Test your AI before going live

Prerequisites

Save your configuration, finish document processing, and prepare expected answers approved by the business. Use controlled customer records for bookings or other actions.

Run a repeatable test

1. Open Config → AI Configuration → Test AI Responses.
2. Ask a known FAQ, a policy question with an exception, and a product question. Compare each with the correct source.
3. Ask for information absent from all sources. Check that the AI does not invent a policy, price, or guarantee.
4. Repeat in English and Arabic, including the language a customer actually uses.
5. Ask for human support and verify the operational handoff on the target channel.
6. Test a tool operation with complete inputs, then missing/invalid inputs. Verify the real remote record rather than trusting a fluent confirmation.
7. Test an unavailable tool/provider and confirm the agent does not falsely report success.
8. Repeat critical tests through a real connected channel. The test panel does not prove provider delivery or channel permissions.

Expected result and recovery

Record question, expected answer, observed answer, source version, and pass/fail. Correct the owning source or tool configuration, then rerun failed cases. Keep affected unattended actions disabled until the result is understood.