Test and improve a voice workflow¶
Choose a few typical conversations as test cases and repeat them after making changes. Check both the Agent’s replies and the messages or data your team receives.
Choose test cases¶
| Scenario | Expected outcome |
|---|---|
| Complete request | Correct information, confirmed details and intended handoff |
| Missing or corrected number | Agent asks for the number and confirms it with the caller |
| Ambiguous intent | Clarifying question or DEFAULT Agent |
| Known knowledge question | Answer supported by the current source |
| Unknown question | Honest fallback without an invented policy |
| Tool error or timeout | Agent explains the failure and offers the planned next step |
| Recipient declines forwarding | Caller returns to the intended fallback |
| Silent caller | Summary and delivery follow configured rules |
Add cases for personalized-greeting fallback, conditional delivery and each Flow intent when used.
Run a test¶
- Test the integration in its editor with test data.
- Start a fresh call and observe it in Live Calls.
- Check transcript, routing, context and tool timing.
- Review the final summary and actual delivery.
- Record pass/fail with the call identifier and configuration change.
Editor tests can contact real systems, send messages or incur usage. Use a test destination that you control.
Improve one problem at a time¶
Change the instruction, source or setting that caused the problem. Repeat the same test and related cases. Check that answers are correct, the request is completed and the waiting time feels reasonable.
Before a broad change, retain the previous configuration so your team can restore it. Use the go-live checklist before moving from controlled calls to production traffic.