Financial projections that survive scrutiny
Nobody believes your projections, and they are not supposed to. What a reader is testing is whether the projections are constructed in a way that can be interrogated — because a model you can argue with is evidence of a business you have thought about.
3 min readUpdated
Build revenue from drivers, not from a growth rate
A growth rate is unfalsifiable. Twenty per cent a year cannot be argued with, because there is nothing underneath it to argue about. A driver build can be argued with in both directions, which is exactly why it is worth more.
| Asserted | Built |
|---|---|
| $1.2m in year one, growing 20% a year | 210 covers a day × 62% seated × $38 average × 26 days |
| Nothing to challenge | Four numbers, each of which a reader can test against their own experience |
| Every downstream figure inherits the guess | Change the ticket and margin, tax, cash and coverage all move |
The build differs by business model — seats and churn for a subscription, billable hours and utilisation for a service firm, traffic and conversion for a shop — but the principle does not. Revenue should be a small number of things that multiply together, each of which you could defend for thirty seconds.
Three statements, and why the third one matters
A profit and loss on its own tells you whether the business makes money on paper. The cash flow tells you whether it survives. The balance sheet is the one most plans omit, and it is the one that proves the other two are internally consistent: if assets do not equal liabilities plus equity in every period, something in the model is wrong and you do not yet know what.
Three things that only show up when the statements are linked:
- The gap between profitable and funded. Carrying receivables means the cash date is later than the profit date, sometimes by months.
- The working-capital cost of growth. Growing fast while paying suppliers in thirty days and collecting in sixty consumes cash exactly when it looks like things are going well.
- The real effect of a loan. Interest is on the profit and loss, principal is not — the repayment shows up in cash and nowhere else.
Monthly for year one, without exception
Annual figures hide the month you run out of money. A business that ends year one with cash in the bank can still have been six weeks from failure in month four, and the annual view shows none of it. Any reader who has seen a few plans will ask for the monthly detail; providing it up front is a small effort that removes an entire round of questions.
The ratios a reader computes anyway
Founders submit plans; banks compute ratios. If the ratio is not in the document the reader works it out, and works it out with assumptions you did not get to choose. Putting it in the plan is how you keep control of the framing.
| Ratio | What it tests |
|---|---|
| Debt service coverage | Whether cash available covers scheduled debt service, with a margin the programme sets |
| Current ratio | Whether short-term assets cover short-term obligations |
| Debt to equity | How much of the risk you are carrying yourself |
| Owner compensation | Whether the business supports the person running it |
| Break-even, on profit and on cash | Two different dates, and both get asked about |
Where projections quietly go wrong
- A cost line that never changes for five years. Rent escalates, insurance rises, maintenance climbs after the warranty ends.
- Headcount that grows with revenue but never with a start month, so the payroll cost lands a year before the hire would.
- A margin outside the band for the industry with no explanation. It may well be defensible — but unexplained, it reads as an error.
- Tax as a flat percentage of a positive number, ignoring the losses carried forward from the first two years.
- Rounding inside the model rather than at the edges, which is how a schedule ends up repaying a different amount than was borrowed.
Every rule here is enforced by the product.
The checks this article describes are the checks that run before a plan can be exported. Free to generate and read.