Finding Knowledge Gaps from Call Reviews and Caller Questions
How to spot the answers your team and AI agents are missing by reviewing calls, grouping repeated questions, and closing each gap with a clear, owned answer.
Your callers tell you what they need to know every day. The trouble is that most of it vanishes into recordings nobody has time to replay. A knowledge gap is a question callers ask that your team, your scripts or your AI agent cannot answer well. Finding those gaps and closing them is one of the cheapest ways to shorten calls, reduce transfers and make callers feel looked after.

This guide shows how to turn call reviews and caller questions into a steady stream of fixes. It pairs with Building a Contact Center Knowledge Base and, for AI agents, RAG for Voice Agents.
What a knowledge gap looks like on a call
Gaps rarely announce themselves. They show up as small behaviours that repeat. Train reviewers to note them whenever they listen to a call.
Signals that point to a gap
- Long silencesThe agent is searching for an answer while the caller waits.
- Hedging or guessingPhrases such as "I think" or "I believe so" about facts that should be certain.
- Unnecessary transfersA call passed on because the first person did not know.
- Repeat callsThe same caller returning because the first answer was incomplete.
- Conflicting answersTwo agents giving different replies to the same question.
- AI agent declinesA voice agent saying it cannot help on a topic callers regularly raise.
Not every gap is missing information. Sometimes the answer exists but is hard to find, out of date or written in terms callers do not use. Note which kind you are looking at, because the fix differs.
Collect the evidence
Use several sources, because each sees different things. Call reviews give rich detail on a few calls. Transfer and repeat-call data show where the pattern is wide. Agent feedback reveals the questions people dread. Search logs from your knowledge base show what staff look for and fail to find.
- Add a field to your review scorecard: "Did the agent have the answer they needed?" with a note on the topic.
- Ask agents to tag calls where they had to hunt or escalate.
- Track the reasons for transfers and look for ones with the same cause.
- Look at repeat calls within a short window from the same number.
- For AI agents, list the questions they could not answer and the ones they answered with low confidence.
Keep the log simple: the question, the topic, the date, the source and how many times it has appeared. A spreadsheet is enough to begin with. What matters is that the same question is counted once and not rediscovered every month, so check the existing list before adding a new line.
From call to closed gap
- NoticeA reviewer or agent flags a question that was hard to answer.
- RecordLog the question in the caller's own words, with the topic.
- GroupMerge similar questions so you see how often each one comes up.
- RankPut the most frequent and highest-impact gaps first.
- AnswerAn owner writes one clear, checked answer.
- PublishAdd it where agents or the AI agent will find it.
- CheckWatch whether the question still causes trouble.
Rank the gaps
You will find more gaps than you can close at once, so rank them. Two measures do most of the work: how often the question comes up, and what it costs when handled badly. A rare question about a legal matter can matter more than a frequent question about opening hours. Use a simple grid rather than a complicated score.
Which gaps to close first
| Frequent | Occasional | |
|---|---|---|
| High impact (money, safety, legal, loyalty) | Close first, with review by an expert | Close soon, with a clear owner |
| Low impact (convenience, small detail) | Close next, often a quick win | Leave for later or note in a list |
Sort the gaps by type too. Some are plain missing facts. Others are policy questions that need a decision before any answer can be written. If a gap needs a decision, send it to the person who can make one, and do not let agents improvise in the meantime.
Writing an answer that closes the gap
A good answer is short, specific and written in the words callers use. Lead with the answer, then give the conditions. Say who owns it and when it was last checked. Avoid long policy extracts; agents on a live call need a few clear sentences and maybe a next step.
What every answer needs
- The question as callers phrase it, plus common variants.
- A direct answer in one or two sentences.
- Conditions or exceptions, if any.
- What the agent should do next, such as a transfer, a form or a callback.
- An owner and a review date.
Who needs to be involved
Closing gaps is a team effort, and giving each group a clear job keeps it moving. Agents are the front line: they know which questions cause trouble and are best placed to flag them. Reviewers see patterns across many calls. Content owners, often product, policy or operations leads, write and approve answers. A coordinator keeps the list and the dates.
A hypothetical example: a reviewer notices three calls in one week where agents hesitate over how a delivery change works. They log the question, the coordinator finds two more from the transfers list, and the operations lead writes a short answer with the conditions. A week later, reviews show smoother calls, and the topic leaves the list. Nothing here needs special tools, only a routine.
- Give agents an easy way to flag a gap in under a minute, or they will not do it.
- Thank people when a flagged gap is closed, so flagging feels worthwhile.
- Agree a target time between flagging and publishing a fix.
- Review the list in a short weekly meeting, not by email.
Close the loop
A gap is not closed when the answer is published. It is closed when the problem stops. Return to the evidence after a few weeks: have transfers on this topic fallen, do reviewers still hear hedging, do repeat calls on it drop? If not, the answer may be hard to find, unclear or wrong. Fix that before moving on.
Gap review routine
- New gaps logged from reviews and agent feedback
- Similar questions grouped and counted
- Top gaps ranked and given owners
- Answers published where agents will find them
- Older answers checked for accuracy
- Previous gaps rechecked for improvement
Doing this in Call
Call has a Suggestions page under Sources for exactly this job. It offers three views: Gaps, Caller questions and Suggested Q&A. In Gaps, a question appears when your agent could not answer a caller from its sources and said so; you can filter by not covered by sources, low confidence or declined by the agent, sort by most critical, then press Add as Q&A and write the answer, or Dismiss a question you will not answer. In Caller questions you see what callers asked, grouped with how often, and can write the missing answers. Test calls are marked so they do not mislead you.
Members can read sources, and workspace managers can add, change or delete them. For human agents, call summaries and transcripts in Call history help reviewers spot hesitations; see AI Call Summaries for how summaries help review.
Frequently asked questions
How do I find knowledge gaps without replaying every call?
Use signals: transfers, repeat calls, agent tags, review scorecard fields and, for AI agents, questions they could not answer. Replay only the calls those signals point to.
How many gaps should I try to close at once?
A handful. A short list closed well beats a long list left half done. Rank by frequency and impact.
Who should own an answer?
A named person who knows the subject, such as a product lead or policy owner, and who can say when it changes.
What if a gap needs a policy decision?
Send it to the person who can decide, record the decision and then write the answer. Until then, tell agents what to do, such as escalate.
How do I know a gap is really closed?
The signals that revealed it fade: fewer transfers, less hedging in reviews and fewer repeat calls on the topic.
Do AI agents need the same process?
Yes. The AI agent can only answer from what it has been given, so the same cycle of finding, answering and checking applies.