Statistics and analytics
Reading what your AI front desk is actually handling, so decisions come from your own numbers rather than impressions.
What is reported
| Measure | Tells you |
|---|---|
| Call volume | How much is arriving, and when. The shape of your week matters more than the total. |
| Outcomes | What conversations turned into - refunds, bookings, orders, escalations, or nothing |
| Escalation rate | How often a person was needed. A useful proxy for whether knowledge is adequate. |
| Call quality | First reply latency and interruption handling - see call quality signals |
| Usage | Consumption against plan allowances - see usage and limits |
Questions worth asking of the data
Numbers on their own do not tell you what to change. These do.
- Is the escalation rate rising?
- Usually missing knowledge rather than a worse model. Read the transcripts behind the escalations.
- When do calls actually arrive?
- If a third arrive outside your hours, your out-of-hours handling matters more than you think.
- What is the AI failing to answer?
- Every unanswered question is a candidate for your knowledge base.
- Is latency drifting?
- Compare across days, not calls. A single slow call is noise.
Reading the numbers honestly
Small samples mislead. A 50% escalation rate across four calls tells you almost nothing. Wait for a volume that means something before changing configuration on the strength of it.
See also
- Dashboard tour - where the statistics module sits
- Health checks - configuration readiness, as distinct from performance
- API overview - pulling records into your own reporting