Tableau Agent Best Practices


Tableau Agent does its best to understand the intent of your questions as well as the context of your data and your business. But without data that is optimized for AI, Tableau Agent can get things wrong.

This article describes some steps you can take to make sure you get the best results from Tableau Agent.

General Information

Use clean, well modeled data

Messy data is hard to analyze and Tableau Agent is focused on analysis, not cleaning and prepping your data.

  • Do: Clean and prep your messy data before analysis. Try using Tableau Agent in Tableau Prep(Link opens in a new window) to clean your data.

    • Don't: Expect an analytical agent to do data prep.

  • Do: Use relationships in data modeling.

    • Don't: Hardcode joins

    • Don't: Use data blending

  • Do: Use certified data sources and update them for AI consumption.

  • Do: Hide unnecessary fields. If a user shouldn't build a viz with it, the agent shouldn't see it.

Use clear field names and descriptions

Tableau Agent doesn’t understand the data in the same way a human would. Without good names and descriptions, Tableau Agent may misinterpret a field. Remove any guesswork about how the data fits into an analysis.

  • Do: Use unique, descriptive, and complete field names.

    • Don't: Leave a field name as "Age" when it could be "Case Age at Close"

    • Don't: Leave abbreviations or ambiguous fields like "P_ID" or "Name"

  • Do: Provide a description that defines what the field is.

  • Do: Include the business context and expected synonyms.

    • Don't: Assume all your users will always refer to the field by its official name

  • Do: Describe how the field is typically used in analysis and how it's different from similar fields.

    • Don't: Copy and paste descriptions across similar fields

  • Do: Maintain the descriptions for standard Salesforce fields.

  • Do: Document any unique business logic or data anomalies (such as how to interpret nulls, or when to use "Invoice Date" rather than "Purchase Date").

Pre-define calculated fields

While Tableau Agent can help create calculations through natural language, if a user asks for a calculation incorrectly or vaguely, an incorrect calculation can result.

  • Do: Create any common calculations ahead of time so you can ensure consistency and accuracy.

Specify field properties

The agent relies on field properties as analytical guardrails. Default or misconfigured properties can lead to poor analytical outcomes.

  • Do: Verify field roles (dimensions and measures) and data types (such as date, numeric, boolean).

    • Don't: Leave a field such as Postal Code as a numeric field (measure) if it should be considered an attribute (dimension).

    • Don't: Leave dates stored as text.

  • Do: Set a default aggregation.

    • Don't: Assume the user or agent will know to use average with "unit price" rather than sum

AI readiness checklist

Use this checklist to implement best practices when preparing your published data source and workbooks for AI consumption.

Model Structure

Check box Clean data

Check box Relationships instead of joins

Check box No data blending

Fields

Check box Field names are specific and contextual

Check box All fields have useful, differentiated descriptions

Check box Unnecessary fields are hidden

Check box Common or expected calculations already exist

Check box Fields have the correct properties: role, data type, default aggregation

Back to top
Thanks for your feedback!Your feedback has been successfully submitted. Thank you!