A clear prompt helps a Board AI Agent understand the analysis you need, the data scope to use, and the format of the answer.
Use these best practices to write clear and repeatable prompts in the Board AI Agent chat.
This article covers:
You don't need to apply every best practice in this article to write an effective prompt. Use the recommendations that best fit your task.
What makes a good prompt
A good prompt gives the Agent enough context to answer with the right data and the right level of detail.
Write prompts that include:
The business goal.
The time period.
The version or scenario, such as Actual, Budget, or Forecast.
The Entities to analyze.
The measures or KPIs (key performance indicators) to use.
The expected output, such as a summary, table, chart, or ranked list.
Avoid vague prompts that do not define scope or output.
Use the RTCROS checklist
Use the RTCROS checklist to write complete prompts.
Role. Tell the Agent which business role or perspective to use.
Task. Explain what the Agent must do.
Context. Add the business context, such as time period, scenario, or Entity scope.
Rules. Define ranking rules, materiality rules, formulas, or sign conventions.
Output. Tell the Agent what format to return.
Steps. Ask the Agent to show the analysis steps when needed.
Use a clear prompt structure
A structured prompt is easier for the Agent to follow.
Use this pattern:
Goal.
[Explain the analysis goal.]
Scope.
[Define time period, version, scenario, Entity scope, and relevant measures.]
Rules.
[Define calculations, sign conventions, ranking rules, or materiality thresholds.]
Output.
[Define the format of the answer.]
Example
Goal.
Analyze the cash flow variance between Actual and Budget.
Scope.
Use the current Screen Selection. Compare the current month with the same month in the prior year.
Rules.
Rank the top 5 drivers by absolute variance. Use the standard sign convention from the linked Dataset.
Output.
Return a short summary and a table with driver, variance, and explanation.
Define Selections and scope
Tell the Agent which Selections to use when the scope matters.
You can define scope with:
Time period.
Version or scenario.
Entity scope, such as legal entity, region, country, product, or business unit.
Currency or reporting view.
Measures or KPIs.
Selections changed manually in Play Mode are not applied.
Define ranking and materiality rules
Use ranking and materiality rules when the Agent must identify what matters most.
For example, ask the Agent to:
Rank the top 5 variances by absolute value.
Focus only on variances greater than a defined threshold.
Separate positive and negative variances.
Explain only the drivers that have the largest impact.
Clear ranking rules help the Agent avoid long answers with low-impact items.
Specify formulas and sign conventions
Specify formulas when the calculation could be interpreted in different ways.
For example, define whether variance means:
Actual - Budget
or:
Budget - Actual
Also define sign conventions when needed. For example, specify whether a positive variance is favorable or unfavorable.
Use Dataset Attribute descriptions and TXT guidance files to help Agents understand customer-specific KPI definitions, acronyms, and business terminology.
Define the expected output
Tell the Agent what type of answer you need.
You can request:
A short summary.
A table.
A chart.
A ranked list.
A step-by-step explanation.
A narrative for review or presentation.
Example
Return the answer as a table with 4 columns. Use Entity, Actual, Budget, and Variance. Add a short explanation below the table.
Request tables and charts
You can ask the Agent to return data in a table or chart when that format helps the analysis.
For tables, define the rows, columns, measures, and sorting rule.
Example
Show a table by Legal Entity and Month. Include Actual, Budget, and Variance. Sort by the largest absolute variance.
For charts, define the chart goal and the measure to visualize.
Example
Create a chart that shows monthly revenue trend for Actual and Budget for the selected year.
Refine an answer with follow-up prompts
You can use follow-up prompts to narrow or expand the answer.
For example, ask the Agent to:
Drill into one Entity.
Explain one driver in more detail.
Change the time period.
Change the output from a summary to a table.
Add a chart.
Compare the result with another scenario.
Follow-up prompts are useful when you want to continue the same analysis without restating all context.
For guidance on how to write Suggested Prompts that Board AI Agents can display to End Users, see Create and configure Board AI Agents.