Research

Generate useful analysis questions for a spreadsheet

Turn a dataset and decision into questions its fields can actually answer

6 min setupTested with:ChatGPTReviewed: 2026-08-28

Add your context

Your text stays in this browser. AILesson Prompts does not send it to a model or server.

Decision or purposeColumns and definitionsKnown data limits

Your prompt

Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt

Propose analysis questions that this spreadsheet can support.

Decision or purpose:
[decision]

Columns, definitions, units, time coverage, and row grain:
[schema]

Known limits:
[limits]

First state what one row represents and which questions cannot be answered. Then propose prioritized descriptive, comparison, trend, distribution, and diagnostic questions. For each, list the decision relevance, exact fields, calculation or PivotTable design, required filters, expected output, and a caveat. Do not propose causal conclusions, forecasts, or group comparisons that the supplied data cannot support. End with the five questions that offer the highest value for the least analysis effort.

Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.

From input to outcome

A worked example

See how concrete context turns this recipe into a usable result

Actual input

Decision or purpose: A support manager must decide where to improve response operations next month.
Columns and definitions: One row per closed ticket from Jan–Jun 2026. Fields: TicketID, CreatedDate, ClosedDate, Channel, IssueType, Priority, FirstResponseMinutes, ResolutionHours, Reopened (TRUE/FALSE), CSAT (1–5 or blank).
Known data limits: Only closed tickets are present. CSAT response rate is 38% and varies by channel. Staffing hours and agent identity are absent. Duplicate TicketID has not yet been checked.

Example output

One row represents a closed ticket, so the file can describe completed-ticket outcomes but cannot measure open backlog, staffing productivity, individual-agent performance, or causal effects of channel. Verify TicketID uniqueness first.

High-value questions include: (1) Which Channel × IssueType groups have the highest median first-response and resolution time? Use medians and ticket counts because long tails are likely. (2) Did those medians change month by month, using CreatedDate month consistently? (3) Which groups have the highest reopen rate, with numerator and denominator shown? (4) How do response and resolution distributions differ by Priority? (5) Where are the extreme delays concentrated? (6) What is CSAT coverage by channel before comparing scores?

For question 1, build a PivotTable with Channel and IssueType as rows, count of TicketID plus median measures calculated separately, and exclude invalid or missing durations. The result guides process investigation, not staffing blame. For CSAT, report response rate beside the score and avoid ranking channels when coverage differs materially.

Best first five: duplicate check; monthly response trend; Channel × IssueType medians; reopen rate by issue; CSAT coverage. They use available fields, directly support prioritization, and require no unsupported attribution.

Why this works

1
Starting from row grain prevents totals and rates from mixing incompatible units.
2
Linking each question to a decision filters out interesting but unusable analysis.