Scott Foster, an IP litigator and partner at Lewis Silkin, was prepping for trial one night when he hit a wall. He was spending hour after hour copying text from transcripts and PDFs, reformatting each excerpt to match his Word documents, and manually building citations. Across a full trial that work could balloon to 50 hours: difficult to bill, easy to write off, and always at risk of a wrong page number slipping through. He mentioned it to Jack Newton, founder and CEO of Clio. Jack immediately saw the problem was massive and solvable, and brought in Morgan Maguire, a lawyer by training with 15 years building legaltech products, to build it.
‍
In the summer of 2023, the three of them started building LexSelect. The first product was a Microsoft Word add-in that automated the copy-paste-and-cite workflow. What had taken 50 hours now took 4.

Building the Word add-in taught them something. The citation workflow was solvable. But to do anything useful with a legal document, you first had to make it readable, and every tool in the stack was failing at that, flattening documents into text and discarding the structure that gives them meaning: hierarchy, defined terms, cross-references that only make sense in context of the whole.
‍
That same failure, multiplied across billions of pages, was the constraint underneath every serious AI deployment in legal. Not model quality. Input quality. The more autonomous AI becomes, the more essential the data layer beneath it is. So they followed the problem to its root and built the layer.