Test and monitor
Understand your analytics
Compare call counts, duration, latency, outcomes and recorded usage.
Choose the correct scope
Open Monitor → Analytics. Select the call or chat view, date range and agent filters. Use breakdowns to compare the relevant cohort rather than combining browser tests, phone calls and different agents in one unexplained average.
The dashboard includes counts, average duration, latency where recorded, concurrency, session statuses, sentiment and direction. Usage charges reflect recorded usage; check Billing for your account's rates and balance.
Interpret latency carefully
A single average hides slow outliers and does not isolate the reason for delay. Compare the same model, language, prompt length and tool path. A calendar lookup or CRM write may be slower than a simple question, even with the same voice model.
No latency value means the measurement is unavailable; it should not be read as a zero-millisecond turn. Caller network quality, microphone capture, endpointing, provider response and audio playback can all affect perceived delay.
Make a change you can measure
- Select a representative agent and period.
- Inspect a few slow or failed sessions in Call history.
- Determine whether the problem is recognition, instructions, an integration or routing.
- Change one setting or prompt section.
- Test the new draft and publish an accepted version.
- Compare a similar set of calls after release.
A positive sentiment score or completed status does not prove a booking was correct. Use AI quality assurance and source-system receipts to check task success. Custom dashboards can organize metrics; they cannot reconstruct audio or tool results that were never recorded.