3. AI report
Use Reports → AI to review agent handling, resolutions and handoffs. These figures help you identify where knowledge or handoff guidance may need attention; read them alongside example conversations.
The same component appears in AI agent → Performance for the selected workspace. Without explicit report dates, it uses a 30-day lookback.
Metrics
| Metric | Definition |
|---|---|
| AI conversations | Conversations whose stored AI-window start falls in the range. |
| Confirmed | Confirmed AI resolutions whose resolution timestamp falls in the range. |
| Assumed | Assumed resolutions recorded at eligible AI inactivity closure in the range. |
| Handoffs | Conversations whose stored handoff timestamp falls in the range. |
| Resolution rate | (Confirmed + Assumed) divided by AI conversations; zero with no denominator. |
| Handoff rate | Handoffs divided by AI conversations; zero with no denominator. |
| Model resolved signal | Conversations whose stored model-resolved timestamp falls in the range; a model signal is not customer confirmation. |
AI per workspace
The chart shows AI conversations, Resolved and Handoffs. The table shows Workspace, AI conversations, Confirmed, Assumed, Handoffs, Resolution rate and Handoff rate.
Handoff reasons
This section counts recorded handoff reply runs by reason, using their run timestamps.
Example results
Illustratively, 40 AI conversations, 10 confirmed resolutions and 20 assumed resolutions produce a 75% resolution rate. Eight handoffs produce 20%. These numbers are an arithmetic example, not measured performance.
Comparing rates and counts
Keep these differences in mind:
- Resolution and handoff rates use different event dates from AI conversations. They may refer to different sets of conversations and can exceed 100% over a narrow period.
- Handoff reasons count reply runs; Handoffs counts conversations. Their totals do not have to match.
- Some direct handoffs do not produce a model reply run.
Data exclusions
Spam is excluded from the conversation aggregates. The reasons query uses reply-run records and does not apply an equivalent conversation-spam filter. Deleted conversations and their runs are removed. Test-console calls are not customer conversations in these aggregates.
Related guides: Report controls, Handoff, Credit windows.