ChatGPT for CFOs: What It Is Good For and What Never Goes In
ChatGPT is good for a CFO's words and checks: first drafts of board commentary and investor updates, explaining and fixing spreadsheet formulas, summarising contracts and long documents, and arguing against a decision before you take it. Five things never go in: personal data like payroll, price sensitive numbers before they're public, bank and customer details, passwords and keys, and any figure that leaves the building before a person has tied it back to the ledger.
The two ways finance teams get ChatGPT wrong
The first is the ban. A CFO sends a memo saying nobody uses it, and within a month half the team is pasting supplier lists into a free personal account on their phone. A ban doesn't stop the use. It stops you seeing it.
The second is using it as a calculator. Someone exports 4,000 ledger lines, pastes them in and asks for the gross margin by product. The answer comes back in 10 seconds, it looks right, and nobody can say how it was worked out. If that number reaches a board pack and it's wrong, you've lost the board's trust in every AI number for a year. A chat window is a drafting tool. Your ledger is the calculator.
The one page list I give every finance team
Before anyone types a prompt, the team gets one page with three columns. It takes about an hour to agree with the controller and the head of FP&A, and it replaces the memo.
| Use freely | Use with a check | Never goes in |
|---|---|---|
| Rewriting your own board commentary | Summarising a customer contract | Payroll, salaries, anything naming a person |
| Explaining a formula or a macro | Aggregated management accounts | Results, deals or guidance before they're public |
| Drafting a policy or a process note | Variance explanations from your own notes | Bank account numbers, card and customer details |
| Objections to a decision memo | First drafts of investor updates | Passwords, API keys, system logins |
| Plain English versions of tax or IFRS notes | Benchmarks and market research | Any number going out unchecked against the ledger |
1. A business account before the first prompt
Personal accounts are out from day one. Buy a business or enterprise plan, read the clause on training in its terms yourself, and make sure it says your inputs aren't used. You also want one admin who can see who has a seat. For a 12 person finance team that's 12 seats on one invoice. Buy them before the ban memo would have gone out.
2. The red column is a hard line
If you're listed, price sensitive results and deal terms don't go in until they're announced, whatever the plan says. I've been through 3 FT100 listings and the rule I took from them still holds: if you wouldn't email it to an outside adviser without an NDA, it doesn't go in a chat window. Payroll is the one people slip on, because it's just a spreadsheet. It isn't. It's personal data, and UK GDPR cares where it went.
3. Every number is tied back before it leaves
ChatGPT can write a sentence about a number. It can't be the source of one. The rule I use is plain: any figure that reaches a board pack, a bank, an investor or HMRC is ticked against the ledger or the bank by a named person. A draft paragraph with 6 numbers in it gets 6 ticks. That takes 5 minutes and it's the cheapest control in the department.
4. Five tasks, 30 days, then decide
Don't roll it out to everything. Pick 5 tasks from the left column, one per person, and run them for 30 days. My usual 5 are month end commentary, the investor update draft, formula checks on the forecast model, contract summaries for the controller, and a devil's advocate pass on any spend over your approval limit. At day 30, ask each person how many hours it saved and keep the ones that saved more than 2 a week.
From my operating seat
I'm Hayat Amin, an AI-native CFO, AI operator and AI coach for CFOs who teaches finance leaders how to build AI into finance operations. I spent twenty years in the C-suite, with three exits and three FT100 listings. Now I build and run AI agent systems inside companies, across finance, sales pipeline, onboarding and compliance evidence, and I coach CFOs one to one on the same work.
I use a chat window every day, mostly for the writing. In a sale process or a board cycle, the slow part is turning numbers you already trust into words a buyer or a board will read. That's where chat earns its seat. The numbers themselves come from the ledger, and inside the companies I run they come from agents that read it overnight, like the one that posts a cash runway figure before the finance team logs in. Chat drafts the paragraph about the runway. It never calculates it.
The CFOs I coach usually move through it in that order: 30 days of chat for drafting, then the first agent that reads the ledger on a schedule. Skipping the first month is how you end up with an agent nobody on the team trusts.
What are the best ChatGPT prompts for a CFO?
The prompts that work give it a role, the material and the shape of the answer. Paste last month's commentary and this month's variances, then ask for the same structure in under 300 words. Paste a formula and ask what it does, line by line. Paste your decision memo and ask for the three strongest objections a non-executive director would raise. A one line prompt with no material gets a generic answer.
Is it safe to put financial data into ChatGPT?
On a personal account, treat it as leaving the company. Use a business or enterprise plan whose terms say in writing that your inputs aren't used for training, and read that clause yourself. That's the only kind of account I let a finance team use. Even then, aggregated management numbers are fine and the five red items are not: personal data, price sensitive results, bank and customer details, credentials and unchecked figures.
Should a CFO use ChatGPT or Claude?
For drafting and checking, either works, and the plan and the rules matter more than the brand. I build my finance agents on Claude because they run on a schedule against the ledger with a rules file loaded every time. For a CFO starting out, pick the one your company already pays for on a business plan and use it every day for a month.
Where I come in
This is the work I do with CFOs one to one and inside finance teams. We write your one page list in the first session, pick the 5 tasks, and at day 30 we decide which one becomes your first agent. If you're ready for that step, read the first five finance agents, and if you want the learning order first, start with AI training for CFOs. The full map is at AI for CFOs, what the end state looks like is at the AI-native CFO, and the sessions are at coaching for CFOs. What would go in your red column that isn't on mine? Start at meethayat.com and tell me.