Data Q&A Agent Considerations and Troubleshooting

Considerations

• The agent is read-only. It can't modify data, create or alter objects, or execute DDL statements.
• The agent respects your data access permissions. You can only query data that you have permission to see based on your user permissions and Data 360 data access settings.
• Always validate critical business decisions by reviewing the generated SQL query and the agent's assumptions. The agent's confidence score indicates how certain it is about its interpretation of your question.
• If the agent's interpretation doesn't match your intent, rephrase your question or provide more specific details about fields, time ranges, or filters.
• Responses are influenced by Metadata Knowledge Graph and Semantic Enrichments Indexing status. If metadata changes haven't been indexed yet, the agent may not be aware of new fields or objects.
• Add semantic enrichments to improve the agent's understanding of your business context, terminology, and calculation methods. See Guide Data Q&A with Semantic Enrichments.

Troubleshoot Data Q&A Agent

If the agent provides unexpected results or fails to answer your question, check the following:

Check Data Q&A Indexing status

From Setup, search for Data Q&A Indexing in Feature Manager and verify it's enabled. Indexing runs every 2 hours when metadata changes. If you recently added fields or objects, the agent sometimes isn't aware of them until the next indexing run completes.

Review active semantic enrichments

Conflicting or overly broad semantic enrichments can cause the agent to misinterpret your question. Go to Semantic Enrichments and filter by Active status. Check if any enrichments contradict each other or apply too broadly to your data model.

Verify field mappings

If the agent can't find the data you're asking about, verify that your objects are correctly mapped and that the field names match what you're referencing in your question.

Rephrase your question

If the confidence score is low, try rephrasing your question with more specific field names, explicit time ranges, or clearer filter conditions. For example, instead of "show me revenue," try "show me total amount_usd__c from Opportunity where stage equals Closed Won." Or define a semantic enrichment that outlines API name relationships.

Check the generated SQL query

Click View More Details to see the SQL query the agent generated. Look for incorrect table joins, missing filters, or field name mismatches. If the SQL is wrong, the issue may be a semantic enrichment providing incorrect guidance.