A request such as “analyze this month's revenue” is not ready for analysis. You need to know who uses the result, what decision they face, how metrics are defined, the unit of analysis, and exclusions.
A minimum analysis brief includes the decision question, metric and formula, grain, comparison period, filters, source of truth, freshness, assumptions, and output format. AI may surface gaps but must not invent business definitions.
When AI supports SQL, ask for both the query and validation tests: row counts, duplicate keys, missing values, reconciled totals, and edge cases. A query is not correct merely because it runs.
Your output is a one-page brief confirmed by the requester before analysis begins. Preventing rework is more valuable than generating SQL quickly.
Try it now
- 1Choose one real analysis request in your queue
- 2Write the decision to support and the output user
- 3Confirm the metric, grain, comparison period, filters, and source of truth
- 4Ask AI to flag ambiguity and propose clarification questions
- 5Have the requester confirm the brief before querying data
Act as an analysis-brief reviewer. For this request [request], ask only the questions needed to confirm the decision, user, metric, grain, comparison period, filters, source of truth, freshness, assumptions, and output. Do not write SQL until the critical fields are confirmed.
Practical AI for Data Analysis Workbook
Use the fill-in canvas on page three to complete this lesson with your own work task.
Download the workbook ↓Ready for the deeper practical lessons?
Return to the course overview and register. Aptus AI will send five deeper lessons over 14 days, each with a real-work assignment and quality check.
Join the free email course