What does Gap Analysis detect? Gap types, severity, and root cause
A gap is a moment where the caller wanted one thing and the AI did another — with an attributable cause and a recommended fix. Every gap is classified three ways: a type that says what kind of failure it was, a severity that says how badly it hurt the caller, and a root cause that points to where the fix belongs.
What counts as a gap
Not every imperfect call is a gap. Gap Analysis flags a moment only when there's a clear mismatch between what the caller expected and what the agent did, and it can attribute that mismatch to a cause it can recommend a fix for. That's what makes each gap actionable rather than just a complaint.
Gap types
Each gap is tagged with one of six types, so you can tell at a glance whether the fix lives in your prompt, your knowledge base, or your call flow.
| Type | What it means |
|---|---|
| Knowledge | The agent didn't have — or didn't surface — information it needed to answer. |
| Instruction | The agent didn't follow a rule or step defined in its prompt. |
| Execution | The agent had the right intent but carried out the action incorrectly. |
| Communication | The information was right, but unclear, incomplete, or poorly phrased to the caller. |
| Expectation | The caller expected something the agent was never set up to do. |
| Policy | The agent acted (or held back) for a compliance or business-rule reason. |
Severity
Severity tells you what to fix first — it reflects how much the gap hurt the caller experience or your agency's goals on that call.
| Severity | Roughly means |
|---|---|
| Critical | The caller was actively failed — wrong information or a lost opportunity. |
| High | A clear miss worth fixing soon. |
| Medium | A noticeable issue with limited impact. |
| Low | A minor or cosmetic shortfall. |
Root cause
Every gap carries a root cause — for example, prompt logic or missing knowledge — that connects the failure to where you'd change something. The root cause is what makes the recommended fix specific: it's the difference between "the agent was wrong" and "this line in your prompt sent it down the wrong path." How Gap Analysis compares your transcript against your prompt and knowledge base explains where those causes come from.