AI adoption: accountants aren’t trained to try things twice.

If a tool you’d just bought produced an unbalanced journal, you’d take it back to the shop.

Of course you would. Accountants are trained for things to be right. The whole job runs on a kind of zero tolerance for output that doesn’t tie out – which, in a profession built on accuracy, is exactly as it should be.

But AI works the other way round.

Every time I’ve heard of someone getting real value out of these tools – and honestly, that’s still mostly outside finance – they had to iterate. Multiple times. They asked for the thing. The thing was rubbish. They asked again, slightly differently. Still rubbish. Eventually, on the fifth or sixth go, something useful turned up.

If you’re like me, you might have seen comparisons with an intelligent intern, but in my experience the intern is actually quite stupid in a lot of ways so it’s tempting to fire them - but if you keep going, it gets smarter fast.

That’s just how it works. The first answer is almost always wrong (sometimes spectacularly wrong!). The skill, if there is one, is in being prepared to keep going past the first wrong answer.

Which is precisely what an accountant’s instinct rebels against. If you’ve spent years training yourself that accuracy is all that matters, the AI workflow feels deeply suspicious. Why am I retrying it? Why am I trusting it more this time?

Reasonable questions. You’re not actually trusting it more. You’re just plugging the holes, finding more, and plugging those ones too – a different skill from “running” a piece of software. It actually has a lot more in common with software development.

It’s a big mind shift. Carrying on from that first answer doesn’t come naturally. Keep going though!