Best Practices When Using Tableau Agent

Tableau Agent is a trusted AI assistant that helps new analysts curate, explore, and visualize data faster and more efficiently with conversational AI. By leveraging generative AI and statistical analysis, Tableau Agent understands the context of your data, guiding you through tasks like preparing your data in Tableau Prep, documenting it in Tableau Catalog, and creating visualizations or exploring data in Tableau Cloud Web Authoring.

A key element of its functionality is the collaboration between advanced AI and human oversight. One of the most critical aspects of Tableau Agent—a human is always involved in the process, ensuring that all proposed responses are thoroughly checked before they are accepted.

Tableau Agent Best Practices

Whether you’re prepping a data source or exploring data, by following these best practices you can be confident that assistance is available every step of the way.

Make Sure Data is Well Curated

When using Tableau Agent, well-curated data and well-defined schema increase the likelihood of more accurate responses. To better curate your data, you should hide unnecessary fields, add clear labels and field descriptions, and specify data types, all of which will improve response accuracy and help Tableau Agent pilot provide deeper insights.

For example, if you have a “year” field in your data and want to do time-based yearly analysis, make sure the field is identified as a date rather than an integer.

Additionally, different verticals may define metrics or synonyms differently, so if you have distinct definitions, be explicit in your field descriptions.

Give Explicit and Clear Instructions

Tableau Agent is action-oriented and tries to determine user intent immediately, so try to be explicit about what you want it to do for you. While you don’t need to reference specific field names when conversing with Tableau Agent, it is important to explain your intent.

Example: Instead of telling it to “rank sales by state”, clarify whether you want Tableau Agent to create a calculation to compute rank or render a visualization showing the ranks of state based on sales figures.

Break Down Complex Objectives into Steps

Tableau Agent cannot update a data model and generate a visualization in one single step. To get the best results, break multi-step queries into separate actions. This two step approach improves accuracy and enhances Tableau Agent’s ability to deliver the most relevant insights. Maximizing the benefits of AI involves understanding current capabilities and adapting your queries accordingly. Please note: We plan to address the functionality in future updates.

Example: if you want to analyze “Profit Margin” but only have fields for “Total Revenue” and “Cost,” break this task into two steps:

1. Create the Calculation by asking Tableau Agent, “Create a calculation called Profit”
2. Generate the Visualization by telling it to, “Show me how Profit has varied over time.”

Next, if you want to analyze which product sub-categories had most growth in profit in the last year year, break this task into two steps:

1. Create a calculation by asking Copilot “Calculate the growth in profit for product subcategories over the past year”
2. Generate the Visualization by asking, “Show me product sub-categories with the highest profit growth” 
    

Manage Session Context

Tableau Agent remembers the context of your previous questions and commands up to a certain limit (currently about 32,000 characters). If you want to start fresh and not have Tableau Agent consider your earlier interactions, start your analysis on a new sheet. This will begin a new session.

Example: If you’ve been asking Tableau Agent about sales data and now want to switch to analyzing marketing data without any previous context affecting the analysis, start a new sheet. This ensures it doesn’t carry over any information from the sales analysis.

Simplify Measures for Rendering

It’s common to ask for extremums in data analysis, such as “top products,” “best salespeople,” or “highest quarter.” In these cases, it’s often ambiguous to AI how to evaluate what is the “top,” “best,” or “highest”. Tableau Agent is biased to action and uses its best judgment to pick a measure. If you know what measure you intended, specify it explicitly.

Example: Instead of asking Tableau Agent to show you the “top products,” ask it to show you the “top products using profit.”

Improving AI outcomes with Tableau Agent

It’s important to note that the responses from Tableau Agent may vary. The LLMs (large language models) powering Tableau Agent are non-deterministic, meaning answers to the same question can differ.

We believe in building trust and transparency with our customers, users, and DataFam. By following these best practices in data preparation and analysis, you can improve the accuracy of insights generated by Tableau Agent. However, like humans, AI is not flawless and can make mistakes.

If you encounter inaccurate responses or insights from Tableau Agent, please use the built-in feedback button to share your experience. Our product engineering team utilizes this feedback to address bugs and implement new features to improve accuracy and comprehension of user intent.

Tableau Agent is designed to continuously learn and adapt. As it processes more data and receives user feedback, its accuracy and capabilities will only get better over time.

Embrace the future of data analysis with Tableau Agent and unlock the full potential of your and your data. Watch the demo to learn more about Tableau Agent, exclusively available in Tableau+.