# Voice go-live checklist (cheat sheet)

> Latency, barge-in, entity accuracy, consent, escalation and cost checks before a voice agent takes real calls.

Source: https://conversationalaidesign.com/cheatsheets/voice-go-live-checklist.html. Author: Nelson Abacherli Kawakami. Reviewed November 3, 2026.

## Measure on real calls
- End-to-end latency from the caller stopping to the agent starting, with your own prompt and voice, under load
- Barge-in stops speech at once and the agent listens
- End-of-turn detection neither cuts people off nor leaves dead air

## Get the details right
- Entity accuracy on names, numbers and codes, not only overall accuracy
- Read-back confirmation for important values
- Another channel to send a code by text
- Accents, noise and slow speakers tested (see the voice localization module)

## Law and trust
- Say it is an AI in the first sentence
- Recording and consent rules checked for each country and state
- Escalation to a person and its hours
- Behavior when speech recognition fails twice

## Cost and operations
- Cost per resolved call measured on real traffic
- Monitoring for latency, interruptions, escalations and failures
- A kill switch and a tested fallback

Related modules: Voice assistants (https://conversationalaidesign.com/modules/19-voice.html), Voice localization: STT and TTS across languages (https://conversationalaidesign.com/modules/20-voice-localization.html)
