Skip to content
PromptifyLab

Make sense of an expense spreadsheet

By Aamir1 min read

Find inconsistent categories and questions worth investigating, while keeping the arithmetic in your spreadsheet.

Before you start

A small, anonymised expense table with date, category, description and amount columns. Remove account numbers and identifying details.

The workflow

1. Prepare a clean sample

Start with 20–50 rows rather than your entire financial history. Use consistent date and currency formats. Keep an untouched copy and write down the total in your spreadsheet.

2. Ask for exceptions, not a financial verdict

Have the assistant identify ambiguous categories, possible duplicates and missing values. Require the original row number and the reason for each flag. Similar transactions are not automatically duplicates.

3. Check each flag against the source

Compare every suggested correction with the original record. Recalculate totals with spreadsheet formulas. Do not use a language model’s arithmetic as your only check.

4. Turn findings into an action list

Separate confirmed corrections from questions for the person who owns the records. Record the change and reason, then rerun your spreadsheet totals.

A starting brief

Replace the bracketed fields. Paste it into ChatGPT, Claude, Gemini or another assistant – or press Fill in.

Starter prompt
Review this anonymised expense table for data-quality issues.

Return: row number, issue, evidence from the row, and a question or suggested check.

Look for missing values, inconsistent categories and possible duplicates.

Do not delete rows, assume a transaction is wrong, or give tax or investment advice.

Do not invent totals. Mark uncertain findings for human review.

TABLE: [PASTE A SMALL ANONYMISED SAMPLE]

Open in

Make it concrete

Illustrative check: two rows share the same date and amount. Flag them for review, but keep both until you confirm whether they represent separate purchases.

Check the result

A useful result points to specific rows and leaves a clear audit trail. A confident judgement with no source row is not useful. This workflow checks data quality; it does not determine tax treatment or financial decisions.

Keep going

Sources

Product capabilities are based on these official pages. Workflow steps and sample briefs are editorial suggestions, not benchmark results. Checked 25 September 2026.