Skip to main content
Documentation
Documentation
Adaptive Planning
Last Updated: 2026-03-13
Reference: Data Exploration: Best Practices for AI Prompts

Reference: Data Exploration: Best Practices for AI Prompts

This feature is only available to Planning Agent customers in the U.S., Canada, Europe, and Singapore. For more information about the Planning Agent, contact your Account Executive.
Effective prompts reduce time to insight. Use these practices to structure queries for data look-ups or complex variance breakdowns
  • Be clear and specific. Use explicit language to avoid ambiguity.
    Example: "Show me 6000 Revenue by Product Category for Q1 2024."
  • Provide context. Add details such as account, time, level, version, dimension, or attribute for precise responses.
    Example: "Detect anomalies in North America Gross Margin Percentage over FY 2024 using Working Budget data"
  • Use keywords as filters. Use words like ‘by’, for’, ‘over’, and ‘between’ to add granularity.
    Example: "Compare Actuals to Budget in Headcount for Operations between January 2025 and December 2025.”
  • Provide onscreen context. To anchor the response, reference specific data elements either visible on your screen or available in your model .
    These examples assume that the Operating Expenses account and the Department dimension are visible on your page:
    • “Summarize Operating Expenses by Department for Q2 2025”
    • “Which Department had the most T&E Expenses over Q2 2025?”
  • Use conversational language. Phrase your prompts as questions and type in natural language.
    Example: “What are the key revenue drivers for FY 2024?”
  • Clarify your expectations. Include constraints or parameters for more targeted responses.
    Example: “List the top three regions with the highest revenue growth over FY 2024.”

Include Contextual Anchors

For most accurate responses and to avoid ambiguity, includes these anchors in your prompts:
  • Account: Use the exact names or codes such as 4100 Product Revenue or 6100 Payroll.
  • Time: Specify the period such as FY 2024, Q3, or Last 6 Months.
  • Levels or dimensions: Define the organizational scope such as Marketing Department or North America Region.
  • Version: Clarify if you are looking at Actuals, Working Budget, or a specific forecast.

Apply General Prompting Techniques

Apply these techniques to improve the AI performance:
  • Specify the format for the response output. Example: "Breakdown Travel Expenses by Department for Jan 2026 and present the results in a table."
  • Use "Chain of Thought" for calculations. If you're prompting for a complex metric, guide the agent through the steps. Example: "Calculate Revenue per Headcount by taking Total Revenue and dividing it by the Ending Headcount for the Sales Level."

Use @ Mention Option

To ensure the Planning Agent identifies the exact account, level, time, or dimension you want to analyze, use the @ mention option:
  • In the prompt, enter the @ symbol to display a selection menu.
  • Select an element type such Account, Level, Time, or a specific dimension or attribute.
  • Start entering the element name for the agent to populate the closest matches from your model.
  • When dealing with elements with similar names (Headcount, Ending Headcount, FTE), select a specific value from the list to provide the exact context needed for an accurate response.

Apply Adaptive Planning-Specific Techniques

  • For accounts or dimension names that contain spaces or special characters, wrap the names in double quotation marks to identify these as single elements.
    Example: Show trends for "Total Salary and Wages" by "Cost Center".
  • Use keywords such as 'by', 'for', 'over', and 'between' to act as natural filters for your data.
    Example: Compare Product and Services Revenue by Department by month for Q1 2026.
  • For a deeper dive into the data, reference the "leaf level".
    Example: To automatically see the most granular data available in your hierarchy, prompt the agent to "breakdown by leaf level".

Refine and Iterate Interactively

Generative AI is an iterative technology. If the first response isn't perfect, refine your approach:
  • Iterate without abandoning: If a response is too broad, add a specific detail such as a specific Region and prompt again.
  • Verify against the source of truth: Before sharing AI-generated summaries and charts in executive presentations, cross-check their validity against your standard matrix reports.
  • Refresh for accuracy: If you just updated data in a sheet or changed report parameters, refresh your browser. The Planning Agent provides the best suggestions based on your most recent refresh.

Manage Complexity

  • The agent can pivot data using multiple dimensions such as by Customer by Month by Region. However, after pivoting the data with two or three dimensions, the visualizations becomes difficult to read. For complex inquiries, limit to the most impactful drivers.
  • The agent can calculate new metrics (that don't exist in Adaptive Planning) on the fly such as "% of Revenue". If the calculation is complex, explicitly name the accounts (instead of using acronyms or short names) you want the agent to use.