There is no useful universal target without a defined task mix, denominator and verification method. Set targets against your own baseline and separate conversational containment from successful resolution.
For this desk’s measurement framework:
| Metric | Recommended definition |
|---|---|
| Autonomous task completion | Unique eligible tasks completed correctly without human rescue ÷ all eligible tasks assigned. |
| Assisted task completion | Unique eligible tasks completed with human rescue ÷ all eligible tasks assigned. |
| Booking integrity | Claimed confirmed bookings meeting every required rule ÷ all claimed confirmed bookings. |
| Handoff success | Required handoffs reaching the intended staffed destination with necessary context ÷ all required handoffs. |
| Repeat contact | Customers contacting again about the same unresolved task within a predefined window. |
| Total cost per successful task | All attributable workflow costs ÷ unique verified successful tasks. |
Report total routed demand alongside eligible demand so that exclusions remain visible. Freeze eligibility rules before the evaluation; do not remove difficult calls after seeing their outcomes.
Track appropriate refusals and safe escalations as separate useful outcomes. Do not silently reclassify them as completed bookings.
Verification matters: Vapi explicitly warns that a tool reporting success does not independently prove an external system changed. (docs.vapi.ai)
The right dashboard allows an operator to trace a claimed success back to the actual record and identify who corrected it, if anyone.