How to write a business plan
Most guides give you a table of contents. A table of contents is not the hard part — the hard part is that a business plan is a document with an audience of one, and almost nobody writes it that way.
5 min readUpdated
Decide who is reading it before you write a word
A credit committee, a seed investor and a visa adjudicator want three different documents. They share a structure and share almost nothing else. The credit committee is testing whether the debt can be serviced in the worst year you are willing to admit to. The investor is testing whether this can be twenty times bigger. The adjudicator is testing whether the document is responsive to a specific set of regulatory requirements and whether it reads as credible.
Write for one of them. A plan that tries to satisfy all three reads as generic to each, and generic is the specific failure mode that gets a plan skimmed rather than read.
| Reader | The first thing they check |
|---|---|
| Bank or SBA lender | Whether cash flow covers debt service with room to spare, and whether the owner is paid |
| Seed investor | Whether the market claim is derived or asserted, and whether the unit economics survive one question |
| Immigration adjudicator | Whether the document is responsive to the regulation and whether it reads like a template |
| You, internally | Whether the plan changes any decision you were about to make anyway |
Write it in the wrong order on purpose
The executive summary goes first in the document and last in the writing. It is the only section most readers finish, and it can only be written once you know what the model says — which you will not know until the model exists.
The order that works:
- 1. The operating model
- How you actually make money, as drivers rather than a growth rate. Covers per day, conversion, average ticket. Seats, price, churn. Billable hours, utilisation, rate. If you cannot describe revenue as a small number of things that multiply together, you do not yet have a plan — you have an ambition.
- 2. The numbers
- Three statements, monthly for at least the first year, that move together. Not a revenue line with a percentage growth rate: a model where changing the price changes the margin, the tax, the cash and the coverage ratio in the same pass.
- 3. The market and the competition
- Built from the bottom up, sourced, and cross-checked against what your own model says you will sell. Do this after the model and the two will be reconcilable. Do it before and you will be tempted to work backwards from a number you liked.
- 4. Operations, team, risks, regulation
- The sections that turn a forecast into a business. The staffing plan has to be the same headcount the payroll carries. The risk section has to name risks that would actually change the outcome, not a list of generic ones.
- 5. The executive summary
- Now, and only now. One page. The business, the ask, the use of funds, the three numbers that matter, and what happens if you are wrong.
The six places plans actually fall over
- Owner compensation shown as zero
- The most common silent failure in the category. A lender substitutes a market salary and recomputes; the plan that looked serviceable no longer is. For an immigration filing it is worse, because the whole marginality test turns on whether the business supports the applicant. Put a real number in, even if you do not intend to draw it in year one, and say so.
- Prose that contradicts the spreadsheet
- The text says forty per cent growth; the model says twelve. The text says eight employees; the payroll carries five. Nobody does this deliberately — it happens because the prose was written once and the model was revised four times. It is also trivially findable, which is why it costs so much credibility when a reader finds it.
- A market section with no arithmetic
- Taking a published industry total and claiming one per cent of it is the single most recognisable pattern in an unfundable plan. Count the buyers you can actually reach, multiply by what they spend, and show the working.
- A downside that only cuts revenue
- Holding the hiring plan and the marketing budget constant while revenue falls thirty per cent is arithmetic, not a scenario. A real downside moves demand, pricing, timing and spend together, and leaves genuinely committed costs like the lease fixed.
- Annual-only figures in year one
- A yearly total hides the month you run out of cash. A reader who has seen a few of these will ask for the monthly detail, and asking is the point at which they stop believing the annual number.
- No statistic with a date on it
- An uncited market claim is the most commonly cited reason plans are rejected, and AI-written plans are now notorious for citing reports that do not exist. Every number you did not compute yourself needs a source, a publisher and the date you read it.
How long it should be
Long enough to answer the questions your reader will ask and no longer. In practice: fifteen to twenty-five pages of narrative for a lender or an investor, plus the financial statements as an appendix. An immigration filing runs longer because the regulation asks for more.
A useful test: delete any paragraph that contains no number, no name, no date and no place. If the section survives that with nothing left, it was never saying anything.
Before you send it
Read it as the person receiving it. Then check, specifically:
- Every figure in the prose appears somewhere in the statements, with the same value
- The headcount in the narrative equals the headcount on the payroll equals the org chart
- Owner compensation is a visible, non-zero line in every year
- Every statistic you did not compute carries a source and a date
- Year one is monthly, and no month has cash going negative without a financing line to cover it
- The downside case moves at least five drivers, and you can say why each one moves
- The break-even date on paper and the break-even date in cash are both stated, and they are different
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.