Plan Your Tableau Agent Implementation for Conversational Analytics

Define how Tableau Agent can help you do your work

Think of your day-to-day experience as an analyst. What inquiries about data do you find yourself answering frequently? What questions and answers do you wish an agent could handle for you?

As the analyst, you're the expert on your data, your business, and the types of questions an analytics agent must answer. Think of Tableau Agent in Tableau Next as a junior analyst who you're mentoring. The time you put into gathering use cases, getting the right data, enriching your semantic models, and providing context for the agent will make this analytics agent a highly valuable asset that ultimately ends up saving you time in the future.

Before you turn it on, consider the best use for Tableau Agent in your business. The Data Analysis subagent that powers Tableau Agent is highly effective at exploring and answering "what" and "why" questions about data. It can answer the most commonly asked data questions and save time for your analysts.

1. Identify common analytics questions for the agent. Tableau Agent is best suited to answer general analytical questions about your data.

   • Trends: Questions about growth or changes over time, often broken down by region, portfolio, or product line.
   • Rankings and KPIs: Queries to rank performance, such as identifying the best-performing campaigns or sales representatives.
   • Comparisons: Analyzes how different factors, like products or accounts, contribute to deals compared to previous periods.
   • Breakdowns: Explores how the total is broken down by different categories.
   • Follow-up Reasoning: Ask a series of sub-questions to drill deeper into an analysis.
2. Collect and create a list of example test questions (utterances) for Tableau Agent. What are common business questions your organization must be able to ask about data? Gather questions from business users, use your own experience, or have an LLM generate a list of likely queries.
3. Determine the data you need for the questions and use cases you have chosen. This process includes making sure you are connected to the data you need in Data 360. The data may require some transformation, either in the source data (for example, Snowflake) or in Data 360.

Set up Tableau Agent in Tableau Next

In this phase, you'll configure users, turn on organization features, and create and activate an agent for analytics and visualizations in Tableau Next.

1. Configure Tableau Next for Agentforce and AI. This step includes enabling Tableau Next, Einstein, and Agentforce and assigning user permission sets.

2. Turn on settings for agentic analytics in Tableau Next. These settings make the Analytics and Visualization agent template and its topics and actions available in Agentforce.

   Optionally turn on settings to support building an AI-ready semantic model.

3. Create an agent with the Analytics and Visualization V2 template in Agentforce Studio.

For detailed steps on this phase, see Set Up Tableau Agent in Tableau Next and Create a Tableau Agent for Conversational Analytics.

Make your semantic model AI-ready to provide context for Tableau Agent

In this phase, you'll follow best practices and use tools available in the Semantic Model Builder to get the semantic model ready for use with an analytics agent.

Your goal in designing the semantic model is to provide business context for the agent, and to make it easy to explain questions, answers, and the semantic model without assistance. If a human analyst who is new to the semantic model can understand it without assistance, Concierge will likely be able to do the same.

1. Become familiar with the concepts in Design a Semantic Model for AI-Readiness. This topic describes best practices to consider when building a semantic model.
2. If you haven't already, add and build a semantic model in Tableau Next. This includes adding data model objects (DMOs) and relationships, either manually or with AI assistance. If needed, create calculated fields. You can do this manually, or with AI assistance using Einstein or Data Pro in the Semantic Model Builder.
3. Define metrics in the semantic model for high-impact KPIs and highly-used calculated fields or field measures. Also, review any visualizations, dashboards, and metrics you have created in Tableau Next based on the semantic model.
4. Add descriptions to reduce ambiguity in objects and calculated fields. You can do this manually, or using AI-generated descriptions.
5. Add business preferences to define business-specific knowledge within the semantic model.
6. Enable the semantic model for use with an analytics agent created from the Analytics and Visualization template. This step is required to be able use the analytics agent.
   

   Note Admins can create fallback lists of semantic models per agent for scenarios where the semantic context isn't clear. For more information, see Set Up Agent Scoping for Tableau Agent.

For detailed steps on this phase, see Build an AI-Ready Semantic Model.

Optimize, test, and calibrate the semantic model

Implementation is an iterative process rather than a linear one. During every step of defining and refining the semantic model, use the feedback available in the Semantic Model Builder to check your results, review your assumptions, ask new questions, make adjustments, and try again. As you fine-tune your approach, each improvement can help lead to better accuracy and responses.

The Semantic Model Builder can offer suggestions for optimizing different parts of the model. Manual testing helps you determine if the model is complete and generates the queries you are expecting. Q&A Calibration makes it easier to test questions in Tableau Agent and create a list of verified questions to improve agent accuracy.

During this phase, optimize, test, and refine the semantic model to recognize questions and provide meaningful answers about the data using Tableau Agent. The goal of this process is to improve the accuracy of responses through a continuous cycle of testing and refinement.

• Use Semantic AI Model Optimization (Beta) suggestions to improve semantic model health. This tool runs a series of automated checks against your semantic model and calculates an overall model health for AI-readiness.
• Use Q&A Calibration (Beta) to test questions in Tableau Agent and classify responses as Verified or Inaccurate. Where a response is inaccurate, investigate the reasons suggested by Q&A Calibration. You can also review the query logic used for the response. What query did it use? What parts of the query are wrong? Next, you can calibrate your model using the suggested calibrations, tailored to your questions. Calibrations include adding or updating business preferences, fields, and calculations.

Note You must have analysts and administrator permissions to use Semantic AI Model Optimization and Q&A Calibration.

As you make changes to the semantic model, test questions again in Tableau Agent and continue testing and updating until you get the responses you expect.

For detailed steps on this phase, see Optimize and Test the Semantic Model.

Progressively roll out Tableau Agent in your organization

When you are done with the testing phase, start progressively rolling out Tableau Agent in your organization to learn more about the actual questions that people ask—not just what you thought they'd ask.

1. Start with 2-5 coworkers who enjoy testing analytics software solutions in your organization. For example, start with your analyst colleagues. Note what questions they ask the agent and how it answers. Are the answers Tableau Agent provides useful?
2. Expand your set of test questions over time as you see the types of questions real users are asking.

   As you continue to test and get improved results, try a pilot with 20-50 people in your organization. Continue to monitor the questions, and test and calibrate the semantic model.

   Add common questions with accurate responses to Q&A Calibration and mark them as Verified. This helps the agent learn and respond to similar questions more accurately, in language that is tailored to your organization.

3. When you feel confident about how Tableau Agent is performing, try rolling it out to an entire group or department, and then the whole organization.

   As you roll out Tableau Agent, give your users guidance and context to help them be successful when asking questions. Educate them about what kinds of questions they can ask and what types of questions aren't supported. For more information, see Ask Questions with Tableau Agent in Tableau Next and Supported Questions and Surfaces for Tableau Agent.