Go-live checklist¶
A successful test call shows that the basics work. Before using the agent with real callers, verify the complete outcome, including failure handling and the team's follow-up.
1. Define a successful call¶
Write down what must be true after a successful call. For a service request: the correct name, confirmed callback number and reason reach the intended team exactly as configured.
Assign someone to review calls and maintain the instructions, knowledge and destination details.
2. Run the test set¶
| Test | Expected result |
|---|---|
| Ordinary request | Correct information and intended action |
| Caller corrects a detail | Latest confirmed value is used |
| Missing or withheld phone number | Agent asks for a callback number when needed |
| Unknown question | Agent states its limits and follows the fallback |
| Request for a human | Configured forwarding or callback process |
| Recipient does not answer | Clear outcome and useful fallback |
| API or lookup fails | No invented success or missing-data assumptions |
| Outside business hours | Intended route in the configured timezone |
| Caller hangs up early | Post-processing behaves as configured |
Use testing and optimization to record results. Repeat affected cases after each change.
3. Check configuration¶
- The Persona describes identity and tone. The prompt includes it with
{persona}. - The prompt explains the task and what to do if information is missing or an action fails.
- Recognition and voice work for the intended languages.
- Knowledge sources are current and answerable questions were tested.
- Tool prerequisites, credentials and actual destinations are correct.
- Processing profiles are assigned to the intended Agents or Flows.
- Routing, timezone and overrides match operating hours.
- The receiving team has verified messages and transfers.
- Access and personal-data handling match your operating policy.
4. Observe the first calls¶
Use Live Calls when you need current behavior and Call History for the completed result. Review a small sample promptly and fix recurring failures before increasing use.
5. Change one thing at a time¶
Note what you changed and why. Start a new call and use the same questions as before to see whether the result has improved.
Next: Troubleshooting connects symptoms to the right configuration area.