Will AI replace accountants? 30% fewer jobs in 3 years

Will AI replace accountants? The production work, yes. It already does ours. My bet is that Australia has 30% fewer accounting jobs within three years, and that an accounting firm running on AI can serve all of its clients with half the people. The accountant who signs off on the books stays, because that signature is what you are actually paying for.

That bet comes from running Hourglass on AI, not from a report. Accounting has clearly defined rules for what you can and cannot do. It is also an industry where getting it wrong costs real money, with penalties and a reputation on the line. The first part is why AI takes the work. The second part is why a human keeps the last word.

What will you still pay an accountant for?

At Hourglass we call any AI job a human sandwich. A human at the start, AI in the middle, a human at the end. The goal is to make both human layers as thin as possible.

The first layer is the prompt. With good context already set up (think a company brain that knows your business) the prompt gets very short. AI does the bulk of the work in the middle. The last layer is the approval, and that is the one a client pays for.

Because the service an accountant sells is risk transfer. You do not want to be on the hook yourself for the accuracy of your accounts, the same way you do not want to carry your own legal risk. So you always want a person doing the final review, and you will keep paying well for that stamp.

What happened when we let AI do our own books?

We did not read about this. We ran it on ourselves.

With Xero you can point AI at reconciling accounts. It is the easiest use case there is. Point it at receipts and invoices and have it reconcile them against the bank. It works like a detective. It finds what matches, finds what is still missing, cleans up the books and follows up with you on the gaps.

Our involvement was the prompt up front. Then a few questions from the AI. Then it was done. The job that used to mean a person dragging transactions around, sorting, finding the misses and going back and forth to work out what an expense was even for, had become a prompt and a short conversation.

Then our accountant opened the finished books.

One run of our own books, so not proof that AI accounting never errs. But it is the test that matters, because it is the top slice of the sandwich checking the middle. The production work passed the sign-off untouched. That is the whole prediction in one example.

Which accounting jobs does AI replace first?

Junior and middle production roles. That is exactly where the work AI does well sits, and those roles get crushed.

What remains is the top and a different kind of middle. The sign-off and the deep accounting knowledge sit with the senior people. The middle layer becomes account managers more than accountants, people who help the senior partners sell and hold the client relationships. The firm of the near future is about final approval and client relationships.

So there are two bets in this post, and they are different.

1. An individual firm needs 50% of its people to serve 100% of its clients. I think that one is already true.
2. Australian accounting employment falls by about 30% within three years. That one has a brake on it.

Why hasn't your accountant's bill dropped yet?

Not because of the AI. Because winning customers is hard.

Firms will move to AI, 100%. The question is how fast they can reset their prices and win new clients off the back of it. Switching costs in accounting are high. How often do you actually see your accountant? Most people stay unless the gap is big.

My estimates of the gap an AI-heavy firm needs to open, for a business owner choosing an accountant.

Michael Batko's estimates of the discount an AI-heavy accounting firm needs to offer. Personal estimates, not market data.

So the race is not about who has the AI. Every firm will have it. The race is about which firms have the hustle to cut price, grow fast and own the market while the others adjust. If none of them do, my three-year timeline is wrong.

Where do you still need a human accountant?

Two places.

The final sign-off or check always sits with the accountant. That is the risk transfer, and it is the whole point of paying a firm.

The second is anything without a black-and-white answer. Advice on entities, cost and tax optimisation, the "you don't know what you don't know" questions. Those are better discussed in person with someone.

What would I do before my next accounts hire?

If I owned a 20 to 100 person Australian business and was about to hire another accounts person, I would not post the job first. I would interview the team and document what the role actually needs to solve. The repetitive, rules-heavy parts are where AI automation pays back. I would automate as much of that as possible, or move to an accounting firm that is deep in AI, because I would expect it to be more accurate and probably cheaper too. That is what our AI audit is for.

Then I would keep an accountant for the two things above. The sign-off, and the conversations where the rules do not settle the answer.