At the demo. Some lawyers have become very good at producing something that looks impressive in a meeting and very bad at turning it into something a fee earner actually uses six months later. Everyone can build a prototype now, that part got easy. The part that stayed hard, and that most firms skip, is product thinking: the UX, the adoption path, the workflow it has to fit into. Without this, that prototype will just never be used.

The second failure is mistaking a vendor purchase for a strategy. Buying something off the shelf and calling it innovation isn’t building anything, it’s renting the same commodity model everyone else has access to. The differentiation was never in the model. It’s in the proprietary judgement and workflow you build around it, and that’s exactly the layer firms tend to underinvest in.

The third is governance, or the lack of it. Most firms have no kill discipline. Ideas get funded on enthusiasm rather than evidence, nobody owns the thing once it ships, and pilots just sit there half-supported because there was never a named owner or a route to production in the first place. A prototype that nobody kills and nobody scales isn’t a failure you notice immediately. It’s a quiet one. It just sits there, burning budget and credibility, until someone finally asks why the firm has twelve AI pilots and nothing in production.