Why most law-firm AI pilots stall - and the few that don't

Every firm I speak to has run an AI pilot. Very few can tell me what happened next. The demo lands, a partner is impressed, a small group gets access for ninety days - and then the thing quietly evaporates. No decision to kill it, no decision to scale it. It just stops being talked about.

After watching this pattern repeat across firms of very different sizes, I’ve stopped believing it’s a technology problem. The models are good enough. The pilots stall for reasons that have almost nothing to do with the AI itself.

The pilot is designed to impress, not to decide

Most pilots are scoped to produce a “wow” moment. That is exactly the wrong goal. A pilot should be designed to answer a single question: does this change the economics of a real task that real fee-earners do every week?

If you can’t name the task, the baseline, and the threshold at which you’d roll out before you start, you haven’t designed a pilot. You’ve designed a demo with a longer runway.

Nobody owns the boring middle

The gap where pilots die is the boring middle: the work between “this is impressive” and “this is now how we do things”. That work is unglamorous - integration, permissions, training, a champion in each team, a feedback loop when the output is wrong.

The firms that succeed assign an owner to that middle who is measured on adoption, not on having run the pilot. The firms that fail assume the technology will cross the gap on its own. It never does.

The few that don’t stall

The pilots that scale share three traits:

• A painfully narrow first use case. One task, one team, one measurable outcome. Breadth comes later.
• A named owner with time protected to do the work - not a side-of-desk volunteer.
• A pre-agreed decision point. Everyone knows in advance what evidence would justify a rollout, and what would justify stopping.

None of that is about the model. It’s about treating adoption as the actual product.

What to do on Monday

If you’re about to run a pilot, write down the task, the current cost of doing it, and the number that would make you roll it out firm-wide. If you can’t, pause. You’re not ready to pilot - you’re ready to demo, and demos don’t change how a firm works.

The technology is no longer the hard part. The hard part is the same as it has always been in professional services: changing what people actually do on a Tuesday afternoon.