Best Practices for Adding Business Preferences

Formatting and Structure

Here's how to format your business preferences to ensure the Analytics Q&A can process them correctly.

• Clearly separate each preference. Start each business preference on a new line that starts with a '#' symbol. This helps the system understand and process each preference individually.

  Example:

  # This is the first preference

  # This is the second preference

• Keep it short. Write concise, simple instructions to avoid confusing the agent.

Content and Purpose

To help Analytics Q&A perform effectively, follow these best practices when writing your business-specific preferences.

• Add business-specific knowledge. Use preferences to guide Analytics Q&A with information about your business, such as:
  • Domain-specific guidelines.
  • Unique terminology for your organization.
  • Information not included in the semantic model schema.
• Be explicit and specific. Describe the unique logic or concepts of your domain clearly and concisely. Guide Analytics Q&A with specific fields from your semantic model instead of assuming it understands the data model.
• Avoid ambiguity and conflicts. Ensure that you don't have multiple preferences that conflict with and confuse the agent. The more clarity you provide, the better the results.
• Break down complex instructions. Separate complex instructions into simpler, distinct preferences to improve the agent's performance.
• Follow examples. Review the provided examples for guidance on format and content. Adapt these examples to fit your needs. Consider these common patterns.
  • For interpretation

    # ‘Largest’ or ‘smallest’ accounts refer to the accounts with the highest or lowest total_amount, respectively

    # When a user asks about new leads, return leads created in the last week

    # When referring to opportunities, TPS is short for total product sales

  • For output

    # When asking about sales data, always sort by account name

    # When asking about leads, always return the lead's name and status

    # When analyzing deals, always compare to the previous quarter

Considerations and Limitations

When adding business preferences, keep these limitations and behaviors in mind.

• Limited instruction size. You can add up to 50 instructions, with each instruction limited to 300 characters. Keep in mind that adding more instructions increases the size of the agent's prompt, which can result in slower response times.
• Limited scope of impact. Business Preferences currently only affect Analytics Q&A conversations in the analytics agent, not other Tableau Next experiences.
• Preferences are guidance, not commands. Analytics Q&A uses business preferences as guidance, but it's not guaranteed to follow them in every response.
• User input has priority. Analytics Q&A prioritizes a user's direct input over a business preference setting. User input includes:
  • Specific preferences in the user's input.
  • Context from recent conversation history.
  • Filters applied on the page.
• No feature control. You can't use business preferences to turn off, modify, or prioritize specific Analytics Q&A features in the analytics agent.

Testing and Iteration

Optimizing business preferences is an iterative process. Experiment with different instructions and observe the agent's responses to learn what works best for your model and your users.

• Start small: Begin with simple business preferences before adding more complex logic.
• Test your agent: After adding a business preference, test the agent’s response.
• Iterate gradually: Add more business preferences one by one, testing the agent’s response after each addition.

Permissions and System Behavior

Understand how business preferences behave in different systems and environments.

• Preference inheritance: Base semantic models or extended applications don't automatically pass down business preferences. Define them specifically for each semantic model you're working with.
• How preferences move between systems:
  • Packaging and syncing: When a semantic model is packaged in a data kit, its business preferences are included and shared with other connected organizations.
  • Promotion to production: Business preferences move to the production environment along with the semantic model.
  • Resolving conflicts: If a conflict between preferences occurs during a promotion, an admin must select which version to keep.

When to Use a Business Preference

Business preferences are the specific rules, jargon, and logic that a particular team or business unit applies to the data. They add a layer of business context, telling Analytics Q&A and the analytics agent in Tableau Next how to interpret and use the data to answer domain-specific questions.

Key elements of a business preference include:

• Unique jargon: They define company-specific terms (for example, "A 'hot lead' is any lead with a score over 90").
• Business logic: They explain how different data fields relate to each other based on business rules.
• Usage guidelines: They provide instructions on how data should be applied in specific reports or analyses.

When Not to Use Business Preferences

Business preferences provide semantic guidance to the agent, but they don't perform calculations or control an agent’s core behavior. For more complex needs, use the recommended tools instead.

The following table lists business preferences that are unsupported or should be handled differently. The examples provided show what not to do.