Start with the job to be done

The working assignment is to build and validate a June regional-revenue formula in Excel 365 for an analyst who must fill the formula down safely. That sentence is narrower than “use AI for Excel.” It identifies a deliverable and a reviewer. Write a definition of done with three layers: the output must satisfy the audience's need; factual or functional claims must be traceable; and a named person must own the final decision. Keep raw columns untouched. Use helper columns for cleanup so every transformation is reversible, and never accept a formula whose result you cannot reproduce on five known rows.

Start by separating tasks. The model may draft, classify, transform, compare, or suggest. It may not silently approve, publish, grade, deploy, cite, or consent on someone's behalf. For this assignment the authoritative material is column letters and headers, three sample rows, desired cell, Excel version, locale separators, date boundaries, and blank/error behavior. Anything absent from those inputs is either an explicit assumption or an unanswered question.

Use this prompt as a realistic starting contract:

Excel 365, comma separators. A=Date, B=Customer, C=Region, D=Revenue; F2 contains a region. In G2, total D where C=F2 and A is in June 2026. Give one formula, explain every condition, and provide normal, boundary, and blank/error tests. Do not invent columns.

Notice what the prompt does: it states the setting, limits the output, names forbidden behavior, and requests evidence that can be reviewed. It does not ask the model to “make it amazing.” If a constraint matters, make it testable. Replace “be accurate” with a source boundary, formula check, test command, rights ledger, or approval step.

A useful response would look like this: A SUMIFS formula using >=DATE(2026,6,1) and <DATE(2026,7,1), with an explanation that the exclusive upper bound includes every June timestamp. That description is intentionally observable. “Looks good” is not acceptance. The operator must calculate a tiny hand-checked table, test June 1 and June 30, inspect dates stored as text, confirm absolute references, and compare the total with a filtered sum. Keep the source material beside the draft so review means comparison, not memory.

Run the work through describe → draft → test → explain → fill. Stop when an authoritative input is missing, a high-risk claim lacks evidence, private material cannot be safely removed, or the proposed action exceeds the permission granted. Escalation is successful workflow behavior, not model failure.

Common framing mistakes are wrong locale separators; text dates; shifted ranges; hidden filters; formulas filled down with relative criteria cells. Prevent them by writing a one-paragraph job card: user, decision, deliverable, source of truth, constraints, reviewer, and stop condition. This card becomes the anchor for every later prompt.