Three drivers, eighteen months, one page. The value is in the assumptions you can defend, not the projections you cannot.
14 min read · Level 3
A model's job is not to be right. It is to make your assumptions explicit enough that being wrong teaches you something. A founder who projects £2M of ARR and reaches £600k has learned nothing if the model was one number in a deck; the same founder learns a great deal if the model said "40 new customers a month at 4% churn" and reality said 14 at 7%. Build the model so that reality can contradict it specifically. Vague projections cannot be wrong, which is exactly why they are worthless.
A financial model projects revenue, costs and cash forward from a small set of stated drivers.
For an early company: new customers per month, price, churn, gross margin, and a hiring plan. Eighteen months. One page. Everything else is derived.
Because it converts a plan into consequences. "Hire two engineers and push marketing" becomes a runway number, a break-even date and a required growth rate — which is when you find out whether the plan is possible.
And because it is the artefact that makes a fundraising conversation concrete. Not the projections — investors discount those heavily and always have. The assumptions: an investor who disagrees with your churn number is having a real conversation with you, and that conversation is the point.
The minimum viable model — five drivers, one page.
DRIVERS Now Assumption New customers / month 12 +2 per quarter Price / customer / month £200 flat Monthly churn 4% →3% after onboarding work Gross margin 78% flat Hires — GTM in M3, eng in M7 DERIVED (month 12) Customers 184 → 341 MRR £36,800 → £68,200 Gross profit £28,704 → £53,196 Fixed costs £45,700 → £59,700 Monthly profit (£16,996) → (£6,504) Cash £214,000 → £71,000 Runway ~11 months
What this model is actually telling you: with the current plan you are still losing money in month 12 and down to eleven months of cash. So either the plan changes, or you raise around month 8.
That is a decision, produced by five numbers.
Then the assumption test. Which driver is least defensible? "New customers 12 → 20/month" is the whole story — the model works because of it and there is no evidence for it yet. Churn 4% → 3% is a second bet on unbuilt work.
So the model's real output is: go and find out whether 20 new customers a month is achievable. Not a spreadsheet — a research task.
⚠ A model with more than about eight drivers stops being understandable and starts being a place to hide. If you cannot explain every input in a sentence, delete it.
First-time founders build models that reverse-engineer a desired outcome. Pick £2M ARR, work backwards to the growth rate that produces it, present the growth rate as a plan. Experienced investors spot this immediately, and worse, the founder starts believing it.
The second mistake: fifty tabs. Complexity is not rigour. A model nobody can hold in their head does not get challenged, and unchallenged assumptions are the only dangerous kind.
The third: building it once for a fundraise and never updating it. A model compared monthly against actuals is a learning instrument. One built for a deck is a decoration with formulas.
New customers, price, churn, gross margin, hiring. Everything else derives. If you cannot explain an input in one sentence, remove it.
Customers × price, month by month. Never a percentage of a market — that is a wish with arithmetic attached.
Beside every driver: what it is now, what you assume, and why. Blank evidence is the model's most useful output.
Long enough to include a raise; short enough that the numbers are not fiction. Five-year models persuade nobody who matters.
Change each driver by 20% and see which moves the outcome most. That one is your real risk, and usually your next research task.
Model versus reality, per driver. This is what turns it from a document into a thing you learn from.
Before a raise, before a hiring plan, and at each planning cycle. Update monthly once it exists.
Pre-product-market fit, a detailed model is false precision — you do not know your churn or your acquisition rate, so every output is a guess compounded twelve times. Model the cost side, which is real, and hold the revenue side loosely.
It is widely acknowledged among early-stage investors that seed-stage financial projections are not treated as forecasts. What gets examined is the assumption set: the implied conversion rates, the sales-hire productivity, the churn. A founder whose model implies each new salesperson closes far more than the industry norm is signalling something about their judgement, not their ambition.
Assume every number will be tested for plausibility, not accuracy. Defensible and modest beats impressive and unexplainable.
A frequent pattern among founders who navigate their first raise well: alongside the spreadsheet they can state the business in one sentence — something like "we add roughly twelve customers a month, they pay £200, about 4% leave, so we need X to break even." The spreadsheet supports the sentence rather than replacing it.
If you cannot say your model out loud in one sentence, it is too complicated to be useful to you and too complicated to be believed by anyone else.
Create a free account to read the rest and apply it to your own startup.