How to Use Tableau Agent for Data Analysis

Tableau Agent turns a blank worksheet into a working visualization from a single typed question. You connect to a data source, open the Agent panel from the toolbar, and describe what you want to see. The feature is available in Tableau Desktop 2025.1 and later, in Tableau Cloud web authoring with a Tableau+ subscription, and in Tableau Server 2025.3 and later. A free trial gives you access to the full feature set without a paid license.

What Tableau Agent Actually Does

When you start a session, Tableau Agent indexes your data source. It reads field names, data types, and samples up to 1,000 unique values per field. That index is what allows it to interpret a natural language request and map it to the correct dimensions, measures, and chart types in Tableau.

From there you can ask it to:

• Build a visualization ("show me the top 10 customers by order value last quarter")
• Choose the right chart type ("show me the distribution of delivery times")
• Perform time series analysis ("which month had the largest growth in new signups versus the prior month?")
• Create calculated fields ("calculate the difference between order date and ship date, rounded up to weeks")
• Explain an existing calculation in plain English
• Filter and sort data inline ("only show the Northeast region and sort by revenue descending")

In February 2026 Tableau launched Dashboard Narratives as a beta feature. Once enabled by a site admin, it lets any dashboard consumer click a button to get a plain-English overview of what a dashboard contains, plus a ranked list of the top insights across every chart on that dashboard. This is the first time Tableau Agent can operate at the dashboard level rather than the individual sheet level.

Requirements and Access

Tableau Cloud requires Tableau+ and the AI in Tableau setting turned on for web authoring by a site administrator. If you do not have a paid subscription, the free trial at tableau.com/products/trial includes Tableau Agent and is the fastest way to test it on your own data.

Tableau Desktop (version 2025.1 or later) activates Tableau Agent when you are signed into a Tableau Cloud site that has Tableau+ enabled, or a Tableau Server site with AI configured. The Agent icon appears in the toolbar next to Show Me. If the icon is grayed out, confirm that AI in Tableau is turned on for web authoring in your site settings and that you are connected to a data source on a worksheet tab.

Tableau Server (version 2025.3 or later) supports Tableau Agent through a direct connection to OpenAI using your own API key. Unlike Tableau Cloud, Server does not route through the Einstein Trust Layer, so PII masking is not applied automatically. Sensitive data handling is the responsibility of the OpenAI account configuration in that setup.

Supported data types: extracts (.hyper), CSV, TXT, XLSX, and live database connections. Cube data sources are not supported. When using data blending, Tableau Agent operates only on the primary data source.

Step-by-Step: First Analysis with Tableau Agent

Step 1: Prepare your data. Tableau Agent performs best when field names are descriptive and distinct. If your spreadsheet has columns named "Col1" or "Measure," rename them before connecting. Remove fields you will not use or hide them in Tableau before opening the Agent panel, because Tableau Agent will not reference hidden fields.

Step 2: Connect your data source. Open Tableau Desktop or navigate to Tableau Cloud web authoring. Connect to your data source. For best performance, use an extract rather than a live connection. You can upload a CSV or Excel file directly from the Tableau Cloud web authoring start page.

Step 3: Open Tableau Agent. Navigate to a worksheet tab. Click the Tableau Agent icon in the toolbar (next to the Show Me button). A conversation pane opens on the right side of the screen.

Step 4: Ask your first question. Type a specific, concrete question about your data. Good examples: "What were total sales by product category in Q1 2026?" or "Show me which sales representatives had the highest close rate last year." Vague prompts like "analyze my data" produce weaker results. If Tableau Agent does not know how to handle a high-cardinality filter value, it will ask you to select the specific values you want.

Step 5: Iterate with follow-up prompts. Tableau Agent treats each worksheet as a single conversation. You can refine the chart by adding follow-up messages: "Only include the top 5 categories" or "Switch to a bar chart sorted by value descending." The conversation persists for the current session but clears when you close the workbook.

Step 6: Use Dashboard Narratives (beta). If your site admin has enabled Dashboard Narratives in Tableau Cloud, open any dashboard that uses a published data source and click the Tableau Agent icon in the toolbar. Select "Get overview" for a plain-English summary or "Get insights" for a ranked list of findings across all charts on that dashboard. This is useful for sharing a dashboard with stakeholders who need a text summary alongside the charts.

Getting the Best Results

Three things consistently improve Tableau Agent output quality. First, set the correct data type and field role for every column before starting a session. A date column tagged as a string will confuse date-based analysis. A numeric field tagged as a dimension will not aggregate correctly. Fix these in the data source before opening the Agent panel.

Second, break complex requests into discrete steps. Ask for a calculated field first, confirm the formula looks correct, then ask for a visualization that uses that field. Asking for both in one prompt increases the chance of an error in either the calculation or the chart.

Third, be explicit about aggregation. Instead of "show me revenue," ask "show me the sum of revenue by month." Tableau Agent can infer aggregation from context, but specifying it eliminates a common source of incorrect results.

According to Tableau's internal testing data, users who provide descriptive field names and use extracts rather than live connections see measurably faster response times during the indexing phase at session start. On live connections to large databases, the initial indexing step can take 20 to 40 seconds for data sources with more than 50 fields.

What Tableau Agent Cannot Do

Tableau Agent does not pick a data source for you, perform data modeling or joins, build dashboard layouts, answer general consulting questions ("how should I structure my analysis?"), or create interactive elements like parameter controls and action filters. It is designed for ad-hoc exploration and initial visualization authoring, not full dashboard construction.

It also does not retain memory across sessions. Closing a workbook clears the conversation. If you reopen the workbook later, Tableau Agent starts fresh with no knowledge of what was built in prior sessions. This differs from tools like VSLZ AI, which retains full context between sessions so you can pick up mid-analysis without re-explaining your dataset.

Multilingual Support

Starting in 2025.1, Tableau Agent supports English and a subset of additional languages. In 2026, Salesforce added support for Dutch, Swedish, Thai, Simplified Chinese, and Traditional Chinese. If your site or desktop application is set to a language outside the supported list, responses default to English. When using a non-English language setting, use the exact field names from your data source in your prompts to avoid mismatches.

Security and Privacy

In Tableau Cloud, all Tableau Agent requests route through the Einstein Trust Layer. No customer data is used to train the underlying LLM. Conversations are not retained after the session ends. The Einstein Trust Layer also supports PII masking, which can be configured to redact sensitive field values before they are sent to the model.

In Tableau Server using a self-managed OpenAI connection, the Einstein Trust Layer is not used. Data passes directly to OpenAI's API, and data governance depends on the OpenAI account's settings and retention policies.

Summary

Tableau Agent reduces the time from connecting a data source to producing a first visualization. The setup path for Tableau Cloud is sign up for the free trial, connect a data source, and start typing questions in the Agent panel. For Tableau Server, an admin needs to configure the OpenAI connection first. The February 2026 Dashboard Narratives beta extends the capability from individual sheets to full dashboards, making it practical to generate plain-English summaries that non-technical stakeholders can read alongside the charts.