AI Guides › Money & Business
Reading Your Own Numbers Before You Ask AI To Explain Them
By Nigel Guy · 3 min read
It's tempting to paste a spreadsheet or a set of accounts straight into an AI assistant and ask "what does this mean" before you've looked at it yourself — it feels efficient, and the summary that comes back always sounds coherent. The problem is that a coherent-sounding summary of numbers you haven't looked at yourself is very hard to sanity-check. You end up trusting a reading of your own business that you couldn't personally confirm or challenge, because you skipped the step where you'd have noticed if something looked off.
The rule: form your own rough read of the numbers first, then use AI to check, extend, or challenge that read — never as the first pass over data you haven't looked at yourself.
The mechanism
- Look at the raw numbers yourself before opening any tool. Not a deep analysis — just enough to notice the shape: is revenue up or down on last period, roughly, and does anything jump out as obviously wrong.
- Write down your own one-line take before you ask anything of an assistant — "margin's tighter than last quarter, not sure why" is enough. This is what you'll check the AI's answer against.
- Then hand over the detailed numbers and ask for the breakdown — what's actually driving the change, which line items moved most, what the trend looks like over a longer period than you could hold in your head.
- Compare its answer to your own one-liner. Agreement is a good sign, not proof. Disagreement means one of you is wrong, and it's worth finding out which before you act on either version.
- Only then use it to model forward — a forecast, a scenario, a projection — because a forward model built on numbers you haven't sanity-checked just extends whatever error is already in the reading.
What to skip
Skip pasting in a full year of transactions on the very first pass and asking for "insights" with no question of your own attached — a vague prompt against a large, unfamiliar dataset is exactly the setup where an assistant will produce something confident and plausible-sounding that you have no independent way to check. And skip treating a forecast as more reliable than the historical read it's built on; a beautifully modelled projection is only as good as the numbers underneath it.
What to skip
Skip skipping this when the numbers are bad news, too — the temptation to hand over a rough quarter without looking closely yourself first, hoping the summary softens it, is exactly when independent verification matters most.
Guardrails
- This isn't about distrusting AI's arithmetic — a competent tool will total a column of numbers correctly. It's about not outsourcing the judgement of what those numbers mean before you've formed any judgement of your own.
- Anything feeding into a filing, a tax return, or an external report needs the same check by a qualified accountant that it would need without AI in the loop — a coherent summary isn't the same as a compliant one.
- Real numbers, especially anything tied to customers or payroll, deserve the same care over what you paste into a tool that any other sensitive business data would.
- If your own read and the AI's disagree and you can't work out why, that's the point to bring in a second human opinion, not to just pick whichever answer you prefer.
All 751 AI guides · JulieMango plans from £17/mo