Use AI to actually get good at Excel.

What you'll be able to do

• Turn a plain-English request into a working Excel or Sheets formula
• Debug #REF!, #VALUE!, and #N/A errors by asking the right way
• Clean and reshape messy data without doing it by hand
• Get AI to explain what an inherited spreadsheet actually does
• Sanity-check AI formulas so you never hand up wrong numbers

Inside the path

A focused set of five-minute lessons. Each one ends with a hands-on exercise, not a quiz you can guess.

1. Describe it, don't derive it 5 min

   The prompt pattern that turns "I want the total for each region" into the exact formula.

2. Fix any formula error 5 min

   Paste the error and the formula; learn how to get a fix and an explanation you'll remember.

3. Clean messy data fast 6 min

   Split names, standardize dates, dedupe, and reshape without manual find-and-replace.

4. Understand a sheet you inherited 5 min

   Get AI to explain nested formulas and logic so you can trust and change them.

5. Check before you trust 5 min

   The quick tests that catch an AI formula that's confidently wrong on the edge cases.

Try a sample exercise

This is the kind of card you'd practice inside Iro: you do the thinking, then get feedback.

Why AI is so good at spreadsheets — and where it slips

Spreadsheet tasks are a sweet spot for AI: the goal is usually easy to describe in words, and the answer is a small, testable piece of code (a formula). Describe what you want and your columns, and a model like ChatGPT or Claude will hand you the formula, explain an error, or suggest a cleaner structure.

Where it slips is the details humans skip: it can invent a function that doesn't exist in your version, get a range off by one, or assume your data is cleaner than it is. None of that is a reason to avoid AI; it's a reason to test the formula on a few rows and read what it actually does before you trust it on the whole sheet.

Excel or Google Sheets — the skill is the same

Almost everything here works in both Excel and Google Sheets; you just tell the model which one you're using so it picks the right function names. The transferable skill is the same: describe the goal, give real context (cell references, a sample row, your version), ask for an explanation, and verify. Learn it once and it works wherever your data lives.

A worked example: messy CRM export to revenue by rep

Start by describing the sheet, not the problem. Say you've pulled 8,000 rows out of a CRM: column A is a Date that arrived as text, B is Rep, C is Amount with currency symbols stuck in it, D is Stage. You want revenue per rep per month for closed-won deals. Here's the loop that gets you there in about ten minutes.

1. Give the model the shape of your data

"Excel 365. Headers in row 1, data in A2:D8001. A = Date stored as text like '03/14/2026', B = Rep name, C = Amount stored as text like '$1,240.00', D = Stage." Those two lines remove most of the guessing, and they're the step almost everyone skips.

2. Fix the types before you do any math

Get helper-column formulas that turn C into a real number and A into a real date, plus a way to spot the rows where that conversion fails. Numbers stored as text are the most common reason a perfectly correct SUMIFS comes back as zero.

3. Ask for the formula and the reasoning

"Write a SUMIFS for G2 that totals the Amount helper column for the rep in $F2 where Stage is 'Closed Won' and the date falls in the month in G$1. Use mixed references so I can drag it across and down, and explain each argument." Now you have something you can extend, not just paste.

4. Check it against rows you already know

Filter to one rep for one month and add the amounts up yourself. Matching numbers mean you can drag the formula out across the grid. If they don't match, paste both back and ask what the formula is excluding; blank stages and trailing spaces turn up here almost every time.

What wastes the most time when people use AI on a spreadsheet?

The biggest time sink isn't the model being wrong, it's asking it a question it was never going to answer well. Five habits account for most of the lost hours.

• Asking it to do the arithmetic. Pasting 400 rows and asking for the total invites drift. Ask for the formula instead and let the spreadsheet compute; the tool that's genuinely good at math should do the math.
• Naming the tool but not the version. XLOOKUP, LET, TEXTSPLIT and LAMBDA arrived with Microsoft 365 and aren't in Excel 2019 or 2016. "Excel 2019, no dynamic arrays" in your first line saves a round trip through a #NAME? error.
• Accepting a formula you can't read. If you can't explain the nesting to a colleague, you won't be able to fix it in three months. Get a LET version with named steps, or split it into two helper columns.
• Describing the data as you wish it were. Merged cells, trailing spaces, blank rows in the middle of a range, dates in two formats. Say what's actually there, or paste ten real rows and let the model see it.
• Forcing a formula where a tool fits better. Some jobs belong to a pivot table, to Power Query, or to QUERY in Sheets. Have the model pick the approach before it writes a single formula.

One more worth naming: don't ask for a whole model in a single prompt. Build it one column at a time and check each one as you go. You'll spend your debugging time on a formula you understand instead of a wall of nesting you inherited from a chatbot.