Implementation bottleneck in legal AI
Healthcare company associate GC on where legal AI products break down
The main bottleneck in legal AI is not the demo, it is the messy work of fitting a tool into real legal workflows without breaking on day two. In lean in house teams, a product has to handle intake, first pass review, approvals, and user mistakes with almost no hand holding. If setup drags, support is slow, or edge cases stall the process, even a strong demo loses the deal because the team still has to do the rescue work itself.
-
This interview ties rejection to implementation, not novelty. The legal lead describes Luminance as promising document based playbooks, but requiring so much setup and training that the team never got it fully working, with support often taking 24 to 48 hours and still not fixing the problem.
-
Large firms report the same pattern at bigger scale. Even for Harvey and Legora, deployment typically takes about six months because security, procurement, onboarding, training, and internal rollout all have to clear before a pilot is live. Legal AI vendors that act like this should happen overnight tend to misread how legal buying actually works.
-
The clearest contrast is workflow embedded tools. Spellbook is built inside Microsoft Word and sells around the contract job itself, which makes it easier to start with one user and expand. That bottom up path fits in house teams better than broad legal AI platforms that ask buyers to add another separate app and another layer of change management.
The next winners in legal AI will look less like standalone chat products and more like tightly embedded workflow software with real implementation muscle. The bar is simple, a non legal employee should be able to submit a request, get routed correctly, recover from mistakes, and keep context intact, with legal stepping in only where judgment is actually needed.