Buying AI is easy. Getting lawyers to change how they work is not.

The most repeated observation in legal technology this year is that the difference between a novice and an expert user of the same software is far larger than the difference between competing products.

That should be an uncomfortable finding for everyone in this industry, including us. It suggests most of the argument about which platform is better is happening at the wrong altitude.

What we see across firms supports it. Take any two teams with identical access to identical tools. One has restructured how a document review actually runs. The other has a group of individuals who occasionally paste something into a box and are mildly pleased with the result. The gap in output between those two teams is enormous, and no procurement decision created it.

So why does adoption stall? Having watched a lot of it, the reasons are more mundane than the industry pretends.

Lawyers do not know what the tool is capable of, and there is no obvious way to find out. The capability is not visible from the interface. You have to already know that something is possible in order to try it.

They try one badly framed request, get a mediocre answer, and file the whole category under not good enough yet. This happens constantly and it is entirely rational. A lawyer who gets a poor answer from a junior twice stops asking that junior.

Associates do not know what they are allowed to upload. In the absence of a clear firm position, the safe behaviour is to use the tool for nothing that matters, which means using it for nothing.

Partners have no time to learn, and no incentive structure that rewards spending three hours getting worse at something before getting better.

And firms run one training session, watch usage spike for a fortnight, and record adoption as complete.

The common thread is that every one of those is a change management problem, and firms keep treating them as software problems. A more expensive tool does not fix any of them. Neither does building your own, which is worth noting given how much money is currently moving in that direction.

The interventions that seem to work are small and unglamorous. Choosing one workflow that a team does every week and rebuilding just that one properly. Making it explicit, in writing, what may and may not be uploaded, so the default stops being avoidance. Having someone senior and visibly busy use it in front of others, which does more than any training deck. And measuring something other than logins.

The firms getting real value are not the ones that bought best. They are the ones that decided, specifically, what they were going to do differently on a Tuesday.