Reports
The reports page works with nothing to configure: it is computed from real conversation records. The period can be seven days, thirty days, ninety days, this month or last month.
Measures
- Bot resolution rate — the share of conversations closed by the assistant alone
- Unanswered question rate — the share of questions the assistant could not answer. This figure is free; the system writes it every turn and replies written by your team are excluded
- Median response and P90 response — response time, shown as a median and a 90th percentile rather than an average, because a few slow turns distort the mean
- ≤1 min response and ≤5 min response — the share of replies inside those thresholds
- Average first response — the mean delay to the first reply in a conversation
- Satisfaction — the share of customers who found what they needed. Given as a percentage, not an average
- Converted requests — how many requests were closed
- Off-hours requests — requests that arrived outside working hours
Breakdowns
New conversation trend, request funnel, score distribution, lifecycle stage, conversation status, hour and day density, most frequent topics and channel performance.
Hour and day density feeds straight into your staffing plan: when requests cluster, scheduling people and shifts around that becomes meaningful.
Most frequent topics is computed only over analysed conversations. The card says so and also shows the total conversation count for the period. With no analysed conversations the card is empty — that means the analysis has not been run yet, not that data is missing.
Export
Export is CSV and opens directly in Excel and Google Sheets. The contact and request dump covers all history. The conversation and message dump covers the selected period.
Export produces the same data as Data and privacy; the difference is that you can pick a period here.
Reports are free
The whole reports page is free and makes no new model call. The separate Conversation Analysis screen, which extracts topic, sentiment and quality per conversation, is different: running an analysis there is a billable model call.