Lease abstraction at deal-team speed.
We compressed a 6-day legal lease abstraction cycle into a same-day workflow, by automating the extract-review-route pattern with a human-in-the-loop interface.
Numbers we agreed to move.
The challenge
Lease abstraction was the bottleneck on every deal. Reviewers were doing first-pass extraction by hand, then second-pass review for accuracy — a pattern that scaled with headcount and nothing else.
What we did
We mapped the extract-review-route pattern, then shipped a focused tool: model does the first pass with confidence scores, reviewers see only the fields below a threshold, every override trains the model. No "AI assistant" panel grafted onto the side of an existing tool.
Where it landed
Reviewers stopped doing first-pass extraction entirely. The system handles 96% of fields above confidence threshold; the remaining 4% surface as a focused review queue with full source citations.
"We didn't think we were going to like it. Two months in, we won't ship without it."