The accountants on stage in London were sure AI is no threat. What if they're wrong?

I had a bit of a feeling of meeting the ostrich with its head in the sand in London.

At The Digital Accountancy Show, several accountants and a founder sat on stage in front of a packed room. The founder was on his third company. The session was called "The AI Doomsday Clock: How far are we from AI replacing accountants?". The accountants were quite certain. They had tested both ChatGPT and Claude, and the conclusion was clear: The technology simply wasn't good enough. It made too many mistakes, so the accounting profession had little to fear.

It made it worse that this was said from a stage. From a stage, an opinion becomes a message.

Then the founder took the floor, and he was at least as certain, but his conclusion was the exact opposite.

"You should be extremely worried."

His argument was simple: It's only a matter of time before a technology company offers the entire accounting process through AI agents. And he made a point I think the profession should take seriously: It's the client who decides whether she wants to use an accountant. In Norway, a company is responsible for its own accounts, but is not generally required to buy accounting services from an accounting firm. If the technology eventually does the work well enough, why wouldn't more companies choose to handle it themselves?

The client is in the driver's seat. And it's also the client who decides whether she wants anyone in the passenger seat.

I have to admit I was more worried about the accountants' reaction than about the founder's prediction.

The accountants had a real point: in accounting, precision isn't something you can trade away, and today's models make mistakes. But that's an argument about how far the technology has come, not about where it's heading. Testing today's model and drawing conclusions about the future is measuring the starting point instead of the direction.

We've worked with AI for a long time at Capassa, so it wasn't the technology that scared us in London, but the reaction to it. What we see from Capa, our own AI agent, is that it is already far faster than us humans at going through all the clients an accountant is responsible for, finding errors and deviations, and making recommendations. It analyses, systematises, documents and communicates far faster, and in several areas better, than a human can. Among other things, we've written about how Capa found an error in our own numbers.

A general-purpose chatbot, especially the free version, is something different from a model built to analyse the client's own accounting data. It's the latter the accounting profession should measure itself against.

London confirmed something we've been focused on for a long time: What can be automated will be automated, and then it's not enough to defend the value of the tasks that are disappearing. The value has to be created in other ways.

The client is rarely interested in how the accountant reconciles the accounts. The client wants to know whether the business can afford to invest in a new tractor, hire more people or go after a new market. Where are the risks? Where are the opportunities? What should we do now?

The client isn't necessarily paying for the analysis, but for help making better decisions.

If the accountant can help provide those answers, ideally before the client has even spotted the need, I believe the profession has a fantastic future.

But if not? Then it's worth remembering that accounting firms don't have a monopoly on financial insight. The banks have the clients. The accounting systems have the data. The consultants have the advisory expertise. And the technology companies are building solutions that can combine more and more of this. All of them can become competitors for the value the client actually wants to pay for.

That's why I hope Norwegian accounting firms won't let today's limitations in AI become an excuse for complacency. Because the decisive question isn't whether ChatGPT can do an accountant's job today. It's whether the client will still need to buy the same services from an accounting firm a few years from now.

Three questions for your next management meeting

If I were in an accounting firm and had heard both sides of that debate, I would spend the next management meeting on three questions:

1. Which of our tasks could a client do herself with today's AI, if she wanted to?
2. Which of them will she be able to do herself in two years?
3. And what do we deliver that the client would still need, even if an AI agent could keep the books?

If the answer to the last question is empty, that's where the work lies.