Describe the calculation you want in plain English and get a working Excel or Sheets formula back — then make the AI explain it cell-by-cell, so you never paste a formula into a live workbook that you cannot defend in a meeting.
Workflow area: Spreadsheets & Data · Tier: Foundation · Hands-on time: ~25 minutes
Formulas are where office AI pays off fastest: the syntax is fiddly, the documentation is dense, and a model translates "average deal size per region, but only closed deals" into AVERAGEIFS in seconds. The discipline that separates a power user from a future incident is the second half of this playbook — you never paste a formula you don't understand, and you always test it against a case where you already know the answer.
#N/A, #REF!, #VALUE!) and the fix usually comes back with the reason.Describe the sheet, then the goal. The model cannot see your workbook. Column letters, what each holds, and one example row beat any amount of adjectives:
I am in [Excel / Google Sheets].
My sheet:
- Column A: order date (e.g. 14-Jul-2026)
- Column B: region (North / South / East / West)
- Column C: status (Won / Lost / Open)
- Column D: deal value in ₹ (e.g. 250000)
In cell G2, I want: the average deal value for the region named
in F2, counting only "Won" deals from the last 90 days.
Give me the formula, then explain each part in one line each,
then tell me one edge case where it could give a wrong or
misleading answer.
Read the explanation before the formula. Good output:
=AVERAGEIFS(D:D, B:B, F2, C:C, "Won", A:A, ">="&TODAY()-90)
- AVERAGEIFS(D:D, ...) — averages deal values where ALL
conditions match
- B:B, F2 — region must equal the value in F2
- C:C, "Won" — status must be exactly "Won"
- A:A, ">="&TODAY()-90 — order date within the last 90 days
Edge case: if no deals match, it returns #DIV/0! — wrap in
IFERROR to show "no data". Dates stored as text will silently
fail the date condition.
The edge-case line is the professional's habit: every formula has one, and asking for it up front means you find out before the review meeting.
Test against a known answer. Before trusting the formula, verify it on a case you can check by hand: filter the sheet to one region manually, eyeball the matching rows, and confirm the formula's number. One verified cell buys confidence in ten thousand.
Decode before you modify. For inherited formulas, reverse the direction:
Explain this formula step by step, in plain English, assuming I
know basic Excel but not these functions. Then tell me: what
breaks if rows are added above row 5?
[PASTE FORMULA]
Never edit a formula you cannot narrate. If the explanation surprises you, the workbook's author may have made the same mistake — check the output, not just the syntax.
Ask for the robust version. Once the formula works, one more prompt: "Make it robust: handle empty cells, wrap errors with a clear message, and note if a newer function (XLOOKUP, FILTER, LET) does this more cleanly." Modern functions are usually simpler and easier for the next person to read — and the next person is often you.
ARRAYFORMULA, QUERY and IMPORTRANGE are its home turf where Excel answers differ — always state which app you are in.Formula questions rarely need real data — column descriptions and fake example rows work just as well. If you do paste rows into an external tool, use dummy values for anything financial or personal; pb-12 has the full pre-flight.
Keep a "formulas I now understand" page in your prompt library (pb-05) — the formula, the plain-English explanation, and where you used it. It becomes your personal function reference, written in your own examples, and the fastest onboarding doc your successor will ever get.