AI Financial Modelling: Where It Helps and Where It Lies

AI helps in financial modelling when it builds the structure, writes and audits the formulas and drafts the assumptions book, while every input comes from a source you can name and every sum is done by the spreadsheet. It lies when you let it invent an assumption or do arithmetic in a chat window, so I give it a copy of the model, a one page spec and 4 checks it must pass before anyone opens the file.

Why most AI models fail the first review

The common first try is a prompt like "build me a 5 year model for a 120 person SaaS company". Ten minutes later there's a tidy workbook with 30% growth, 8% churn and a 72% gross margin. Nobody asked where those numbers came from. The model picked figures that look like a SaaS company, and they look right because they're average. Your company isn't average, and a lender or a buyer will ask for the source of each one in the first hour.

The second failure is quieter. Someone pastes 36 months of revenue into a chat and asks for the total or the growth rate. The answer comes back to the decimal place, and it's wrong often enough that I never use one. A language model predicts the next word. It doesn't add. When a wrong total goes into a board deck, the CFO loses more trust in one meeting than the AI saved in a quarter.

The five rules I use for AI in a financial model

1. Write a one page spec before the AI touches anything

The spec says what the model is for, who reads it and what decision it supports. Then it lists the tabs, the time axis (monthly for 24 months, then annual to year 5 is my default), the revenue drivers by name, the cost lines and the 3 outputs that matter, usually cash low point, EBITDA and covenant headroom. It takes me 45 minutes with the CFO. Without it, the AI designs the model it has seen most often, and you spend 2 days pulling it back to your business.

2. Let the AI build structure and formulas, never values

This is where it earns its place. From the spec it lays out the inputs tab, the calculation tabs and the 3 statements, and writes every formula, in about an hour against 2 to 3 days by hand. The rule is strict: it writes formulas that point at input cells, and it types no number into a calculation cell. Every hardcode it adds goes on the inputs tab with a label, and it lists every cell it changed at the end of each session.

3. Every assumption carries a source and an owner

The assumptions book is a table: the input, its value, where it came from and the person who signs it off. The AI drafts this table well. It reads the last 24 months of the ledger, the CRM and the payroll file, and proposes each value with the query behind it. What it may not do is fill a gap with a plausible guess. If there's no source, the cell says "owner to set" and turns red. In a typical build that's 6 to 10 inputs, and they're the ones the CFO should be thinking about anyway.

4. Run 4 checks before a person opens the file

The balance sheet balances in every period. Opening cash ties to the bank to the penny. The last 12 actual months rebuild to the management accounts within 1%. No formula on a calculation tab contains a typed number. The AI runs these and reports pass or fail with the cell reference. If any fail, it fixes and reruns, and it never adds a plug line to make the balance sheet close. A plug is how a model lies in a way nobody spots until due diligence.

5. Use it hardest on review and scenarios

The best use I've found is the model you inherited. Give the AI a 40 tab workbook and it will map every input to every output, find ranges that stop one row short and explain the logic in plain English in an afternoon. Scenarios work the same way. It runs downside, base and upside on a copy, changes only the inputs you name, and writes one paragraph on what moved the cash low point. The live model never changes without a person.

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.

In a sale process, a buyer's team asks about the model's assumptions far more than its formulas. Every growth rate has to trace to a contract, a pipeline report or a cohort. That's why I run the assumptions book as a source table. In the setup I build, the AI rebuilds the 24 month model each month from the closed ledger, flags any input that drifted more than 10% from actuals, and the CFO decides whether to change it. The model got faster to build. The judgement stayed with a person who signs it.

Can AI build a financial model in Excel?

Yes, and Excel is where I'd keep it. Give the AI a one page spec and it will lay out the tabs, the time axis and the formulas for a 3 statement model in about an hour, against 2 to 3 days by hand. It still gets 1 or 2 links wrong in most builds, so the balance check, the cash tie out and a person reading every formula on the input and output tabs stay in place.

Can Claude build a financial model?

Claude can build one, and it is the model I use most for this work because it reads a long workbook and explains it back well. Give it the spec and a copy of the file, never the live one. Ask it to write formulas, not values, and to list every cell it changed. If it hands you a finished number in the chat, rebuild that number in the sheet before you trust it.

Will AI replace financial modelling jobs?

It will replace the typing, not the judgement. The hours an analyst spends wiring tabs, fixing ranges and formatting will drop by half or more inside 2 years in most finance teams I see. The person who decides what growth rate is defensible, who signs the model a lender relies on and who answers the buyer's questions in due diligence is still needed, and is worth more.

Where I come in

This is the work I do with CFOs one to one and inside finance teams. In the first session we write the spec and the assumptions table for the model you rely on most, and within 2 weeks the AI is rebuilding it on a copy with the 4 checks running. The forecasting side is in AI financial forecasting, the commentary that sits on top is in AI for FP&A, and the cash view it feeds is in AI cash flow forecasting. The map of the whole topic is at AI for CFOs, the end state is at the AI-native CFO, and the sessions are at coaching for CFOs. Which input in your model has no source written next to it? Start at meethayat.com and tell me.