Language: English
Start 14-day free trial

Guide

How to use ChatGPT for financial analysis: review the assumptions

Updated

ChatGPT is most useful in financial analysis as a reviewer of your model's assumptions, not as the author of your numbers. Paste the assumptions and the key outputs, and ask for five things. Every assumption with its value, unit, source and who set it, with "guess" where the source is blank. Any result typed in as a number instead of calculated. Each driver moved up and down one at a time, ranked by how much it moves the output. Each top driver compared with your own history. And the three assumptions the forecast rests on most, with what would happen if each is wrong. Then check every figure it quotes against the spreadsheet.

Short answer

  • Review, don't build: give it the assumptions and outputs.
  • List every assumption with its source; blank means "guess".
  • One input at a time, ranked by effect.
  • The three that matter most, with the downside of each.

The brief to copy

Assumptions review brief

"Review the assumptions in this forecast. Key output: [e.g. cash at month 12].

List every assumption, including numbers typed into formulas: what it is, value, unit, period, source (history, contract, plan, guess), and who set it. Blank source means 'guess'.

Separate drivers (inputs) from results (calculated). Flag any result typed in as a number.

Move each driver 20% up and down, one at a time, and show how much the key output changes. Rank them.

Compare the top drivers with our history: [paste]. If there's none, write [check: no history].

Name the three that matter most, with the evidence, the downside if wrong, and what would raise confidence. Note unit or timing problems. Summary in two or three sentences.

Assumptions and outputs: [paste]"

A worked example

Illustration: a made-up business, round numbers. Lina, finance lead at a 20-person software company, pastes next year's forecast: cash at month 12 is the key output. From the review:

From the review

Typed-in result: revenue for months 7 to 12 is hard-coded, not calculated from new customers × price. [fix]

Sensitivity (±20%, cash at month 12): monthly new customers ±210,000 · churn ±90,000 · price ±60,000 · salaries ±40,000.

Plausibility: the plan assumes 30 new customers a month; the best month so far was 18. [needs a reason]

The three that matter most: new customers a month (30, no history behind it; if it's 18, 40% lower, cash at month 12 falls by about 420,000), churn (2% a month, from the last six months), price (from the current price list).

Structural: two sales hires start in month 1 with no recruiting time, and their quota adds revenue from month 2.

Summary: The forecast ends the year with cash to spare only if new customers reach 30 a month, nearly double the best month so far. That and the hard-coded second-half revenue are the main risks.

Lina checks each figure in the spreadsheet, replaces the hard-coded revenue with the formula, and reruns the plan at 18 new customers a month before it goes to the board.

Before you use it

  1. Check every number it quotes against the spreadsheet.
  2. Give it your history for the top drivers; without it, plausibility is a guess.
  3. Use a business account that doesn't train on your data by default.
  4. Leave out client names and anything your contracts restrict; see client data and AI in finance.
  5. Keep the decision with people; the review shows where to look.

Next step: keep the brief

Every forecast and budget gets the same review. Finance prompts, saved shows where to keep it, and Claude skills for finance covers the skills to set up first.

Skills for this

All skills

Questions

Can ChatGPT do financial analysis?

It's most reliable as a reviewer of your model, not its author. Paste the assumptions and the key outputs and ask it to list every assumption with its source, flag results typed in as numbers, test each driver up and down one at a time, compare the drivers with your own history, and name the three that matter most. Check every number it gives you against the spreadsheet.

What is sensitivity analysis in a forecast?

Changing one input at a time, by a set step such as 20% up and down, and recording how much the key output moves. Ranking the inputs by that change shows which assumptions the forecast really depends on.

Is it safe to paste a financial model into ChatGPT?

Use a business account that doesn't train on your data by default, and leave out client names and anything your contracts restrict. For many reviews you only need the assumptions and totals, not the whole workbook.

AI skill

Have your AI challenge the assumptions in your financial model

Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. Financial model assumptions review has your AI review each forecast: each assumption with its value, source and owner, drivers kept apart from results, sensitivity one input at a time, plausibility against your history, and the three assumptions that matter most.

Use this skill in your AI

14 days free · No card needed

Sources

  1. HM Treasury: The AQuA Book, guidance on producing quality analysis for government (30 July 2025) — an assumptions log: each assumption described, its effect and reliability quantified, when and why it was made, who made it and who signed it off; record any sensitivity analysis (read 2026-11-05)
  2. Corporate Finance Institute: What is sensitivity analysis? (read 2026-11-04)
  3. Anthropic Privacy Center: commercial products — not used for training by default (read 2026-11-05)
  4. OpenAI Help: Data controls — personal plans train on chats by default unless turned off; read through search-result text (read 2026-11-05)