How-To Guide

How to Use AI to Write a Business Plan (Without the Made-Up Numbers)

A seven-prompt sequence that turns your own figures into a plan a lender would actually read — plus the four things AI will invent if you let it

B Biztrategy Published 29 October 2026 · 7 min read
Top view of colleagues discussing charts and graphs on a wooden table during a business meeting.

Every so often something forces a small business owner to write a business plan. A bank wants one before it will look at a loan. A landlord wants one before handing over a lease. A new partner wants to see where the thing is actually going. Or — most often — you just want one quiet afternoon to work out what the next twelve months are supposed to look like, instead of reacting to whatever lands in the inbox.

AI is genuinely good at this job, but not in the way the hype suggests. Ask ChatGPT or Claude for “a business plan for my cleaning company” and you will get eleven pages of confident, plausible, entirely generic text, complete with a market-size figure it invented and a competitor section that names nobody. That is exactly the complaint owners keep making about AI advice: every guide says use AI, almost none of them say how. This is the how — a sequence you can run in about three hours that produces a plan a bank manager would actually read.

What AI is actually good at in a business plan

Before you start, it helps to be honest about the division of labour. A business plan has two kinds of content: things only you know, and things that are mostly structure and prose. AI is excellent at the second category and dangerous in the first.

Where it earns its keep:

  • Structure and completeness. It knows what a lender expects to see and will tell you which of the ten standard sections you have skipped.
  • Turning notes into prose. You dictate six messy bullet points about how you win customers; it returns three tidy paragraphs you then correct. This is where most of the three hours gets saved.
  • Interrogating your assumptions. Asked properly, it is a tireless, slightly annoying sceptic — which is more than most of us get for free.
  • Producing variants. One plan, three audiences: a six-page version for the bank, a two-page version for a prospective partner, a one-page internal version for your team.

Where it will quietly let you down: your actual numbers, your actual local market, your named competitors, and any judgement call about risk. Treat every figure it produces as a placeholder until you have replaced it yourself.

Gather your inputs before you open a chat window

The single biggest determinant of output quality is what you paste in at the start. Thirty minutes of gathering beats two hours of prompting. Before you begin, pull together:

  • The last 12–24 months of revenue, by month if you have it, and your gross margin.
  • Your current fixed monthly costs — rent, salaries, software, insurance, loan repayments.
  • Customer counts: how many you have, average spend, roughly how many you lose a year.
  • Your three nearest real competitors, by name, and what they charge.
  • Any existing material — an old plan, your website copy, a pitch you once gave.
  • The specific thing this plan is for: “a €60,000 equipment loan from Banco Sabadell” is a much better brief than “growth”.

If your figures are spread across a bank export and three spreadsheets, get them into one place first. A plan built on numbers you are not sure of is worse than no plan.

One practical note on privacy: if the plan contains salaries, customer names or anything you would not email to a stranger, use a paid team-tier account where your inputs are not used for training, not a free consumer login.

The seven-prompt sequence that produces a real draft

Run these in a single conversation or project, in order, so the model keeps your context. Edit as you go rather than at the end — correcting section two before section three is written saves you re-writing section six.

1. Set the brief.

I run a [business type] in [city], founded [year], with [n] employees and roughly €[x] in annual revenue at a [y]% gross margin. I need a business plan to [specific purpose and reader]. Here are my figures and notes: [paste everything]. Before writing anything, ask me the eight questions you most need answered to write a credible plan.

Making it ask questions first is the step most people skip, and it is the one that stops the generic eleven pages. Answer the questions in plain language — typos and all.

2. Outline only. Ask for a section-by-section outline appropriate to your reader, with a word budget per section and a note of which sections need data you have not yet supplied. Approve or reshuffle the outline before any prose is written.

3. The business and the offer. Have it draft the company description, what you sell, and who buys it — using only facts you provided. Add: “Mark anything you inferred rather than took from my notes in square brackets.” Those brackets are your to-do list.

4. Market and competitors. This section needs the tightest leash. Ask it to describe your market using only the three competitors you named and your own customer data, and to state explicitly where an external figure would normally go, rather than supplying one. Then go and find those figures yourself — using AI for market research covers how to do that without ending up with invented statistics.

5. Operations and team. Mostly dictation. Describe how a job actually runs from enquiry to invoice; let it turn that into prose and flag the dependencies you have not mentioned.

6. Financial narrative. Note the word narrative. You supply the numbers; it writes the paragraphs that explain them and points out where the story and the spreadsheet disagree.

7. Adversarial review. The highest-value prompt in the whole sequence:

You are a commercial lending officer at a Spanish savings bank with 20 years of experience. Read this plan and list the ten questions you would ask that it fails to answer, plus the three claims you would not believe without evidence. Be blunt.

Expect this to hurt slightly. It is cheaper than finding out in the meeting.

Where AI will invent things — and how to catch it

Four failure modes account for almost every embarrassing business plan written with AI. Check each one before the document leaves your laptop.

  • Market statistics. “The European pet grooming market is worth €4.2 billion and growing at 11% annually” reads beautifully and may be entirely fictional. Every external number needs a source you have personally opened. If you cannot find one, cut the sentence.
  • Growth curves. Left alone, models write optimism: 30% year-on-year, three new hires, a second location. If your last three years averaged 6%, a 30% projection is not ambition, it is a credibility problem.
  • Competitors. It will cheerfully describe firms that do not exist, or get real ones wrong. Only name businesses you have checked this week.
  • Regulatory claims. Anything about licensing, employment law or tax treatment must be verified against a primary source or your accountant. Do not let a sentence about compliance into a document a bank will keep on file unless you know it is true.

The underlying discipline here is the same one that applies to any client-facing AI output: assume fluency is not accuracy. Our guide on preventing AI hallucinations in client work sets out the verification habits in more detail.

Making the financials yours, not the model’s

Do not ask AI to produce your financial projections from scratch. Ask it to help you build them from your own history, then challenge them.

A workable approach: export your monthly revenue for the last 24 months, paste it in, and ask for three scenarios — flat, your stated plan, and a downside where revenue falls 15% — each built by applying a stated growth rate to your baseline, with the arithmetic shown. Then ask the awkward question: “At what monthly revenue do I stop being able to make payroll and the loan repayment in the same month?”

That single figure — your break-even month — is worth more to a lender than ten pages of strategy, and it is the number most plans leave out. If cash timing is the part that worries you, using AI for cash flow forecasting goes deeper on building a rolling 13-week view.

Keep the actual maths in a spreadsheet you control. The plan quotes the spreadsheet; the chat window never becomes the source of truth.

Tailoring the plan to whoever will read it

Once you have one verified master plan, producing versions is nearly free — and this is where AI genuinely beats hiring someone to write it for you, because revisions cost minutes rather than another invoice.

  • A lender wants repayment capacity, security, and what happens if you are wrong. Lead with the financials and the downside scenario.
  • An investor or partner wants the size of the opportunity and why you specifically will capture it. Lead with the market and the unfair advantage.
  • Your own team wants to know what changes on Monday. Strip the market section, keep the objectives, add owners and dates.
  • A grant application wants impact against the funder’s stated criteria, usually in their own words.

Ask for each version explicitly: “Rewrite this as a two-page summary for a potential business partner. Keep every figure identical. Reorder so the opportunity comes first. Cut the operations detail.” Then re-read it — the risk when rewriting is that a verified number gets rounded, re-phrased or quietly dropped.

The bottom line

Used properly, AI changes a business plan from a two-week job you keep postponing into an afternoon’s work: roughly 30 minutes gathering inputs, 90 minutes running the prompt sequence, and an hour verifying every external figure and replacing the invented ones. What it does not do is make the decisions. The growth rate you commit to, the risk you are willing to carry, the thing you will stop doing to make room for something new — those are still yours, and a plan that outsources them reads exactly like what it is.

Write the first draft fast, then spend your real effort on the parts a model cannot know. That is the whole trick, and it applies to most AI work in a small business, not just this document.

Where does your business stand on AI?

Take the free 3-minute AI Readiness Quiz and get a personalised score with your next steps.

Take the Free Quiz →