Cutting Contract Review Time by 73%

A global law firm with 120 lawyers was processing 200+ contracts per month using junior associates for initial review. Each contract took 3-4 hours of manual review time, flagging non-standard clauses and summarising key terms. Partners were frustrated with the inconsistency and cost. The firm had tried two off-the-shelf legal AI tools and rejected both - one was too US-centric, the other hallucinated clause citations that didn't exist.

We spent the first two days reviewing 50 contracts to understand clause taxonomy and risk criteria. The system we built used a retrieval-augmented approach: each contract was split, embedded, and indexed against a curated library of standard commercial law clauses. GPT-5.4 ran the classification and flagging; a human-in-the-loop review step remained for high-risk clauses.

The UI was a simple web interface integrated into their existing document management system. Associates saw flagged sections with citations. Partners got a daily digest of high-risk contracts.