Tips for Improving Performance in Concierge: Analytics Q&A

View supported editions.

For a complete overview of the phases required to implement Concierge, see, Plan Your Concierge Implementation in Tableau Next.

Prepare the dataset intentionally

• Design joins carefully. Avoid many-to-many or non-strict joins (such as BETWEEN) that expand result sets and slow queries. Example: If joining Opportunities to Accounts, ensure join keys are unique (such as AccountID) to prevent result explosion.
• Push complexity into ETL. Heavy calculations or derived fields (such as Forecasted Revenue) should be pre-computed in Data Cloud transforms, not recalculated at runtime.

  Example: Instead of computing forecasted opportunity close rate in runtime, compute it in ETL and expose it as a metric.

• Handle large fact tables (>100 million rows) thoughtfully. Pre-compute aggregations (for example, monthly revenue totals) or apply semantic model filters to reduce data scanned at runtime.

  Example: For a Sales fact table with 500 million rows, pre-aggregate ACV by quarter to help Concierge answer "Show me ACV by quarter" without scanning the full dataset.

Follow best practices for semantic model design

• Include only objects and fields needed for the business.
• Streamlined semantic models ensure faster query generation and improve the accuracy of field mappings.

For in-depth guidelines for designing and refining semantic models in Tableau Next, see Design Semantic Models for AI Readiness and Build an AI-Ready Semantic Model.

Pre-define metrics

Define metrics in the semantic model that directly support frequently asked utterances.

• Metrics provide context that can help guide analytics agents for understanding where and when specific logic applies.
• Defining metrics improves latency by avoiding expensive ad hoc query paths and ensures consistency across your org.

  Example: "What is my ACV?" creates the ACV metric once in the semantic model. Future utterances referencing "ACV" benefit from faster routing and consistent logic.

Ask targeted questions

Ask targeted questions rather than open-ended questions in Concierge. Targeted questions reduce the amount of data that must be analyzed, improve query latency, and prevent timeouts on very large datasets.

• Limit the scope of your questions (utterances) to reduce the query scope.
• Ask questions that clearly specify time ranges. Example: "Show me sales for Q1 2025" vs. "Show me sales"
• Simplify questions to include fewer categories, measures, or aggregations.
• Avoid asking for long lists of records. Ask questions that don't require thousands of records to reduce the query scope. Example: "What are the top 10 opportunities?" vs. "Show me all opportunities?"