How to Build a Knowledge Base Your Agents Will Actually Use

Building a knowledge base is one of those support initiatives that almost every team starts with good intentions and most teams quietly abandon. The content gets outdated. Agents stop searching it because it takes longer to find an answer than to just know it. New agents don't know it exists until they've already trained themselves the hard way.

The difference between a KB that gets used and one that collects dust comes down to three things: the right scope, the right format, and a gap detection system that keeps it current. This article covers all three.

Why Most KB Projects Fail

The three failure modes for knowledge base projects:

1. Too much content, too little organization. Teams import everything - policy docs, SOPs, training materials, product specs - and call it a knowledge base. Agents search for "return policy" and get 47 results, none of which are formatted for quick in-conversation use. The KB becomes a noise source rather than a signal source.
2. Wrong format for the use case. A 1,500-word policy document is fine for a training manual. It's useless when an agent has a customer waiting. KB articles for live support need to be short, specific, and structured for skimming at speed.
3. No update process. Products change. Policies change. Pricing changes. If the KB articles don't reflect current reality, agents stop trusting them. Once trust is broken, agents go back to asking each other instead of searching the KB - and the institutional knowledge lives in Slack threads and people's heads instead of a searchable system.

The 80/20 Rule: Start With Your Top 20 Questions

You don't need 500 KB articles to start seeing results. You need the 20 articles that answer 80% of incoming questions. In practice, every support team has a small set of questions that come in at much higher volume than everything else.

Your top 5 questions alone account for more than half your volume. Writing clear, well-formatted articles for those five questions will have more impact than building 100 articles about edge cases.

How to Find Your Top 20 Questions Using Velaro Conversation Data

The fastest way to find your top 20 is to pull the data from your existing conversations - not guess. In Velaro, you can generate a Topic Report from the Analytics dashboard that clusters conversations by detected intent and ranks them by volume. Alternatively, if you don't have topic detection enabled yet, do a manual audit:

1. Export the last 500 conversations from your most active channel
2. Tag each conversation with a primary topic (the first substantive question the customer asked)
3. Sort by topic frequency
4. The top 20 topics by count are your initial KB scope

This takes 2–4 hours for a team of one. It's the most valuable 4 hours you'll spend on your KB project, because it ensures you build what's needed rather than what seems thorough on paper.

"Don't build the KB you wish customers were asking about. Build the KB for the questions they're actually asking - every day, at scale."

Article Format That Works for In-Chat Suggestions

The format of your KB articles matters as much as the content. An article optimized for in-chat surfacing looks very different from a policy document:

Here's the template that works best for in-chat KB articles:

The Gap Detection System: Using Velaro to Find Unanswered Questions

Even the best initial KB will develop gaps. New product features create new questions. Policy changes make old articles wrong. Seasonal events (holiday shipping delays, pricing promotions) generate new question spikes.

Manual gap detection relies on agents noticing and reporting gaps - which rarely happens consistently. Velaro's automated gap detection works differently:

1. Search query analysis: When agents search the KB and don't click any results (or refine their search multiple times), Velaro flags the search query as a potential gap. These flagged queries are surfaced in a weekly Gap Report.
2. Bot escalation patterns: When the bot escalates a conversation because it couldn't resolve the intent, the bot's transcript is logged. If the same intent escales frequently without a KB match, it appears in the Gap Report.
3. Low-confidence suggestions: When Velaro's KB suggests articles with low confidence scores (below 70%), those suggestions are also flagged - they indicate that an article exists but doesn't match the query well enough to be useful.

The KB Maintenance Schedule

A knowledge base without a maintenance process is a future liability, not an asset. Here's the minimum viable maintenance cadence:

• Weekly: Review Gap Report. Flag articles that need updates. Takes 20 minutes.
• Monthly: Review the 10 most-used articles for accuracy. Check that steps still match current UI or process. Update anything that's changed.
• On product/policy changes: Assign a KB update as a required deliverable of any product release or policy change. The KB update is part of the release checklist, not an afterthought.
• Quarterly: Full audit. Archive articles that haven't been viewed in 90 days (they're not filling a gap). Review article confidence scores and rewrite low-performers.

The weekly review takes 20 minutes. The monthly review takes an hour. This is the total investment to keep a KB accurate and useful. It's not a significant commitment - but teams that skip it watch their KB degrade over 6 months until agents stop using it.

Getting Agents to Actually Use the KB

Technical quality matters less than adoption. A KB nobody searches is worthless. The three adoption drivers that actually work:

1. Integrate it into the chat window. Velaro surfaces relevant KB articles as agents type - they don't have to leave the conversation to search. Passive suggestion is 4x more effective at driving usage than requiring agents to search manually.
2. Show usage metrics. When supervisors review CSAT scores alongside KB usage data, agents learn that using KB correlates with higher scores. Make the data visible in team reviews.
3. Let agents contribute. Agents who can flag gaps and submit new articles feel ownership over the system. Agents who are passive consumers of a curator's choices don't. Add a "suggest an edit" or "report a gap" button to every article.

The Bottom Line

A well-built knowledge base is a force multiplier for your entire support operation - it makes every agent faster, enables AI self-service at scale, and reduces ticket volume without adding headcount. The teams that get the most from their KB treat it as a living product, not a one-time project: they measure gaps, maintain cadence, and build agent ownership. That ongoing investment is what separates a KB that actually deflects tickets from one that collects dust.