Where a clinic booking agent loses the appointment.
A representative Voice Agent evaluation of an inbound appointment line for a multi-clinic group. Illustrative throughout — it shows the depth of a real evaluation, not a named client result.
System evaluated
Inbound appointment voice agent
Domain
Multi-clinic healthcare booking
Reporting period
Single evaluation cycle
Channel
Voice · inbound
The agent books cleanly when a caller knows exactly what they want. The open question was how it holds up under real call behavior — mid-call corrections, spelled-out names, and callers who stack a second request onto a booking.
Scope
160 human-tested calls · 4 caller profiles · 8 scenarios · English
Performance
Weighted overall: 3.6 / 5 — Acceptable, with one limiting weakness.
Diagnosis
Caller
Actually, make that Thursday, not Tuesday.
Agent
Great — I have you booked for Tuesday at 2pm.
Impact. The agent reverts to the original date after a mid-call correction. The caller leaves the call booked on the wrong day.
Recommendation. Treat any correction as the new source of truth and re-confirm the changed value before closing.
Caller
My last name is Nguyen — N-G-U-Y-E-N.
Agent
I have that as Wynn. Is that correct?
Impact. Spelled-out input is miscaptured under accent variation, forcing repeated correction and stretching the call.
Recommendation. Read spelled input back letter by letter for confirmation before saving it to the record.
Caller
Can I book for next week and also check if you take my insurance?
Agent
Sure — what day works for the appointment?
Impact. The insurance question is silently dropped. The caller has to phone back for the half they did not get.
Recommendation. Detect the second intent in the turn and confirm both will be handled before proceeding.
An excerpt of the per-call log — every finding carries a reproducible Call ID, a failure type, and a severity on the 4-band scale.
Resolution
74 findings total — scored on the Voice rubric and ranked by severity, the same way every evaluation reports.
Fix context retention after corrections
Targets the lowest dimension (2.8) and the single biggest source of wrong bookings. Highest expected lift.
Harden spelled-name capture
Reduces the repeated-correction friction that stretches calls and frustrates callers.
Add multi-intent detection on booking
Recovers the secondary questions that are currently dropped mid-booking.
Overall 3.6 / 5 — Acceptable. The agent books reliably on clean, single-intent calls but loses accuracy the moment a caller corrects a detail or stacks a second request, and context retention is the limiting factor. With the three priority fixes it is ready for a supervised pilot on the main booking line; it is not yet ready to run insurance-related or multi-step calls unattended.
Prepared by KNK Global · evaluation services
See this on your own booking line.
A pilot returns an evaluation in this exact shape, scored on your live voice agent under real callers.