# Voice-AI readiness checklist

Do not launch until the owner can evidence each relevant item.

- The call reason is narrow, frequent and measurable.
- The authoritative data source is known and available in real time.
- The system discloses that it is automated and offers a human route.
- Identity requirements are defined before personal disclosure or action.
- Allowed and prohibited actions are enforced outside the model.
- Recording, transcription, retention and outbound-call rules were reviewed.
- Representative accents, languages, background noise and interruptions were tested.
- Latency, silence, packet loss and repeated misunderstanding are monitored.
- A human is available within a defined service level.
- Handoff includes identity state, goal, facts, attempts and unresolved decision.
- Provider events and media connections are authenticated.
- The workflow has a safe fallback during model, tool or telephony failure.
- Inputs, sources, decisions, approvals and actions are recorded proportionately.
- Success is measured by accepted resolution, not deflection alone.