Three quarters of small firms use AI. Fewer than a third have increased revenue.

The argument about whether small firms will use AI is over. It ended quietly and nobody announced it.

So the thing is in the building. What is not in the building is the result. Three quarters of small firms have the tool, and fewer than a third of them have seen revenue go up because of it.

The number we were about to publish, and didn't

A framing was going round while we were writing this: that 72% of small firms use AI, only 10% use it extensively, and that governance — not budget, not technical skill — is the primary predictor of which is which. It is a good line. It is the exact line this series would like to be true.

We went looking for it in Clio's own published material and could not find it. The figures we could find in Clio's own words are the four above, and they do not say that. Clio's own write-up of the gap attributes the mid-sized lead to resourcing instead — firms with "larger budgets and dedicated IT teams".

So we are not going to tell you governance is the cause. We would be quoting a claim we could not source, in a post whose entire argument is that you should check where things come from.

What the numbers do support, stated carefully

There is a second Clio report, on mid-sized firms, published 9 March 2026. Two figures from it: 86% of mid-sized firms report using AI, and 60% report having formal policies guiding AI use.

It is tempting to lay that beside the revenue figure and draw a line. Do not, and here is why: the 60%-with-policies number is about mid-sized firms, and the 60%-with-revenue number is about enterprise firms. Different cohorts, different reports, different questions. Reading them as one sentence is how a correlation that was never measured ends up in a slide deck.

What is left after you take that apart is still worth something. The firms that have written down what they are doing and the firms getting paid for what they are doing are both the larger firms, and the small end has the tool without either. That is a real observation. It is not a finding, and we are going to keep calling it the first thing rather than the second.

(The ABA's 2024 technology survey puts AI use much lower, at 30% of respondents — a different sample asking a different question, and useful mainly for its slope: roughly tripled in a year.)

The thing nobody counted

Here is what none of these surveys measures, and it is the number that would actually change what you do on Monday.

Of the firms that are using AI, how many decided to?

"75% of small firms are using AI to complete legal work" is a count of firms where somebody is using it. It is not a count of firms where a partner sat down, looked at the options, picked one, and told everybody which one. At the small end those are wildly different populations, and the difference between them is a person opening a browser tab.

That is the actual gap, and it has nothing to do with budget. The tool is free or nearly free. The decision is what is missing.

What to do

Before you evaluate a single product, find out what is already being used. One message to everyone, one question: what have you put into an AI tool this month? Not to catch anybody — to get the list, because you cannot write a rule about a set you have not seen.

Watch for two things in the replies. A tool nobody had mentioned, and an account that belongs to a person rather than to the firm. Either one changes what a policy has to say, and neither shows up on a survey.

It takes an afternoon and it does not need to be a lawyer's. An operations lead can send it, collate the answers and hand back the list; the partner's part starts after the list exists.

What it costs to skip is not measured in hours. It is that on the day a client asks which AI tools have touched their matter, the honest answer is that nobody knows. That is the sentence this afternoon removes.

If you would rather someone ran that inventory alongside you, the consulting page is the way in. Either way, get the list first — everything else in this series depends on it.