Why this matters
Every register, bank and online shop exports a spreadsheet, and almost nobody reads it. So decisions about stock, staff and budgets get made on a hunch, and the monthly report someone does build takes an afternoon of pivot tables.
ChatGPT can read twelve hundred rows in seconds and answer in plain words. The risk is trusting it too early: a duplicate order counts a sale twice, a blank cell skews a comparison, and the answer looks just as confident either way.
In this module you learn the order that makes the answers safe to use: make ChatGPT check the file first, decide the fix yourself, ask specific questions as tables, then finish with three takeaways and one thing to double-check before you act.
What you’ll build
A sales review for Maple Hollow Garden Center, a made-up small business with six months of sales in one CSV file: 1,248 orders across 7 columns (order_id, date, category, product, channel, units, revenue).
ChatGPT finds the two problems before any analysis: 3 orders that appear twice and 12 orders with a blank channel. You have it remove the duplicates, label the blanks Unknown and hand back the cleaned file (1,245 rows). Then you get monthly revenue and category growth as tables, a line chart you can download, and three takeaways, each with the number behind it, such as Plants making 45% of revenue.
What you need
- Nothing to start the practiceChatGPT is a practice copy, and the sales CSV is already waiting in a practice file browser.
- A ChatGPT account, to do it for realFile upload works on the Free plan, with limits on how many files you can upload. For company data, use your company’s approved plan.
- A sales export as a CSV or Excel fileMost registers, banks, accounting tools and online shops have an Export or Download button on their sales or transactions report. One row per sale, one column per detail.
- A practice file, if you want one firstThe sample CSV below has the lesson’s first 14 rows with a blank channel and a duplicate planted in it, so you can try the check in real ChatGPT.
Concepts
4 ideas you’ll use in the practice.
Check before you ask
Before any analysis, ask ChatGPT what each column means, how many rows there are, the date range, and whether there are duplicates or blank cells. Tell it: "Don’t give me any insights yet."
Insights built on a dirty file look right and are wrong. Two minutes of checking is cheaper than one bad decision.
You decide the fix
ChatGPT finds the problems; you say what to do with them: remove the duplicates, label the blanks Unknown, keep the rest. Then ask for the cleaned file back, so you have the same data it is using.
Specific questions, as tables
"Revenue by month" and "which category grew most from January to June" beat "what does this tell you?". Tables make every number easy to check.
For questions like these ChatGPT writes and runs a little code over your rows, so the totals come from the data, not from guessing. Still check one or two totals against your own report.
End with what to do
A chart is a picture. Three takeaways, one sentence each with the number behind it, are something your team can act on. Ask for one thing to double-check too, so a one-off dip does not become a bad decision.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- The five prompts (.txt)Check, fix, ask, chart and takeaways, with what each column in the lesson’s file means.
- Sample sales file (.csv)The lesson’s first 14 orders with one blank channel and one duplicate order planted, to practice the check.
Quick check
3 questions to check you’ve got it.
1. You upload a sales file. What do you ask first?
- Which month was best?
- Check the file: columns, row count, date range, duplicates and blanks, and no insights yet
- Make a chart
Show the answer
Check the file: columns, row count, date range, duplicates and blanks, and no insights yet. A duplicate or a blank changes every answer after it, so find them before you ask anything.
2. ChatGPT finds 3 duplicate orders. Who decides what to do with them?
- You
- ChatGPT, automatically
- Nobody, leave them in
Show the answer
You. It finds the problem; you choose the fix and ask for the cleaned file back.
3. Before you cut June’s stock order because June was 30% down, what should you do?
- Act on it right away
- Check it against last June and your own records
- Ask for a bigger chart
Show the answer
Check it against last June and your own records. A dip can be normal for the season. Check the number against your own records before acting.