AI Guides › Money & Business
What A Real Financial Model Needs That A Chat Can't Give You
By Nigel Guy · 2 min read
Asking an AI chat to "build me a financial model for X" produces a
plausible-looking set of numbers in a tidy table. It reads like a model. It
isn't one, because it's missing the parts that make a model actually useful
for a real decision — where the numbers came from, and what happens if
they're wrong.
The rule: a financial model is only as useful as its assumptions register
and its sensitivity range — a single set of numbers from a chat conversation
is a guess with formatting, not a model, until you can see what it depends
on and what happens when those numbers move.
The mechanism: what a real model needs
- An explicit assumptions register — every input labelled (revenue
growth rate, churn, cost per unit) with a stated source, not buried
silently inside a formula.
- Traceable formulas. You, or anyone else looking at it, should be able
to see how a number was derived, not just receive a final figure.
- Multiple scenarios, not one — best, base, and worst case at a
minimum, so a decision isn't hostage to a single guess turning out wrong.
- A sensitivity check on the assumption that matters most — what
happens to the outcome if that one number is off by twenty per cent.
- Your own verified data feeding the real inputs — your actual costs,
your actual prior revenue — not AI's estimate of typical figures for a
business roughly like yours.
Where AI genuinely helps
Use it to build and structure the spreadsheet and formulas — a real time
saver. Use it to stress-test: "what happens to the outcome if this
assumption is wrong by twenty per cent" is a good question to ask it. Do not
let it supply the core assumptions — growth rate, pricing, conversion — from
its general knowledge; those need to come from your own data or a verified
source. Review every formula yourself afterward, well enough that you could
explain it to somebody else.
What to skip
Skip asking AI for "typical" numbers for your industry and treating them as
your actual inputs — those are generic and may not apply to your specific
market or stage. Skip a model built around a single scenario presented as if
it were a forecast rather than a sketch of one possibility.
Guardrails
- This assumes you or someone advising you understands the basics of
financial modelling. If not, it's still worth having a professional build
or check the model for anything involving real money.
- AI-assisted spreadsheet mechanics still need checking for errors — formula
mistakes are easy to introduce and easy to miss in a document that looks
finished.
- Verify any industry benchmark figure independently before using it as a
load-bearing assumption; these figures date quickly and vary considerably
by market segment.
All 751 AI guides · JulieMango plans from £17/mo