How to Write a Business Case Using ChatGPT for a New Tool or Investment

The Six Sections a Business Case Needs

A business case is a decision document with six jobs: state the problem in money, propose a solution, set cost against benefit, explain the alternatives you rejected, name the risks, and make one specific ask. Miss any of them and a reviewer fills the gap with doubt.

1. Problem. What the current situation costs, in money or hours.
2. Solution. What you want to buy, for whom, at what scope.
3. Cost–benefit. The price of the tool set against the price of doing nothing.
4. Alternatives. The other options you weighed, and why each one lost.
5. Risk. What could realistically go wrong, and how you would handle it.
6. Recommendation. One specific ask, with an amount and a date.

HM Treasury’s guidance on developing business cases, which sets the Five Case Model used across UK government spending approvals, organises the same thinking under different labels — strategic, economic, commercial, financial and management. The vocabulary differs; the questions don’t. Any reviewer, anywhere, is asking why now, what does it cost, what else did you consider, what could go wrong, and what exactly do you want from me.

Why one big prompt fails

The instinct is to paste everything into a single request and ask for the whole document. It never holds up. By the fourth section the model has drifted off your constraints, quietly rounded your figures, and slipped back into the marketing register it defaults to whenever it runs out of specifics.

Sequencing fixes it. Each section gets its own request, and each request re-states the facts — the same principle behind chaining prompts across a multi-step task. One more ordering rule: write the executive summary last. A summary drafted before you know your own numbers is just a hopeful paragraph.

Build Your Evidence Ledger First

The Evidence Ledger is a short list of every hard figure your case depends on, written by you, before ChatGPT is involved at all. Costs, hours, hourly rates, headcount, renewal dates, quoted prices — each one from a source you could name out loud in a meeting. It takes about ten minutes.

The difference isn’t length. It’s that the second version has nothing left for the model to invent. Ask for ROI without inputs and you have effectively requested a plausible number, which is exactly what AI hallucination produces. Supply the inputs and the same request becomes arithmetic.

The point of choosing to write a business case using ChatGPT rather than from a blank document is speed, and speed only pays off if the result survives review. A draft that comes back for sourcing has cost you more time than writing it by hand would have.

Re-pasting the ledger each time feels wasteful. Do it anyway. It costs a few seconds and it prevents the drift that makes long AI documents contradict themselves halfway down. Here is the block that opens every session.

You are helping me draft a business case for [tool or investment] at [company or team name].

EVIDENCE LEDGER — these are the only figures you may use:
- People affected: [number]
- Hours lost per person per week to [the problem]: [number]
- Average loaded hourly cost: [currency amount]
- Quoted price of the proposed tool: [amount] per [user per month, or licence per year]
- Current spend on whatever this replaces: [amount] per year
- Decision needed by: [date]

RULES — apply these to every section you write for me:
1. Use only the figures in the Evidence Ledger. Do not estimate, round, benchmark, or infer any other number.
2. If a section needs a figure that is not in the ledger, write [NEEDS FINANCE INPUT: name what is missing] instead of supplying one.
3. Write for a sceptical CFO. Short active sentences. No buzzwords and no adjectives about the product.
4. Do not write the executive summary until I ask for it.

Confirm you understand these rules, then list any figure you think is missing. Do not draft anything yet.

Two lines there do most of the work: the instruction to use only supplied figures, and the instruction to write a visible gap rather than estimate. That second one is a negative constraint, and it outperforms politely asking for accuracy every time. Ending with “do not draft anything yet” matters too — the missing-figure list you get back is often more useful than the draft.

Sanitise Company Data Before You Prompt

On a personal ChatGPT account, your conversations can be used to train OpenAI’s models unless you turn that off. On a ChatGPT Business or Enterprise workspace, they are excluded by default. OpenAI states this directly in its documentation on how your data is used to improve model performance.

The practical version for someone drafting a budget request: assume the chat is not private until you have checked which account you are in. The setting lives under Settings, then Data Controls, labelled “Improve the model for everyone”. Switching it off applies to new conversations, not old ones. Our fuller guide to whether ChatGPT is safe to use at work covers retention and company policy in more depth.

Settings are only half of it. The bigger risk in a business case is the raw material — the signed contract, the salary band, the client name that has no business being in a chat window at all.

Write a Business Case Using ChatGPT: Six Prompts

Each prompt below assumes the grounding block from the previous section is already in the conversation. Run them in order, in one chat, reviewing each output before moving on. The structure underneath is the familiar Role, Goal, Context, Format model — role set once at the top, goal and format restated per section.

Problem first, always. The strongest business cases spend most of their energy on the cost of the current situation, not the features of the new thing. A reviewer who accepts that the status quo is expensive has already half-approved the purchase.

Using the Evidence Ledger and rules above, draft two sections.

SECTION 1 — PROBLEM (150 to 200 words)
Quantify what the current situation costs per year, using only ledger figures. Show the arithmetic on one line so a reviewer can check it. Then describe the operational symptom in plain terms: [what actually goes wrong, e.g. "briefs get re-sent three times because nobody knows which version is current"].

SECTION 2 — PROPOSED SOLUTION (150 to 200 words)
State what we would buy, for how many people, and what changes in the first week. Scope only. No benefits language and no ROI claims in this section.

Context: the decision-maker is [role], and the thing they care about most is [cost, risk, speed, or headcount].

This is where drafts usually fall apart, because it is the one section that cannot be written around. Give the model the table shape you want and it will fill it correctly; leave the shape open and it will write persuasive prose with numbers sprinkled through it.

Using the Evidence Ledger only, produce the cost–benefit section.

Output a table with exactly these columns: Year | Cost of doing nothing | Cost of the proposed tool | Net difference.
Cover [number] years. Beneath the table, show the calculation used for each column in one short line each.

Then write three bullets:
- the payback point, but only if the ledger figures actually support one
- the single largest driver of the difference
- one sentence naming what these figures do NOT prove

If any cell cannot be calculated from the ledger, write [NEEDS FINANCE INPUT: what is missing] in that cell. Do not fill it with an estimate, an industry average, or a typical range.

Reviewers reject proposals that pretend there was only ever one option. Treat “do nothing” as a genuine alternative with a genuine price tag, and the risk section as a credibility test rather than a formality — if you want a standalone version of that section, our guide to producing a risk assessment with ChatGPT goes deeper.

Draft two more sections, same rules.

SECTION 4 — ALTERNATIVES
List the options considered: [option A], [option B], and doing nothing. For each, give one line on what it would cost us and one line on why we rejected it. Treat doing nothing as a real option, priced from the ledger.

SECTION 5 — RISKS
Identify the four most likely risks of going ahead: adoption, integration, cost overrun, and [any risk specific to our situation]. For each, give the risk in one sentence, the likelihood in plain words rather than a percentage, and the mitigation we would genuinely run.

Reject any risk that would apply equally to every software purchase ever made. I want the ones specific to this one.

Leadership reads this paragraph and often nothing else, which is exactly why it should be written after the evidence rather than before it. Our full guide to writing an executive summary with ChatGPT covers the general form; this version is tuned for a spending decision.

Now write the executive summary, using only what appears in the sections you have already drafted. Introduce nothing new.

Rules:
- Maximum 150 words.
- First sentence states the problem in money.
- Second sentence states the ask: the amount, the timeframe, and who needs to approve it.
- Then three short lines: what it costs, what it saves, what happens if we do nothing.
- No adjectives about the product. Delete any sentence that would survive being pasted into a different company's proposal unchanged.

Finish by listing every [NEEDS FINANCE INPUT] marker still in the document so I can chase them.

Red-Team the Draft Before Your CFO Does

The step nearly everyone skips is the cheapest one available. Before the draft goes anywhere near a real reviewer, make ChatGPT play the reviewer — the same rehearsal logic behind preparing for a difficult conversation with AI, pointed at a document instead of a person.

Switch role. You are now [the CFO or finance director] reviewing this proposal, and your default answer to spending requests is no.

Read the full draft above and give me:
1. The five strongest reasons to reject it, ranked by how likely each is to actually come up.
2. For each one, quote the specific sentence or figure in my draft that triggers it.
3. The one question you would ask that my draft currently cannot answer.

Be blunt. Do not soften the objections and do not write my responses for me yet.

Ask for the objections first and the counters second, in a separate turn. Requested together, the model writes weak objections it can comfortably answer. Requested alone, it finds the ones that sting.

Then run one last pass over the whole document for tone. The default AI register is confident, abstract and slightly promotional, which reads as evasive to anyone who signs cheques for a living.

Worth saving the six prompts somewhere you will find them again — a notes file, a ChatGPT Project, or alongside the other free AI work templates we keep on the site. The second business case takes fifteen minutes, not thirty.

Where It Breaks and What to Check

Five failure modes account for nearly every bad output when you write a business case using ChatGPT, and they share one root cause: something the model needed was left for it to supply. Recognising the symptom tells you which part of the chain to rerun, rather than starting the whole document again.

What ChatGPT cannot do here

None of this makes the case for you. What it does is remove the two things that get proposals dismissed before their argument is even read: numbers nobody can trace, and language that sounds like it came from a brochure. Build the ledger once and the six prompts turn a job that used to eat a weekend into something you finish between two meetings.

That is the honest promise of learning to write a business case using ChatGPT — not a better argument than you could make yourself, but the same argument, properly structured, in a fraction of the time. The thinking stays yours; the typing doesn’t. And the next one is faster still: keep the ledger format, keep the six prompts, change the figures.