What is a knowledge gap?
A knowledge gap is any point where a customer needs an answer and your knowledge base can't give the agent, or the customer, a correct one quickly. Most teams picture a gap as a missing article. That is only one of four kinds, and often not the most expensive.
| Kind of gap | What it looks like | Usual fix |
|---|---|---|
| Missing | Customers ask it, no article covers it. Agents improvise, transfer or promise a callback. | Create an article |
| Stale | An article exists but describes an old process, price, form or policy. Agents follow it and the customer calls back. | Update the article |
| Inconsistent | Two or more articles give different answers. Customers get a different answer depending on who they reach. | Merge, correct or retire |
| Unfindable | The right article exists, but its title and wording don't match how people ask, so nobody finds it. | Retitle, add phrasing, link it |
The distinction matters because each kind needs a different fix and a different owner. Writing a new article for an unfindable answer creates a duplicate. Retitling a stale article makes the wrong answer easier to find. The knowledge management glossary defines these and the other terms used across our guides.
What do knowledge gaps cost?
A gap rarely costs much on a single contact. It costs a lot because it repeats. Every customer who asks the same unanswered question adds handle time while the agent searches, asks a team leader or puts the customer on hold. Some of those contacts get transferred. Some end with a promise to call back. Some get a confident wrong answer, which produces a second contact a few days later. To put a dollar figure on one gap with your own cost per contact, use the worked model in the cost of a knowledge gap.
Repeat contact is where the damage compounds. COPC reports customer satisfaction falling from around 90% to around 55% when customers need three or more contacts to resolve an issue (COPC Inc.). Gaps also block self-service. Gartner puts the average assisted contact at $13.50 against $1.84 for self-service (Gartner, 2024), and self-service can only answer what the knowledge base contains.
Some costs never show up in handle time. Inconsistent answers create complaints and compliance exposure, and agents who can't trust the knowledge base stop using it.
Where should a knowledge gap analysis start?
Start from the demand side: what customers asked. Most teams start from the supply side instead, auditing the articles they already have, and that can only find problems inside existing content. It can never find the question nobody wrote an article for.
The common sources each see part of the picture:
- Customer conversations (call transcripts, chats, emails, tickets). The most complete record of demand, because every contact is in it, whether or not anyone searched.
- Search logs (failed, zero-result and abandoned searches). Useful for findability. Blind to every question an agent didn't search for. We cover this in detail in failed searches vs knowledge gaps.
- Article feedback (ratings, comments, flags). Good at catching stale content, but only on articles people opened.
- Agent and team leader requests. High quality, low volume, and biased towards whatever was frustrating this week.
- Quality evaluations. Show where agents gave wrong answers, but sample a small share of contacts.
Use conversations as the backbone and the other sources to explain what you find. A topic that customers ask about often, that agents rarely search for, and that has no article, is usually the biggest gap in the building.
How do you run a knowledge gap analysis, step by step?
- Pick a representative window. A month of conversations across every channel covers billing cycles, weekly patterns and most seasonal topics. Include voice, not just digital channels; many of the hardest questions arrive by phone.
- Extract what each customer wanted. For each conversation, record the customer's intent in plain language: "find out if the excess is waived for a storm claim", not "claims enquiry". One conversation often holds two or three intents.
- Group intents into topics. Cluster similar intents and arrange them into a taxonomy that reflects how customers ask, rather than how your organisation is structured. This makes volume and impact visible by topic. Building a taxonomy from conversations covers this step in depth.
- Check each intent against the knowledge base. Ask whether an article fully answers it, partly answers it, or doesn't answer it at all. Where two articles both claim to answer it, check that they agree, using the approach in duplicate and contradicting articles.
- Size each gap. Count the conversations affected and how difficult they were (see the next section).
- Choose an action. Create, update, link to an existing article, or dismiss with a reason. Record who owns it and when it will be done.
- Check the result. After the fix is published, watch the same topic. Transfers, callbacks and repeat contacts should fall. If they don't, the fix didn't work.
Steps two to four are where manual analysis breaks down. Reading and coding conversations by hand is slow, so teams sample, and sampling misses the long tail of lower-volume topics where many expensive gaps hide.
How do you rank knowledge gaps by impact?
Rank by impact: how many conversations a gap affects, weighted by how difficult those conversations were. A frequently asked question that agents already handle well has little impact. A less common question that triggers transfers and callbacks every time has a lot.
Impact of a gap = conversations affected × a difficulty weighting for transfers, callbacks, unresolved contacts, negative sentiment and repeat contacts on that topic.
For example, using invented numbers: a gap affecting 1,000 conversations a month, a third of them transferred or called back, might be weighted by 1.3 for an impact of 1,300. A gap with 300 conversations and no extra friction scores 300. The first goes to the top even if the second is easier to write. For a business case, multiply by your own cost per contact: at $9, about $11,700 and $2,700 a month.
Flag gaps with compliance or safety risk separately. A low-volume gap where agents give the wrong answer on a regulated product can matter more than its score suggests.
What should you do with each gap?
Every gap should end in one of four decisions:
- Create an article when nothing covers the question.
- Update an existing article when it is stale, incomplete or contradicts another.
- Link an existing article when the answer exists but people can't find it. Fix the title and wording so it matches how customers ask.
- Dismiss with a reason when the question doesn't warrant an article, for example a one-off outage already resolved. If the pattern returns, it should resurface.
Write fixes from the evidence, the same principle that makes AI answers checkable. The conversations that exposed the gap show the exact words customers use, the follow-up questions they ask and the points where agents got stuck. A draft built from those conversations answers the real question in the customer's language. Keep a link from the draft back to its source conversations so a reviewer can check it, and have a subject-matter expert approve it before it is published.
How do you stop knowledge gaps coming back?
Gaps reopen for predictable reasons: a price changes, a form is replaced, a product launches, a policy is reworded in one article and not another. Three habits keep the knowledge base current:
- Run the analysis continuously. New conversations arrive every day, so new gaps do too. Watch for emerging topics, questions that barely existed last month and are growing quickly, and close them while they are small.
- Check content quality on a schedule. Review articles for readability, validity (is it still true?), brand alignment, complexity, duplicates and contradictions. Duplicates are the early warning: two articles on the same topic drift apart until they contradict each other.
- Measure the fix. Track transfers, callbacks and repeat contacts on each topic before and after the article changes. Views and thumbs-up ratings tell you an article was opened. They don't tell you it resolved anything.
If your team follows KCS, these habits are its evolve loop, fed by conversations instead of by agents finding time to write. Closing the KCS loop in a contact centre shows how the two fit.
Do you need a new knowledge base to do this?
No. Knowledge gap analysis is a layer that sits over the platforms you already run. It reads conversations from your contact centre platform or service desk and sends findings and drafts to your knowledge base. Replacing the knowledge base doesn't solve the problem, because a new tool starts with the same content and the same blind spots. A knowledge tool is only as good as the knowledge in it, and AI search won't fix a legacy knowledge base. See adding gap detection to your knowledge base.
This is how Knowledge Quest works. It reads every conversation from Genesys Cloud, Amazon Connect, Freshservice or Zoho Desk, extracts what each customer wanted, and checks it against your knowledge base. Gaps are ranked by impact, with the evidence attached. Knowledge Copilot drafts the article or update from the conversations, citing its sources, and your team approves it. Six quality checks keep existing articles accurate and consistent. Drafts sync to supported knowledge bases, and work alongside other knowledge tools through import and export. You keep your tools; they get better knowledge.
Frequently asked questions
How often should we run a knowledge gap analysis?
Treat it as a standing process with a weekly review. Products, policies and customer questions change every week, so an annual or quarterly audit is out of date by the time it is finished. A weekly list of new and growing gaps, plus a monthly look at whether closed gaps stayed closed, suits most contact centres.
What is the difference between a knowledge gap and a training gap?
A knowledge gap means the right answer isn't available, or isn't findable, in your knowledge base. A training gap means the answer is there and easy to find, but agents don't use it or don't apply it correctly. Check the knowledge base first. If a good article exists and agents still get it wrong, it's a training or process issue.
Can we do knowledge gap analysis manually?
Yes, on a sample. Pull a few hundred conversations from a typical month, write down what each customer wanted, and check whether an article answers it. It works for a first pass, but it is slow, it misses low-volume topics, and it goes stale quickly. Automated analysis reads every conversation and repeats the check continuously.
Isn't a failed search report enough?
No. Failed searches only show gaps an agent searched for, in the words they used. Questions answered from memory, by a colleague or by a guess never appear. See failed searches vs knowledge gaps for the full comparison.
Does knowledge gap analysis work for service desks as well as contact centres?
Yes. The method is the same with tickets in place of calls and chats. The ticket description, the work notes and the resolution show what the requester needed and whether the knowledge base helped. See how it works for service desks.