The problem
A fast-growing legal tech platform had built a strong client base reviewing commercial contracts for mid-market businesses. Their process was entirely manual: a senior lawyer would read each contract, extract key clauses, flag risks, and produce a summary report.
At 4+ hours per contract, the firm was capped at 8 contracts per week regardless of how many clients they took on. Hiring more lawyers was expensive and slow. They needed a fundamentally different approach.
What we built
We designed a three-stage pipeline:
Stage 1 — Structured extraction. A fine-tuned extraction model identifies and classifies clauses (indemnity, liability cap, termination, IP ownership, governing law) from raw contract text. Unlike a generic LLM, this model was trained on the firm’s own contract corpus, so it understood their specific taxonomy.
Stage 2 — Risk scoring. Each extracted clause is scored against a configurable risk rubric. Clauses that fall outside acceptable parameters are flagged with the specific reason — “liability cap below standard threshold” rather than “this looks risky.”
Stage 3 — Report generation. A final LLM call synthesises the extractions and flags into a structured summary report matching the firm’s existing template format.
The entire pipeline runs in under 90 seconds for a typical 30-page contract.
Integration
We built directly into their existing document management system via webhook. When a contract is uploaded, the pipeline triggers automatically. The lawyer receives a draft summary and risk report that they review and sign off — rather than producing from scratch.
Results
- Review time per contract: 4 hours → 22 minutes
- Weekly throughput: 8 contracts → 40+ contracts, same team
- Risk flag accuracy: 94% precision vs. senior lawyer review
- Time to value: 6 weeks from kickoff to production