Two brokers asked me last week when AI is going to start replacing brokers. It is a fair question, and the honest answer is that the numbers we actually have point somewhere else entirely.

Start with how many firms are using it. In late 2023, around 12% of UK businesses with ten or more employees used AI. By June 2026 that was around 35%. Roughly three times as many in under three years. Half of large firms now use it. More than a quarter of the smallest ones do.

That is the number everyone quotes. Here is the one nobody does.

Over the same period, the average number of AI technologies used by a business that had adopted any went from about 1.4 to about 1.6.

Read that again, because it is the whole story in one line. Adoption widened enormously and barely deepened at all. Firms tried it. Almost none of them built on it.

The gap between quicker and better off

The government asked businesses using AI what it had actually done for them. Three quarters said it improved the productivity of their workforce. More than half said they had developed new or improved processes.

Then the same survey asked about revenue. Seventy-seven per cent reported no change at all. Twelve per cent reported an increase. One per cent said it went down.

Same firms. Same questionnaire. Everyone got quicker, and almost nobody made more money.

I do not think that is a contradiction, and I do not think it means the productivity gains are imaginary. It means something duller and more useful. Time saved is not money made. An hour you get back on Tuesday only becomes revenue if there was something waiting to fill it, and in most firms there wasn't, or nobody had decided what.

What it is costing to get there

Cost is the single biggest barrier reported, and this is the part that surprised me: it bites hardest at the top. Ninety per cent of large organisations call high cost a significant barrier. For micro businesses it is 72%.

The firms with the deepest pockets are the ones most likely to say this is expensive. That is not what you would expect from a technology quietly paying for itself.

And the replacement question

Around half of businesses reported that AI had no impact on headcount whatsoever. Among medium-sized firms, just under 7% reported any reduction at all.

Not zero. I am not going to pretend nobody has lost a job to this. But a technology that a third of firms now use, and which has changed the headcount of the overwhelming majority of them not at all, is not a technology that is replacing people. Whatever it is doing, that is not it.

The honest caveat

These are self-reported figures, and the government's own report says so. Businesses are describing their own experience, not submitting audited accounts. A firm that has just spent money on something has reasons to say it helped.

Which, if anything, makes the revenue number harder to explain away, not easier. If there is a bias in self-reporting, it runs towards optimism. And even with the thumb on the scale, three quarters said it made them faster and three quarters said it made them no richer.

What I think is actually happening

There is an enormous amount of money riding on the story that this replaces people, because that is the story that justifies the valuations. The evidence in front of us describes something much smaller and much more ordinary: a tool that makes people quicker at language, adopted broadly, deployed shallowly, and not yet pointed at anything that shows up in the accounts.

That is not a failure. It is what the early part of every useful technology looks like. But it does mean the firms who work out where to point it are going to look very different from the ones who simply bought it.

In part two I want to look at why, and at the one thing in a regulated firm that cannot be handed over no matter how good the tool gets.