What does Gap Analysis compare against? Your prompt and knowledge base
Gap analysis runs after a call's AI analysis completes, then compares the transcript against your agent's prompt and its knowledge base. For a knowledge gap it goes one step further and tells you whether the answer already existed in your knowledge base — a retrieval problem — or was missing entirely — a content problem — so you fix the right thing.
When it runs
Gap analysis isn't a separate scan you have to trigger. It runs automatically after a call's AI analysis finishes, so gaps appear on their own once a call has been transcribed and analyzed. (See how a call is analyzed, start to finish.)
What it compares
For each call, Gap Analysis lines up three things:
- The transcript — what actually happened on the call.
- Your agent's prompt — the instructions, rules, and flow you gave it.
- Your knowledge base — the facts and answers the agent can draw on.
When the transcript shows the caller wanting something the prompt or knowledge base didn't deliver, that mismatch becomes a gap.
Retrieval problem vs. content problem
Knowledge gaps come in two flavors, and the fix is different for each:
| What the analysis found | What it means | How you fix it |
|---|---|---|
| The answer already existed in your knowledge base | A retrieval problem — the agent had the information but didn't use it | Improve how that content is phrased or surfaced so the agent finds it |
| The answer was missing | A content problem — the information wasn't there to begin with | Add the missing answer to your knowledge base |
Telling these apart matters: adding content you already have won't help if the real issue is retrieval, and re-wording existing content won't help if the answer was never there.