Last lesson, you briefed Deep Research to investigate the world outside. This one turns the analysis inward — onto your own data. Imagine you ran a film photography workshop, and you want to check how it went. You've got the files from running it, and one question won't leave you alone: did it actually go well? You'll answer it with ChatGPT — and learn to catch the moments where the answer looks right but isn't. Let's get started!

The one-day film photography workshop you ran wrapped last weekend, and now you want to know how it really went. You've got three separate files:

Three questions are worth your analysis:

This is the job for ChatGPT's data-analysis capability. You upload your files, ask questions in plain language, and, for some analysis tasks, it writes and runs real code behind the scenes to work the data. And then hands you the result as text, tables, or charts, right in the chat.

The instinct is to ask your big question right away. But first, have ChatGPT tell you what's going on in each file — the columns, the row counts, what every field holds. You can't analyze what you haven't looked at, and a thirty-second orientation saves you from answering the wrong question confidently.

It reports back:

The attendance gap is right there — but your question was sharper than a single rate: which kind of registrant showed up, and which didn't? Answering that means getting the files consistent first.

Before that number means anything, the data has to be consistent. The signup dates and the check-in times are written in different formats. The price column mixes "$90," "90," and a few blanks where students came for free. Sort that out first, or every total you calculate inherits the inconsistency.

You don't fix any of this by hand. You tell ChatGPT what consistent looks like and let it do the work:

You start with attendance, and ChatGPT gives you a clean number: 80% — 51 of 64 registrants came through the door. It's tidy and quotable, but it's half the question you asked.

A single 80% hides the thing you care about. Early Bird buyers might have shown up almost to a person, while half the free Student tickets never appeared, and an 80% average blurs that distinction. The decision you're weighing (whether to keep offering student discounts) lives in the breakdown, not the total. So you ask for attendance split by ticket type.

The breakdown shows Student attendance lagging well behind the paid tiers, which makes your second question urgent: Did the student discount earn its place? You ask for revenue per attendee by ticket type. ChatGPT answers without hesitating, and one figure looks off the moment you read it.

The Student tier comes back with a per-attendee amount that isn't zero, even though the students paid nothing. So what did ChatGPT do with those blank prices when it built the number? Maybe it dropped them, maybe it filled a value. Each choice produces a different "revenue per attendee," and it told you neither.

So you ask again, leaving no room for a silent choice:

Now each tier's figure means exactly what it says — and you can finally judge whether the discount paid off.

Your third question relates two things rather than counting one: are the people who rated the workshop highly also the ones who'd return? The feedback file has both a rating and a "would recommend" response, so you ask ChatGPT to put them side by side and chart how the rating relates to return intent.

The chart suggests high raters do lean toward returning — but only 47 people answered, and once you split them by response, some groups are tiny. That's enough to call a signal worth watching.

You've got attendance by segment, a defensible per-tier revenue figure, the rating-and-return signal, and one open question. The last step is shaping it into something you'd actually send your co-organizer — the verified numbers, the chart, and the one remaining question.

That's the last lesson — and a good place to look back at how far you've come. You started by building tools that hold up under real use:

You moved into creating — an app built from a sentence, a full visual identity from a description. And you finished by investigating: briefing deep research, then turning your own data into insightful findings.

You now work with ChatGPT the way a professional does — building multi-step workflows and connecting tools into systems. That's a different level from where most users stop. Now it's time to put it to work! Good luck!