Check an AI-generated formula against expected answers you have worked out independently. A formula that runs and produces plausible numbers may still encode the wrong rule, mishandle blanks or treat a boundary incorrectly.
Microsoft’s IF documentation describes a logical test with different results for true and false. That structure makes a small, fictional example useful for seeing what a meaningful check looks like.
The 9, 10 and 11 examples below each have a purpose: below the threshold, exactly on it and above it. Together they show whether the inclusive boundary has been represented correctly. A small set of well-chosen inputs can make the formula’s behaviour easy to inspect.
Write the rule before the formula
Suppose an invented task should label a numeric score of at least 10 as eligible and a lower score as not eligible. Before accepting a formula, record the expected labels for 9, 10 and 11.
Then decide what should happen for a blank, text or an error value. These cases can have explicit expected answers before the formula is checked.
Build a small test table
| Input | Expected meaning |
|---|---|
| 9 | Below the threshold |
| 10 | Exactly at the inclusive boundary |
| 11 | Above the threshold |
| Blank | Behaviour must be specified |
| Text | Behaviour must be specified |
Compare actual results with these expectations. Read the documented function behaviour when results differ, including the workbook’s locale and data types where relevant.
Test the surrounding workbook
Check copied references, ranges and any assumptions about sorted data or missing rows. A formula can be correct in one cell and wrong after it is filled elsewhere. Keep the checks proportionate to the consequences of the spreadsheet.
The model’s explanation can help describe the logic; the independent examples provide a separate check of the results. Retain the independent examples and actual results so another reviewer can repeat the checks.
Test an actual candidate formula
For the numeric-score example above, a candidate Excel formula is:
=IF(A2>=10,"Eligible","Not eligible")
With numeric inputs, 9 should return Not eligible, while 10 and 11 should return Eligible. The inclusive comparison is the important part of this example. Localised Excel installations may use different function names or separators; enter the supported equivalent for the workbook.
The formula above leaves the treatment of a missing score outside its specification. If a blank must display Awaiting score, add that requirement before choosing or revising a formula. Then extend the expected-answer table to include it. Text entries and error values also need an agreed treatment for the actual workbook.
Check references after copying
Suppose the formula is copied from row 2 to row 3. Its input should follow the intended row when each row contains a separate score. If the threshold comes from one designated configuration cell instead of the literal number 10, the reference to that cell needs to remain attached to that threshold as intended.
Check the first copied row, a later row and any row adjacent to a gap in the data. Record expected and actual answers side by side. This checks both the decision rule and its placement in the sheet, two separate parts of the same task.
Blanks deserve an agreed meaning too. A blank score might mean awaiting assessment, while zero is an actual numeric result. Giving those cases separate expected answers makes the spreadsheet more useful to the people entering and reading the data.
Sources consulted
- IF function — Microsoft; checked 4 November 2026.