Containment rate — calls completed without a human — is the headline metric for Voice AI, but it can hide poor outcomes if agents trap callers or hang up early.
Track escalation quality: did the human receive a clean transcript, collected fields, and a reason code? Warm handoffs protect CSAT when the bot correctly steps aside.
Add task KPIs such as appointment booking rate, lead qualification completeness, and average handle time against a human baseline for the same queue.
Review failure clusters weekly — noisy STT fields, missing intents, and policy refusals — and fix the top drivers instead of only tuning the voice persona.
Digicane instruments Voice AI pilots with containment, handoff quality, and CRM write-back success so buyers see operational truth, not vanity ratios.