Troubleshooting for Conversation Improvement Themes

Troubleshooting steps for errors that might occur in Conversation Improvement Themes.

Scenario/Symptom: The dashboard is not showing any data, one week after the install.

Associated Error Codes/Warning: N/A.

Steps to Resolve:

Make sure all the skills and scheduled jobs are active by following the steps given in Activating Conversation Improvement Themes.

Scenario/Symptom: Execution staging records are stuck on Processing/Failed.

Associated Error Codes/Warning: Flow-related logs in syslog.

Steps to Resolve:

Increase flow logging (preferably in dev).

Environment guidance:

• Dev/Test: Set verbose logging to All configurations and runtime value.
• Prod: Use Runtime states and duration only. Don't enable All configurations and runtime value in production.

Steps:

1. Navigate to Process Automation > Flow Administration > Properties.
2. Set Level of reporting data generated by the flow engine to the appropriate level (see the guidance above).
3. Create a conversation through the Portal and end the conversation.
4. Navigate to Process Automation > Flow Designer > Theme Tagging(subflow) or Primary Request Tagging(subflow).
5. Review flow executions. Identify any actions marked Failed or Error and address the cause.
6. If all actions are Complete and there is no other path indicating failure handling, open a Support ticket.

Scenario/Symptom: Execution staging records Failed when default model is Gemini/Claude/Now LLM.

Associated Error Codes/Warning: Skill Execution timed out.

Steps to Resolve:

This is a known issue with Gemini and Claude as they are slow thinking models, responses may be slow and sometimes can get timed out. To resolve this, reduce the chunk size in system properties by performing the following steps:

1. Navigate to All > System Properties (sys_properties.list) and search for sn_na_thematic.dataChunkSize and reduce it. The ideal chunk size for all the models is:
   • Azure Open AI: 100
   • Gemini: 25
   • Claude: 5
   • Now LLM: 1
2. Move the failed Execution staging record to processing.

Generative AI Logs: If an evaluation is not complete, navigate to sys_generative_ai_log and check for the following definitions to see if there are any errors:

• Theme Extraction
• Themes Update Definition
• Themes Primary Request Tagging

Outbound HTTP:

If no Gen AI logs are created, check the Outbound HTTP logs for any issues exist.

Sys Logs: Check sys logs for any errors related to passing with scope as sn_na_thematic.