by digital tools or AI. The authentic value of healthcare and nursing lies in human-to-human connection. But providing that value requires that staff have sufficient physical and mental capacity to do so. The goal going forward is freeing up time and energy so that more of both can be devoted to the care of patients and their families. --- ## 70% Fewer Inquiries at Peak Season: The CX Strategy Behind MUJI's Growth: MUJI (Ryohin Keikaku Inc.) Source: https://en.helpfeel.com/case-studies/70-fewer-inquiries-at-peak-season-the-cx-strategy-behind-mujis-growth # 70% Fewer Inquiries at Peak Season: The CX Strategy Behind MUJI's Growth - Inquiries cut 70% during peak months - Online active users grew to 150%+ year-over-year, with inquiry volume held flat - FAQ publishing time cut from up to 2 weeks → instant MUJI sells everything from a $0.50 eraser to furniture, apparel, and food across 1,300+ stores worldwide. As omnichannel sales grew, inquiry volume outpaced what a human-staffed support team could absorb, and many online shoppers who wanted to self-serve were still forced to call. ## The Challenge The legacy FAQ was owned by a separate EC division, so updates took up to two weeks to go live and routinely missed the seasonal moments customers needed them. Search returned articles in reverse-chronological order, burying the most relevant answers. Staff couldn't trust it either, driving repeated escalations to product managers. ## Why Helpfeel A cross-functional review with the EC and IT teams set two non-negotiables: the Customer Relations Office had to manage the FAQ independently (no engineering tickets), and search had to reliably surface the right answer. Helpfeel met both. ## What They Did Built a unified content guideline and rewrote every article to a consistent standard during migration. Prioritized new articles by sales volume and inquiry frequency, starting with the highest-traffic categories. Embedded FAQ links directly on high-velocity product pages (e.g. health & beauty). Enabled real-time seasonal content: sunscreen FAQs surfaced in early spring, live the same day instead of two weeks later. ![MUJI booster skincare products on a clean studio surface](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/muji_body.jpg) ## Results During the October "MUJI Week" member campaign (historically a period of sharp inquiry spikes), sales grew while inquiries stayed suppressed, eliminating a long-standing timeliness problem. FAQ bounce rate fell, new-hire onboarding accelerated (new staff resolve inquiries by referencing the FAQ directly), and the Customer Relations Office's internal profile rose sharply. ## Looking Ahead MUJI's north star is zero inquiries (letting customers shop without ever needing to stop and ask), plus FAQ experiences customers return to proactively for genuine added value. --- ## A 40% Lift in Self-Resolution: Using FAQ Data to Make the Product Better: ROLLCAKE Inc. Source: https://en.helpfeel.com/case-studies/a-40-lift-in-self-resolution-using-faq-data-to-make-the-product-better # A 40% Lift in Self-Resolution: Using FAQ Data to Make the Product Better - Self-resolution rate improved ~40% after implementation - No-hit rate reduced from ~6% to 2-3% - Response time cut from ~2 business days to significantly faster ROLLCAKE runs three consumer subscription services: ALBUS (a monthly photo-print album app), Letter, and FLOWER. A pre-implementation analysis found that ~40% of all inquiries were self-resolvable if the FAQ were easier to search. But small phrasing changes broke the old search entirely, and a trial AI chatbot was abandoned for surfacing irrelevant articles. Response times hit ~2 business days, and adding headcount wasn't a scalable answer. ## Why Helpfeel Search quality was decisive: after reviewing Helpfeel on other companies' sites, the team trusted it to match colloquial queries ("can't do X," "Y hasn't arrived") to the right article. Guaranteed Customer Success support was the other factor: they'd been running the FAQ on intuition, not data. Migration was essentially a link swap. ## What They Did Distinguished "things Helpfeel should resolve" from "things that need a human," running a deliberate dual-track model. Tuned keywords from Customer Success analysis, dropping the no-hit rate from ~6% to 2-3%. Fed CS insights into the product: a post-order address-change feature and in-app notifications that keep users from getting lost mid-flow. ## Results Self-resolution rose ~40% and human-handled contacts now arrive with context (the user already tried the FAQ), making each exchange more efficient. Helpfeel's quantitative monthly reports ("this many users hit this problem") are shared at full-team meetings, making it easier for engineers to prioritize fixes. --- ## A 47% Drop in Inquiries (and 2 FTE of Savings) at a Femtech Subscription: mederi Inc. Source: https://en.helpfeel.com/case-studies/a-47-drop-in-inquiries-and-2-fte-of-savings-at-a-femtech-subscription # A 47% Drop in Inquiries (and 2 FTE of Savings) at a Femtech Subscription - Inquiries reduced 47% within one year of deployment - Labor cost savings equivalent to 2 full-time employees - ~460,000 registered users on its messaging channel mederi runs women's-health subscription services including "mederi Pill," an online consultation service where OB/GYNs prescribe contraceptives. Rapid growth to ~460,000 users in roughly three years drove inquiry volume up in tandem, straining a three-person support team. The prior FAQ also required routing every edit through an outside web vendor, making timely campaign updates impossible. ## The Challenge The team also saw inquiry demand itself as a UX signal: if a user had to contact support, that was a gap to close. ![mederi low-dose pill packaging](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/mederi_body.jpg) ## Why Helpfeel Primary criteria were non-engineer editability and high end-user usability. Helpfeel's editing interface and its as-you-type search experience made it the clear pick. ## What They Did Replaced jargon with patient-friendly language across medically supervised articles (every article reviewed by internal pharmacists and physicians). Expanded sensitive health-topic coverage so anxious users can self-resolve without disclosing their name. Routed the inquiry funnel through the FAQ first: users about to submit a question pass through relevant articles, a structural change cited as a major driver of reduction. Tracked customer-satisfaction KPIs (CSAT, first-contact resolution, response time) rather than a blunt zero-inquiry target, appropriate for a service where some contact is healthy. ## Results Inquiries fell 47% in a year, the inquiry rate dropped from a 5-7% baseline, and the team realized ~2 FTE of labor savings. Recurring payment-error contacts were cut by pairing a new FAQ article with a link inside the automated payment-error email. Inquiries that still arrive are treated as "treasure": VoC fed cross-functionally into product changes like a MyPage redesign. ## Looking Ahead mederi plans to integrate Helpfeel with other systems for more personalized delivery, with the FAQ serving as a steady support pillar that helps users take their medication with confidence. --- ## A 61% Drop in Inquiries: Alpen's EC Growth Strategy: Alpen Inc. Source: https://en.helpfeel.com/case-studies/a-61-drop-in-inquiries-alpens-ec-growth-strategy # A 61% Drop in Inquiries: Alpen's EC Growth Strategy - Web-form inquiries reduced 61% - Self-resolvable inquiries reduced 78% - Overall support operating costs reduced ~10% - Inquiry rate (inquiries ÷ orders) down 2 points Alpen is one of Japan's largest sporting-goods retailers, with a loyalty program approaching 15 million members and a medium-term plan to more than double owned-EC revenue. But as orders grew, inquiries grew in lockstep, and unlike marketplace selling, Alpen handles every payment, shipping, and navigation question in-house. ## The Challenge The only self-service tool was a text FAQ with basic browser keyword search that customers clearly weren't reading, and every edit required an external vendor request. ![Alpen](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/alpen_body.jpg) ## Why Helpfeel After a year evaluating chatbots without conviction, the team pivoted to AI-powered FAQ. Across competing AI-FAQ tools, Helpfeel's search precision (reliably guiding customers to the correct answer) was the decisive factor, with leadership satisfied by the ROI case. ## What They Did Made inquiry rate the primary KPI and fed customer feedback into FAQ content continuously. Strategically removed direct inquiry-form links from product pages, replacing them with links into the Helpfeel FAQ. Embedded FAQ links throughout the journey (product pages, cart screen) to resolve questions on the spot without cluttering the buying flow. Pinned a "Shopping Guide" category (payment, shipping, cancellations) in the first visible area. Ran monthly Customer Success meetings to interpret analytics and brief internal stakeholders. ## Results Form inquiries fell 61% and self-resolvable inquiries fell 78%, a number so dramatic the team initially suspected a measurement error. Combined with cost restructuring, support operating costs dropped ~10%, while FAQ page views rose, confirming active usage. --- ## A 95% Drop in Campaign Inquiries: Standing Up Support at a Fast-Growing Marketplace: Kauche Inc. Source: https://en.helpfeel.com/case-studies/a-95-drop-in-campaign-inquiries-standing-up-support-at-a-fast-growing # A 95% Drop in Campaign Inquiries: Standing Up Support at a Fast-Growing Marketplace - Campaign-related inquiries reduced from ~100 to 5-6 per campaign (about 95%) - Response time cut from up to 5 business days to a fraction of that - Inquiry mix shifted from basic how-tos to genuine product-improvement signals Kauche is a group-buying shopping app where users buy jointly (with friends, family, or strangers via social media) to unlock discounts (700,000+ downloads by May 2022). When Haruka Sato joined as the first dedicated CS hire, there were no response templates and contacts took up to 5 business days. The FAQ had gone stale with no capacity to maintain it. Worst of all, every campaign reliably triggered ~100 identical "how do I join?" inquiries, creating predictable dread before each launch. ## Why Helpfeel For a diverse consumer base, the same question comes in many forms: a user wanting to close their account might search "withdraw," "cancel," "deactivate," or "delete." Helpfeel's intent-prediction search handles those variants flexibly. For a small team with no FAQ expertise, the proactive monthly Customer Success meetings were the other deciding factor. ## What They Did Built dedicated FAQ pages for each campaign type (a Customer Success recommendation). Used monthly CS sessions for analytics review, no-hit keyword tracking, and prioritized action, rallying the team around an "eliminate all no-hit words" goal. Published a Vietnamese-language FAQ subset after data surfaced a spike from a Vietnamese influencer's promotion, laying groundwork for future international expansion. ## Results Campaign inquiries collapsed ~95%, and pre-campaign dread turned into confidence. With basics handled by self-service, remaining inquiries became real product feedback, now shared directly with engineers and designers, turning support into a product-development input channel. The CS team grew from 1 to 3 with a positive, improvement-oriented culture. --- ## Efficient self-service at LIXIL: Leveraging VOC Data for Continuous Help Improvement: LIXIL, Inc. Source: https://en.helpfeel.com/case-studies/boosting-efficiency-at-lixil-with-helpfeel-leveraging-voc-for-continuous-improvement # Efficient self-service at LIXIL: Leveraging VOC Data for Continuous Help Improvement LIXIL develops and provides plumbing and home products such as toilets, bathrooms, and kitchens, as well as building materials like windows, doors, interior surfaces, and exterior surfaces.LIXIL's Customer Service Department handles contacts about repairs and product questions after purchase and carries out improvement activities using VOC (Voice of Customer) data to ensure the LIXIL brand is top-of-mind for customers.The Customer Service Department faced the challenge that consumers and construction companies installing their products could not solve their problems through the help center, causing their contact center to be overwhelmed with phone calls.To address this issue, Helpfeel was introduced in 2022. The number of phone calls has decreased, and the speed of creating help pages has improved. We spoke to Mr. Takano, Mr. Nagasaki, and Mr. Kosui from the customer service department about the background of introducing Helpfeel, how it's being used today, and the specific benefits they're seeing. ## Seeking a Customer-First Help Center That Reduces Burden on End Users Mr. Takano, General Manager, Service Innovation Department, Customer Service Department ──Can you tell us about the challenges in the Customer Service Department before introducing Helpfeel? Takano: The team I manage is on a mission to improve the NPS (Net Promoter Score) for the LIXIL brand. We consider LIXIL's customers to be the retail stores that sell our products, the construction companies that install them, and the end users who actually use them. However, our connection with end users was not as strong as with the other types of customers.With the advance of digitalization in recent years, the environment has allowed end users to choose products they like by searching for them, shifting the trend towards end users choosing products themselves rather than just buying the fittings included with a house.In this environment, to have our products chosen, we needed to be more proactive in our approach to end users than ever before, while also being mindful of customer retention. We thought we needed to strengthen our service measures to achieve this.Nagasaki: As Takano said, with this background, we first conducted VOC analysis based on the opinions of end users received at the Contact Center and created help pages tailored to customer needs to improve the self-service rate. However, the search feature of the help center we were using at the time was poor, and customers could not find the information they wanted. Even when customers entered words into the search box, no articles were displayed, and they had to change the words and search again and again.Kosui: On the other hand, there were many cases where dozens of help pages were displayed, and customers often didn't know which article to look at, so I felt that we were placing a burden on our customers.Takano: If customers can't solve their problems using the help center, they will contact us. As a system-related issue, there was a problem with the uneven distribution of workload among members. Because of the history of our company, which was formed through a merger to become LIXIL, the people in charge of receiving contacts were only familiar with specific products. As a result, when we received complex contacts, we had to rely on specific members to deal with them, leading to a situation where the work became personalized and the workload was unevenly distributed among specific members. Mr. Nagasaki, Leader, VOC Improvement Promotion Group, Service Reform Promotion Department, Customer Service Division Nagasaki: I thought there were two main reasons for these problems. One was that the system we were using at the time lacked a good search engine. The other was that we didn't have much SEO know-how, and I think we weren't able to create articles that would allow customers to find the right information when they searched. ──Were there any other issues with the system you were using at the time? Takano: The system had many features, but we weren't really making good use of them. In fact, there were so many features that we ended up giving up on using them halfway through the operation. Instead of having too many features that we couldn't make good use of, we decided to focus on searchability and convenience for customers and began looking into replacing the system with a different one. ──Please tell us about the background to introducing Helpfeel as your new help center. Ms. Kosui, VOC Improvement Promotion Group, Customer Service Department Kosui: We mentioned Helpfeel as one of the candidates for implementation and approached them in the fall of 2021. We were looking for a system with good search functionality. We wanted to create a system that would enable customers to accurately narrow down the information they were looking for when searching.Takano: The main reason we decided on Helpfeel was that, in addition to its excellent search functionality, it was a system that specialized in what we wanted to do, considering that our company has a wide range of products and that the contacts we receive come from a diverse range of sources, including construction companies and end users. Also, we had previously used a help center that was packed with too many features and was too difficult to use, so we had high expectations that Helpfeel would be easy for our members to understand and use smoothly. ![A LIXIL team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/lixil_body.jpg) ## The speed of updating help pages has greatly improved. Data analysis by Customer Success enables the PDCA cycle to run smoothly. ──Were there any particular things you tried to do when you introduced Helpfeel? Takano: At the same time as we introduced Helpfeel, the Service Innovation Department also introduced Scrum as a framework. Until then, we had divided up the roles within the organization and the members had stuck to the roles they had been given, but we moved to a system where each member could work independently. Nagasaki: The launch of Helpfeel was one of the missions of Scrum. I think that the fact that the UI was easy to use and had a low barrier to entry meant that even though we changed the system, there was no major confusion, and the members were able to grow quickly. Looking back now, I feel that Helpfeel and Scrum were a good match.Kosui: In our day-to-day work, we try to publish help pages as quickly as possible. In the past, it sometimes took two months to publish a single article, and we spent a lot of time checking each word.Now, we are conscious of the need to get new content out quickly, and we publish articles within about a week of starting work on them. We have also been able to speed up the VOC analysis process for creating article drafts. One of the reasons for this speedup is Helpfeel's knowledge base, which allows multiple people to edit articles simultaneously. We can now add or update 50 articles, including revisions, as we can finish revisions on the spot while having meetings. I feel that we are constantly renewing our help center. ──Can you tell us your thoughts on Helpfeel's customer success team? Kosui: Thanks to the support of the technical writers, the quality of the help pages has improved, and you also provide us with detailed data analysis on a monthly basis. We are grateful for the figures you have highlighted for us.Takano: We were able to gather a lot of data using the previous system, but because the data was so varied, we didn't know where to start with analysis and as a result, we weren't able to take any action.Now, we are very grateful that Helpfeel has helped us to identify which data we should be looking at in the first place, and have helped us to narrow down the important data and set KPIs that will lead to improvement. You have created a situation where the PDCA (Plan Do Check Act) cycle is working properly. ## Received the LIXIL Award for Energy Conservation and Productivity Improvement ──It's been about a year since you implemented Helpfeel. What kind of effects have you seen? Kosui: The effects are clearly visible. The biggest result is the bounce rate. The bounce rate immediately after introducing Helpfeel was 43%, but it has now improved to below 37%. Takano: The search usage rate has also increased, and the number of phone calls has decreased by 5-7%. I think there are environmental factors at play, such as the fact that customers are now looking at websites more than before due to the coronavirus, but even so, many people are now able to look at websites and resolve their questions before making a phone call. As a result, the staff who receive contacts are not as busy as they used to be. ──What are your thoughts on Helpfeel overall? Nagasaki: There are few cases where people get stuck with the operation, and even new members who have just arrived due to personnel changes are able to use it immediately. In fact, a member who transferred from another department last year is already making the most contributions to help page creation.It has also been well received by other departments within the company. Our products sometimes use difficult terminology to explain their uses and features, and I think customers appreciate the fact that even if they search for a feature using different terminology, the appropriate information is displayed in the search results. There have even been calls to display a pop-up search window on other pages too.And this year, the Customer Service Department received an internal award called the LIXIL Award. I think the award was given in recognition of the fact that we were able to reduce the number of calls and cut back on manpower, and that our productivity had improved. ## Aiming to Provide New Value, By Guiding Customers to Sales Pages and Building Up Knowledge, Rather Than Just Solving Problems ──Please tell us about your future plans. Nagasaki: I'd like to provide more value than just solving customers' problems. We're currently in the process of directing customers to our showroom and sales sites in our help pages, to encourage repeat use. I'd like to contribute to the business in this proactive way, for example by responding to demand for e-commerce sales and home renovations.Takano: In the medium to long term, we would like to make our knowledge more open. We think it would be good if we could publish as much information as possible, including knowledge for professionals who install the products, so that customers, construction companies, and distributors can all see the same information and solve their problems.If this is achieved, all parties involved, both inside and outside the company, will be able to see the same information, so we can expect a reduction in the number of man-hours spent by sales staff and the number of individual contacts, which will also lead to cost reductions. ──Finally, please share a message to companies considering introducing Helpfeel.Nagasaki: Helpfeel is not only an easy-to-use system, but the included support via customer success is excellent. We are able to improve with the advice of Helpfeel's experts, and I think you can expect to see results that exceed the investment you make. --- ## Cancellation Inquiries Down 80%: Hamee's Search-Log-Driven EC Support Model: Hamee Inc. Source: https://en.helpfeel.com/case-studies/cancellation-inquiries-down-80-hamees-search-log-driven-ec-support-model # Cancellation Inquiries Down 80%: Hamee's Search-Log-Driven EC Support Model - 40% self-resolution rate within one month of launch - Scaled through 4-5x sales growth with no added support headcount - Cancellation inquiries cut 80% (≈200/month → ≈40) - Inquiry rate dropped from 5-6% to 4%; return-shipping costs eliminated Hamee builds and sells consumer brands end-to-end, including the iFace smartphone-accessory brand and Pixio gaming monitors, with a US-market global business and a Gen Z focus. When a fan-culture boom and expanded color lineup pushed gaming-monitor inquiries from 100-200 to ~1,000 a month, a 1-2 person team started missing its same-day-response KPI. ## The Challenge For phone cases, the huge range of compatible models made wrong-model purchases common, generating ~200 cancellation inquiries a month. ![Hamee](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/hamee_body.jpg) ## Why Helpfeel Search that resolves vague, casual queries. Ongoing post-launch improvement proposals. Low lift: Helpfeel structured the articles from existing FAQ content, live in ~1.5 months versus a planned 3. ## What They Did Linked warranty-claim and cancellation forms directly from FAQ articles so customers complete the task themselves. Used search logs (which surfaced "cancellation" as a consistent top term) to build internal consensus and ship a new self-service workflow within a month. Deployed real-time FAQ surfacing inside the inquiry form, guiding self-resolution up to the moment of submission. ## Results A 40% self-resolution rate within one month let Hamee absorb 4-5x sales growth without adding staff. Cancellation inquiries fell 80%, eliminating post-dispatch return-shipping costs that had run up to 3x normal delivery cost. Organic search now drives 50-70% of FAQ traffic, with early signs of articles being cited by generative-AI tools. ## Looking Ahead Hamee adopted Helpfeel Agent Mode to unify product pages, FAQs, and manuals, improving satisfaction across the entire purchase journey, not just post-purchase support. --- ## Cutting Pandemic-Surge EC Inquiries in Half: Belluna Inc. Source: https://en.helpfeel.com/case-studies/cutting-pandemic-surge-ec-inquiries-in-half # Cutting Pandemic-Surge EC Inquiries in Half - Inquiries reduced 50% from pre-implementation levels - Inquiry rate (inquiries ÷ orders) down 2 points in the first month - Customer response time cut in half - FAQ page sessions doubled Belluna runs catalog and online retail across apparel, accessories, and home goods, serving primarily women aged 40-70. During the pandemic, stay-at-home demand pushed EC inquiries past 10,000 per month. The team grew to ~70 operators and still couldn't keep up: managers were answering email themselves, and there was no time left to fix the root cause. ## The Challenge The existing FAQ had no search, the chatbot had gone unmaintained, and some routine needs (stopping a newsletter, reissuing a payment slip) forced a support contact, artificially inflating volume. ![Inside a Belluna store](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/belluna_body.jpg) ## Why Helpfeel Triggered by a 2023 redesign of the Belluna Online Store, the team evaluated FAQ systems on maintenance ease, search speed, and support quality. Helpfeel won on all three (especially handling of variant spellings and phrasing), and the structured Customer Success support mattered for a team with little FAQ experience. ## What They Did Rewrote dense, formal articles into concise, friendly copy with images during migration. Used popups to surface the FAQ where customers get stuck. Embedded Helpfeel inside the inquiry form ("chasing FAQ"): articles appear in real time as customers type, promoting self-resolution up to submission. Boldly removed the legacy "Usage Guide" entirely, consolidating all guidance into the FAQ (data confirmed customers self-resolved). ## Results Inquiries fell 50%, response time halved, and FAQ sessions doubled. With routine load gone, the team built an inquiry-analysis framework and front-line operators began proactively proposing improvements, a cultural shift unthinkable during the overload period. --- ## CX transformation: Self-service and sales via website achieving a 96% drop in call volume: Odakyu Travel / Odakyu Electric Railway, Inc. Source: https://en.helpfeel.com/case-studies/cx-transformation-self-service-achieving-a-96-drop-in-call-volume # CX transformation: Self-service and sales via website achieving a 96% drop in call volume Odakyu Electric Railway is known for its "Romance Car" express trains in Japan that connect Hakone, Enoshima, and other prominent tourist destinations with central Tokyo. They began a drastic change in their travel business in 2021. The company completely moved over from a handling a wide range of domestic and international travel for individuals through its brick-and-mortar stores to online sales of travel products focusing on Odakyu rail lines and destinations. In line with this major transformation, customer contact points have also shifted from face-to-face and telephone to website and application. Helpfeel was used to enhance the help center so that questions can be quickly resolved on the same smartphones customers use to purchase and present digital ticket reservations. As a result, telephone inquiries have been greatly reduced under the new structure. We asked Hideyuki Nakamura, Project Manager of the Tourism Business Development Department of the company, about the deciding factors for choosing Helpfeel, the current status, and the effects they saw after introducing Helpfeel. ## Taking the Coronavirus pandemic as an opportunity to shift travel business from brick-and-mortar stores to online. ──First, please tell us about your responsibilities and your involvement with Helpfeel. I originally belonged to Odakyu Travel Inc. and worked as a store manager in its actual stores in Shinjuku and Machida, Tokyo. Currently, I am seconded to Odakyu Corporation, the parent company, and I am in charge of business promotion and customer service under a new system of online sales of products specializing in Hakone and Odawara, which are along the railway lines nearby. Odakyu Travel's website was renewed in June 2023, and it offers "dynamic packages" that allow customers to freely combine transportation, lodging, and local sightseeing. Here help pages for reservations and purchases are provided by Helpfeel, which was introduced in anticipation of the site revamp, and I am also responsible for its operation. ――You introduced Helpfeel at a time when your business itself was undergoing major changes. What was the background to the change and what were the key issues you faced? From 2021, the Odakyu Group has been promoting initiatives to become a "regional value-creation company" by developing new demand for sightseeing destinations along its lines and creating tourism experiences that take advantage of unique regional characteristics.As part of this effort, we began strengthening our travel business with an eye on the after-corona market. Specifically, it was decided to eliminate nearly 20 brick-and-mortar stores that handled a variety of domestic and international travel for individuals, and to specialize in "direct online sales" of travel in and around the Odakyu rail lines.This means that while we will retain our mainstay products such as "Hakone by Romance Car", the new system changed the way we offer our services, as we no longer provide paper tickets and face-to-face customer service.At the same time, the reservation system, staffing, and other aspects were reviewed to best suit direct online sales. We have been studying the best way to provide information on the Web and to reduce the number of telephone calls and emails.Even though Odakyu Travel had a help center on its website when we primarily offered in-person and telephone support, it was rarely used. However, under the new structure the role of the help center became much more significant. How to construct a help center that would enable customers to quickly obtain the information they need became a major challenge ![Odakyu Travel / Odakyu Electric Railway](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/odakyu_body.jpg) ## A new smartphone-only reservation system aiming for ease of use, even for first time customers. ──Can you tell us why you chose Helpfeel as your new help center? We decided to change from a conventional FAQ, which only had a list of questions and answers to one in which related answers are automatically displayed when a search word is entered.We chose Helpfeel after comparing several products that could do this, because we thought it would be the easiest for our customers to use, with features such as the ability to prominently display the most frequently asked questions. It was also important to us that we were able to see how Helpfeel worked in practice with the example of other businesses in our industry.──What kind of information did you want to focus on with Helpfeel? Considering that the majority of people already own smartphones, we have shifted our sales channel to the web, where people can make reservations by themselves from their phones.At the same time, however, our new reservation system had a flow that may be a bit confusing if you are not familiar with it. Therefore, we wanted to make use of help content to ensure smooth operation from the very first time. In particular, we had in mind the handling of tickets booked by multiple people.The Romance Car and other tickets included in Odakyu Travel products are sold exclusively via the Internet, and digital tickets are obtained and presented via an app or other means from one smartphone per person. Similarly, the Hakone Free Pass, which allows unlimited rides within the Hakone Mountains, also has a digital ticket that requires a smartphone for each person.In other words, for each of these products, one person cannot acquire digital tickets for everyone or present them together on behalf of the group, and the representative/companion of the group must follow the prescribed procedure. We considered the possibility of explaining this point online. ――How did you go about formulating help topics and creating answers? First, we created a new set of questions and answers within the company. Specifically, we asked several employees with extensive experience in dealing with customers both in person and over the phone to list “likely inquiries.” The common questions were consolidated into a single question, and we created answers to respond to them.The most frequently asked questions, such as “Can you change the time of the Romance Car on the same day?” were placed at the top of the list, where they could be clicked on quickly, so that users could get answers without having to enter words.Frequently used words such as “Romance Car,” “change,” and “cancel” are also placed at the top of the list, and when clicked, a list of related questions is displayed so that the user can search for an answer among them. ## Call volume was reduced by 96% and the number of phone lines was cut in half ──Please tell us about your current use of Helpfeel. Helpfeel has been in full operation since June 2023, when the reservation site reopened after a three-month preparatory period for site renewal. 116 help pages have now been created, with more to be added. Regarding the ongoing challenge of handling digital tickets for group reservations, we have first provided a general explanation page “How to access digital tickets for each member of a group reservation” in response to questions such as “I can’t find my digital ticket.” For more detailed explanations, links to other content on the site are provided. For example, one page has a link to a PDF with a step by step guide showing how the person who made the reservation can help each member of their group to access their digital ticket on their own device.――Have you seen an effect on your call volume? Although direct comparisons cannot be made due to factors such as the elimination of telephone numbers on the site and fluctuations in travel demand, we had over 3,600 unique visitors to Helpfeel in one month, and we believe that this has contributed to a significant reduction in telephone calls. Based on these results, we were able to reduce the number of telephone lines used to handle customer calls by half and also to reduce the number of staff responding to calls. ## Aiming for more detailed and comprehensive content by delegating updates directly to support staff ──What are your impressions of operating Helpfeel day to day and the support it provides? Compared to basic FAQs, Helpfeel is much easier to use, not only in terms of convenience for users, but also in terms of maintainability for operators, such as the ease of modifying and adding answers.It is good to be able to see the trends of help content usage by visitors to the site in real time. I have been serving customers in person and on the phone for a long time, and the daily results show that help pages serve many more people than I could have ever imagined.In our case, the trend and contents of contact messages were clear from the beginning, so we asked Helpfeel for assistance before and after the introduction to improve the flow so that we could immediately check the message content.Specifically, we asked Helpfeel to adjust the display order of frequently used keywords, such as “cancel” and “change” so that customers could find the answers to top questions, and this improved usability. ――Finally, what is your outlook for the future use of Helpfeel? Odakyu Electric Railway Headquarters Building According to the analysis of user behavior on Helpfeel, it seems that a large number of people are reading our help pages carefully, and it seems that because they are getting answers to their questions, the number of telephone calls is actually decreasing.I intend to share the work of adding pages to Helpfeel in the near future. I would like to further enrich the help content we have available, and make it possible to find all the explanations that are yet to be consolidated in Helpfeel. In addition, we would like to focus on improving support for seniors who browse the website but have difficulty making ticket reservations. We are already holding an Odakyu Travel Smartphone Reservation Class every day at the Odakyu Travel Center in Shinjuku, Tokyo where we invite customers to bring their own smartphones for face-to-face classes. --- ## From 15,000 Monthly Inquiries to Under Half: Moving From "Processing" to Experience: Jimoty Inc. Source: https://en.helpfeel.com/case-studies/from-15-000-monthly-inquiries-to-under-half-moving-from-processing-to-experience # From 15,000 Monthly Inquiries to Under Half: Moving From "Processing" to Experience - Monthly inquiries reduced from a 15,000 peak to a 7,200 low (over 50%) - Self-resolvable share of contacts fell from ~32% to ~20% in one year - Operator handling time cut from 37 to 23 hours/day (−14 hrs/day) - Significant drop in customer harassment / unreasonable complaints Jimoty is a local classifieds marketplace (a US OfferUp/Craigslist analog), serving primarily users in their 40s, mostly on mobile. At its 2021 peak, 15,000 monthly inquiries pushed a team juggling support, post moderation, and abuse reports to the brink. An earlier FAQ restructure cut only 2,000 inquiries because exact-match search failed on real-world phrasing: "can't get in" instead of "can't log in," or a typo like "loggin in," returned nothing. ## Why Helpfeel Across every option evaluated (other FAQ tools and chatbots), Helpfeel projected the highest inquiry reduction, specifically because of its handling of expression variation and approximate matching. ![A Jimoty team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/jimoty_body.jpg) ## What They Did Handed existing FAQ content to Helpfeel's technical writers to seed the search dictionary for a smooth launch. Added screenshots and step-by-step visuals where data showed users read an article but still contacted support. For "I got stood up / last-minute cancellation," answered the real underlying question ("what can I do now?") with reporting guidance. Created explicit "we don't offer this" articles to set expectations and prevent unanswerable inquiries. Pinned a "Jimoty Usage Guide" at the top of the FAQ for basics. ## Results Inquiries fell past 50% to a 7,200/month low. Operator handling time dropped 14 hours/day, letting the team absorb turnover without backfilling, a direct personnel savings. Eliminating phone support plus better self-service measurably reduced abusive, emotionally charged disputes. On the "Resolution Square" community forum, users now link FAQ pages when helping each other. ## Looking Ahead With FAQ-addressable contacts handled by self-service, Jimoty is shifting from "processing" inquiries to improving satisfaction, next adding a chatbot to compress the ~5-hour human response time to minutes. --- ## From 2,000 Inquiries a Month to Under 200: How a Streetwear Retailer Cut Contacts by 90%: B's International Inc. Source: https://en.helpfeel.com/case-studies/from-2-000-inquiries-a-month-to-under-200-how-a-streetwear-retailer-cut # From 2,000 Inquiries a Month to Under 200: How a Streetwear Retailer Cut Contacts by 90% - Monthly inquiries reduced from a peak of 2,000 to under 200 (less than one-tenth) - First-contact resolution rate improved from the 70s to 90% - Team freed to shift from reactive support to proactive customer success B's International runs the EC and retail operations for streetwear brands including XLARGE, X-girl, MILKFED., and SILAS. A four-person team was drowning in up to 2,000 email inquiries a month, routinely carrying work into the next day under constant exhaustion. ## The Challenge The questions were overwhelmingly operational (delivery status, payment, how-to), and each took multiple back-and-forth emails to resolve. The legacy FAQ had no search box: customers had to manually browse categories, a non-starter for a customer base of teens and twenty-somethings who search in casual, everyday language ("when does it arrive?" not "delivery lead time"). ![A B's International team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/bs-international_body.jpg) ## Why Helpfeel Helpfeel's intent-prediction search maps colloquial phrasing to the right answer, the exact mismatch the old FAQ couldn't bridge. Rather than incrementally patch the existing FAQ while still buried in email, the team chose a ground-up replacement. ## What They Did Migrated and keyword-tuned content; split long articles short (90% of EC customers shop on mobile). Pinned the top 3 inquiry types (delivery timing, payment methods, shipping) directly under the search bar, with dynamic ordering. This single change produced an immediate, sharp drop in inquiries. Redesigned the contact form with required fields so every inquiry could be answered in one reply, lifting first-contact resolution from the 70s to 90%. Printed QR codes on store receipts linking straight to the FAQ, routing in-store post-purchase questions to self-service. ## Results Inquiries fell to under 200/month ahead of schedule. Team members previously stuck firefighting email now improve store manuals and run in-person staff interviews. The department's mandate has shifted from reactive support toward proactive customer success and upsell. --- ## LUSH's Ecommerce Contacts Decreased 10% in Four Months; Wrapping Paper Sales Increased 120%: LUSH Japan, Inc. Source: https://en.helpfeel.com/case-studies/lushs-ecommerce-contacts-decreased-10-in-four-months-wrapping-paper-sales-increased-120 # LUSH's Ecommerce Contacts Decreased 10% in Four Months; Wrapping Paper Sales Increased 120% Lush is a pioneer in the field of natural handmade cosmetics, and as more and more companies are adopting the SDGs and sustainable practices, Lush is a brand that is becoming increasingly popular. Lush has 77 stores in Japan and over 900 stores in 49 countries and regions around the world. Lush is loved by many people, and due to its popularity the customer support department for their ecommerce site was receiving 3,200 contacts per month. As part of the company's efforts to create a system that can manage contacts even if the number of customers increases, the company moved its help content from its existing system to Helpfeel in March 2023. In this interview, we spoke to Nozomi Majima, the company's Customer Care Manager, about the situation before implementing Helpfeel and the changes that have occurred since, plus how they plan to use it in the future. ![LUSH Japan's Customer Care Manager speaking during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/LUSH_ec_body.jpg) ## 10 people handle 3,200 contacts per month. We needed a system to manage contact volume even as the number of customers increases. ──What issues did you face with your old FAQ page? Our FAQ page was ineffective because customers struggled to find what they needed. It had a long list of questions and answers on one page, and the search function didn't work well.During busy periods, 10 staff members handled up to 3,200 contacts a month, including phone calls and emails. We often received inquiries about delivery and order placement post-purchase, and we felt bad that we couldn’t always meet our customers' needs promptly.We also used a web customer service tool, but it closed at 6 PM. After that time, we could only direct customers to the FAQ page. With LUSH’s expanding sales channels, we wanted to prevent contact volume from increasing proportionally with purchases, so we considered introducing an easy-to-use help center with a strong search feature.We had already implemented Helpfeel for our internal staff help center and found its features easy to use. The intuitive image updating with copy and paste, plus the similarity between the editing page and the published page made it user-friendly. Therefore, we consulted with Helpfeel about introducing it for customer support as well. ──Did you consider other help center tools? Yes, we considered other systems but were concerned that their UI wouldn’t align properly with the LUSH brand. We wanted to avoid customers feeling they had left the LUSH site when visiting a help page.Helpfeel allowed us to create a UI design that perfectly matched the LUSH brand, and it integrated seamlessly with the main LUSH site. ## Four months after implementation, the contact rate dropped by 10% at the same time as order volume increase, achieving an ideal result. ──What changes have you seen since implementing Helpfeel? From the first month, the contact rate dropped nearly 4%, from 24% to 19%. Over four months, the contact rate reduced by almost 10%, even as orders increased. Our goal was to simply maintain the same contact volume, so this was a pleasant surprise.With fewer contacts, we could focus more on complex customer support issues. We managed to handle increased orders without expanding our contact response team. ──Have there been other changes besides reducing contacts? Yes, we published templates on the help page for specific inquiries, which led to more precise customer contacts. This minimized back-and-forth communication, reduced responses, and improved efficiency. ## Sales increased by 120%! And now it's possible to discover new user trends from search data. ──Are there other examples of using individual help pages instead of a single FAQ page? We transitioned from managing contact responses via a spreadsheet to searching for answers on Helpfeel. The help pages also serve as a manual for our support staff. When responding by email, we sometimes send the help page URL directly.This change encouraged team members to notice opportunities and suggest new help pages actively. ──Were there benefits to having answers on individual pages? We added a link to the Helpfeel page about gift wrapping on the checkout screen. Consequently, orders for gift wrapping items increased by 120% compared to the same period last year.Previously, the help content was hard to notice during order placement. By linking the help page to the checkout, we improved information flow, leading to increased sales. ──Did you gain insights into customer trends from help center data? LUSH has a Bring It Back initiative and a Fresh & Flowers service, but when we looked at the search data for the help center, we found that people were searching for things like “container exchange” and “subscription” rather than the service names.Even if customers do not search Helpfeel by using our official names, it is possible for them to hit the right pages using the words that are familiar to them. We have been adjusting the results based on suggestions from the customer success team at Helpfeel's regular meetings about what words should be linked to which pages.I don't think we would have realized how some customers actually refer to LUSH products until we saw the search log data like this. ──What kind of advice do you get at regular meetings with Customer Success? We receive specific action plans based on data, such as adjusting search keyword hits and creating missing pages. Recently, we discussed a flow proposal that directs customers through the help page before making contact. We rely heavily on their analysis and improvement ideas. ──From now, what are your future plans with Helpfeel? We plan to add more product introduction pages, linking them from the help center. This should help customers resolve their questions smoothly. ──Finally, would you like to share a message to companies facing similar issues? Helpfeel’s customer success team provides detailed advice on improvement actions, allowing you to move forward without worrying about creating better help content. For companies that value brand image, Helpfeel can build a help center and help pages in line with your brand. --- ## Scaling to 4.7M Users With No Added Support Headcount: akippa Inc. Source: https://en.helpfeel.com/case-studies/scaling-to-4-7m-users-with-no-added-support-headcount # Scaling to 4.7M Users With No Added Support Headcount - User base grew ~40% with no increase in support staffing - Inquiry rate (inquiries ÷ reservations) fell from ~1.4% to below 1.0% - FAQ access rate rose from ~40% to over 70% akippa is a parking-space-sharing marketplace with 4.7M+ registered users and 50,000+ spaces. As a startup, headcount is deliberately constrained, and a low-price, high-frequency model means cost-per-inquiry must stay tight: the team operates under an explicit mandate that even if revenue doubles or triples, inquiries can grow only 1.2-1.5x. ## The Challenge The old FAQ only did exact-match search: a broad term like "parking lot" buried the right article, while a slight rephrasing returned nothing, pushing users to contact support. ![A akippa team member during the interview](https://n4h88u4sldz4wmyd.public.blob.vercel-storage.com/images/akippa_body.jpg) ## Why Helpfeel In the demo, Helpfeel understood differently-phrased queries and guided users to the right article, critical for akippa's wide demographic, including many older space owners who need a UI "anyone can use without getting lost." Ongoing post-launch Customer Success support sealed the decision. ## What They Did Set a hard inquiry-rate KPI (≤1.07%) to grow users without growing support burden. Kept articles current with monthly product changes and wrote for plain-language accessibility. Used monthly Customer Success reviews to move from "looking at data" to "identifying the real problem." ## Results User base up ~40% with the inquiry rate held below target, dipping under 1.0% in recent periods, down from ~1.4%. FAQ access rate climbed from ~40% to 70%+. A channel shift from phone to live chat (≈70% cheaper per interaction) further optimized cost, and leadership recognized the scalable, self-service-driven support model. ---