What matters for your decision
Use clear fallback rules for missing evidence, integration outage, and after-hours support. Make the customer-visible message depend on what the next system actually confirmed.
Put this into a pilot
- Define channel-specific fallback to a human, a form, or a later retry and accurately state hours or expected ownership.
- Verify the fallback with synthetic unavailable queue and integration scenarios; check that the user can proceed without repeating the entire context.
- Record uncertain action status for reconciliation, and avoid saying an action succeeded before there is downstream confirmation.