Where the money goes in the AI race?

In 1958, an American economist by the name of Willard Cochrane found something strange.

‍

After farms mechanized and yields skyrocketed, the farm profits did not keep up.

‍

When the initial technology boom happened, the first farmers to jump on the bandwagon reaped huge rewards. Prices stayed the same, but costs plummeted. As soon as the first farmer had made their money, their competitors copied their act, bringing prices back down to the new cost level, erasing any profit. Those who resisted were then faced with the choice between adopting the technology and continuing to struggle against the market forces. And so, the technology treadmill was born.

‍

I think finance is about to step onto the technology treadmill.

‍

Almost 70 years later, an August 2026 Fed paper out of San Francisco predicts the same pattern in the big companies' race to embrace AI. After an initial burst of market share, the early adopters find themselves undercut by a wave of smaller competitors, who sponge up a growing share of the market.

‍

(Though it hardly needs saying, these technologies are getting cheaper far faster than Moore's law. A Stanford University study, the 2025 AI Index, found that the price of using comparable models to GPT-3.5 dropped more than 280x between November 2022 and October 2024.)

‍

But to me, the most interesting part of the technology treadmill is what happens to the early movers. Let's say that there is an accounting practice that provides a certain service to its clients for a price of AED 10,000 per month. It takes the accountants AED 7,000 to provide this level of service, leaving them with AED 3,000 as the amount of money they make per client before overheads.

‍

By using AI, the accountants can provide the same service for AED 5,500, after subtracting the costs of technology and the proportionate share of wages for the workers. The accounting practice has increased the amount of money it makes per client by AED 1,500. But as soon as one of their competitors starts offering the same level of service for AED 8,000, they have to follow suit, cutting their profit per client to AED 2,500, which is AED 500 less than before they adopted AI.

‍

Having spent two years working on month-end closes and reconciliations, I can tell you that speed has its appeal. However, this is where I would start questioning the assumptions in anyone's business case for adopting AI: that we improve while our customers and competitors remain frozen in time.

Count the savings honestly

If you save 20 hours, but your payroll budget remains the same, it is not a cost saving. It is capacity. It is absolutely valid to present it as such, but it should not be added to the slide as a 20-hour cost saving.

‍

The flip side of making work processes disappear is that the bill for them can inflate, especially if the technology gets used elsewhere. Microsoft is already talking about adding usage-based charges on top of its subscriptions in several of its divisions. If they convince their clients to do more analysis, they will make more money, even if the cost per analysis is lower.

‍

What we need to be measuring is the full cost of the entire service provided to the standard our clients expect, including the cost of the accountants' scrutiny of the figures.

What to do with the time

One way to respond to the pressure on billings is to cut your prices. But another option is to use the time saved to improve the service offering.

‍

If one of your clients is about to make a large inventory purchase just before payroll, but two of their customers are late paying them, the accountant can help the owner avoid taking on so much debt. This is, arguably, a more valuable service than simply accelerating the month-end close. At the same price, the client gets the accountant's assurance that they will not find any unpleasant cash surprises waiting for them at the end of the month.

‍

AI makes analysis cheap, but knowing what to do about the results of that analysis is another matter entirely.

Getting off the treadmill

Some farmers later got off the treadmill by switching to organic farming and selling their produce at a higher price. They did not increase their yields or productivity, but they managed to make more money by providing a differentiated product.

‍

The same principles apply to accountants. The value of our services is not going to be permanently tied to how much work we get through, so we need to be thinking about what our clients would want us to spend more time on.

‍

And that, perhaps, is the question I would take to my next meeting about AI: once faster work becomes the norm, what will make a customer choose us?

‍