Troubleshooting: Virtual Agent Not Predicting Topic When NLU is EnabledIssue Even when the NLU model shows a high confidence score for an utterance, the Virtual Agent (VA) is not correctly predicting or launching the mapped topic.Release All Releases supporting Virtual AgentResolutionFollow these steps to troubleshoot and resolve the issue: Confirm NLU Model and Intent Prediction Open NLU Workbench and verify: The model is published.The correct intent is being predicted for the utterance. Validate Prediction Logs Navigate to the open_nlu_predict_log table. Confirm that the correct intent is predicted in the response field. Check VA Topic and NLU Mapping On the Virtual Agent topic: Ensure the Model and Intent fields are correctly mapped.Verify the topic is published (not in Draft). Verify NLU Bindings Go to the sys_cs_topic_language table: Confirm an entry exists for the mapped intent and language.Make sure only one binding per topic per language exists (no duplicates).Check that the application scope (sys_scope) matches the VA topic’s scope.Confirm the correct domain is set, if using domain separation. Topic Properties Ensure the topic has no restrictive roles or conditions under Topic Properties. If present, validate that these are met by the user invoking the VA. Discoverability Setting Check that sys_cs_topic.is_discoverable = true This flag is relevant starting Quebec and must be set for auto-discovery. Use Virtual Agent Designer to Test Navigate to: Collaboration → Virtual Agent → Designer Use the "Test Active Topics" option to simulate user input.Confirm if the predicted intent matches an active topic. Review Intent Feedback Table Open the open_nlu_predict_intent_feedback table: Review the predicted intent and VA’s handling of it. Check for System Errors Review logs in the syslog table for any relevant error messages during prediction or topic launch. Additional Notes If issues persist, re-train and re-publish your NLU model.Always test with a user role that meets the criteria set on the VA topic.Related LinksNatural Language Understanding of Virtual Agent responses Modify Models