Writing effective definitions

Remember these things:

• Summaries are based on the data that the dashlet query returns. The AI does not use data from other dashlets and does not run additional queries.
• The AI analyzes a sample of up to 1,000 rows when a dashlet contains more than 1,000 rows. The summary is based on that sample.
• Summary text supports basic formatting, including line breaks and lists. Summary text does not support rich-text formatting.
• The system returns the same cached summary when the data and definitions remain unchanged. This behavior is expected.
• The AI respects user permissions. The AI generates summaries from data that the user can view.

Even if measures and attributes are not enabled for BQL generation in your Data Model Descriptions, labels and descriptions improve Insight Summaries. The AI uses the additional context to create business-focused summaries instead of using raw system names.

For example, if a measure has the label "Gross Revenue" and the description "Total sales before returns, in USD," the summary displays "Gross Revenue increased by 8%" instead of "Sum_sale_amt increased by 8%."

The examples of good Summary Definition:

• "Summarize the top 3 performing regions by revenue and note any month-over-month changes."
• "Focus on year-over-year comparison and highlight categories with more than 15% growth."
• "Highlight the highest and lowest values, and note the overall trend direction."

The examples of good Anomaly Definition:

• "Flag any category with a decline greater than 10% compared to the previous month."
• "Detect if any region's revenue deviates more than 2 standard deviations from the average."
• "Alert if any product shows zero sales when it normally has consistent activity."