Financial Modelling Best Practices: A Reviewer's Checklist

Review enough financial models and a pattern emerges: the failures are rarely mathematical. The discounting is fine, the growth formulas work. What breaks is architecture — structure, separation, visibility. These are the practices that separate robust models from fragile ones.

1. Separate inputs from calculations

Every hardcoded number is an assumption, and every assumption should be visible in one place, formatted distinctly. A formula that contains a hidden hardcode — =B14*1.07 — is an assumption disguised as logic. It will be missed in review and stale within a year.

2. One timeline, everywhere

Columns are periods, and period N means the same date on every sheet. The most insidious model errors are timing mismatches: revenue recognised annually in one schedule and quarterly in another, interest on opening balances in one place and closing in another.

3. One concept per row

Long formulas that compute three things at once are unauditable. Break calculations into steps a stranger can trace. The cost of an extra row is trivial; the cost of an untraceable formula is a review that fails.

4. Checks that run continuously

Build the balance check and the cash check into the model itself, visible on every sheet, linked to a master flag that turns the moment anything breaks. A check that exists but is never looked at is decoration.

5. Kill circularity deliberately

Interest depends on debt; debt depends on cash flow; cash flow depends on interest. Handle the circle on purpose — iterative calculation with a breaker switch, or a deliberate approximation — never by accident. Undetected circular references corrupt results silently.

6. Document what matters

For every material assumption: what is the number, where did it come from, and who owns it? Documentation is not bureaucracy; it is the difference between a model and an argument.

7. Present ranges, not points

Every material output deserves a sensitivity. A valuation of 1,000 that swings from 700 to 1,400 across plausible assumptions is a different story from one that holds between 950 and 1,050 — and the decision-maker deserves to know which story they are being told.

8. Independent review

The person who built the model is the least qualified to find its errors; authorship blinds. A structured second review — architecture first, then checks, then formulas, then sanity benchmarks — catches what builders cannot see. The full architecture behind these practices is in the Financial Modelling Hub.