Cognify AI Raised $24M to Automate Legal Due Diligence — Here Is How It Works
Legal due diligence is one of the most expensive, time-consuming, and error-prone processes in modern business. In a typical M&A transaction, junior lawyers at Big Law firms bill hundreds of hours reviewing contracts, flagging risks, and building data-room summaries — work that is high-stakes, repetitive, and deeply unpleasant. Cognify AI, a San Francisco-based startup, has spent three years building an AI system that can do this in a fraction of the time, and it just raised $24M to scale.
The Problem, Quantified
The average mid-market M&A deal ($50M–$500M) involves reviewing between 3,000 and 12,000 documents. At a blended billing rate of $400/hour and an experienced associate reviewing roughly 40 documents per hour, that translates to $30,000 to $120,000 in legal fees just for the document review phase — before any substantive legal work begins. Cognify’s system processes the same corpus in 4–8 hours at a cost that its CEO, Maya Okonkwo, describes as “less than the coffee bill for a three-day all-hands.”
How Cognify’s Pipeline Works
The technical architecture is more interesting than a simple “upload PDFs, get summary” pitch. Cognify uses a four-stage pipeline:
- Document parsing and classification. Contracts, board minutes, employment agreements, IP assignments, and NDAs are parsed using a fine-tuned layout model that handles scanned PDFs, handwritten annotations, and multi-column legal typesetting — all common in older data rooms. Each document is classified into one of 47 legal document types.
- Entity and clause extraction. A domain-specific NER model identifies parties, dates, governing law clauses, termination provisions, change-of-control clauses, and non-compete agreements. The model was trained on 2.1M annotated contract clauses, built in partnership with three global law firms who provided ground-truth labels.
- Risk scoring. Extracted clauses are scored against a configurable risk rubric. A “change of control” clause that triggers automatic termination on acquisition is flagged as critical risk; a standard confidentiality clause with a 2-year term is logged but not escalated. The rubric is customisable by deal type and jurisdiction.
- Synthesis and report generation. Finally, GPT-4o (with a custom system prompt tuned by Cognify’s legal team) generates a narrative due-diligence report in the format preferred by the law firm or corporate legal team, complete with citations back to specific document pages.
The Human-in-the-Loop Design
Cognify is careful not to overstate what the system does autonomously. “We are not giving legal advice. We are giving lawyers a much better starting point,” Okonkwo says. The system flags every high-risk clause with a confidence score and a reference to the original document. Lawyers review flagged items, accept or override the AI’s assessment, and their decisions feed back into the model via a continuous learning loop.
Early results from pilot customers suggest that junior associate hours on due diligence drop by 60–70%, and senior associate review time drops by 25–30% (since they are reviewing a cleaner, pre-processed output rather than raw documents). One pilot customer — a regional PE firm doing three to four deals per year — reported saving $380,000 in legal fees over 12 months of use.
The Competitive Landscape
Cognify is not alone. Harvey AI, Clio’s AI features, and Ironclad’s contract AI are all nibbling at adjacent problems. What differentiates Cognify, according to investors, is depth in the M&A due-diligence workflow specifically — not generic contract summarisation — and the proprietary training data from law firm partnerships. “The moat is not the model; it’s the annotated data and the workflow integrations,” says Farida Sadat, partner at Gradient Ventures, which led the Series A.
The $24M will go toward expanding the engineering team (currently 22 people), building deeper integrations with iManage and NetDocuments (the dominant document management systems at large law firms), and launching in the UK and Germany, where GDPR-compliant data handling requirements have historically slowed AI adoption in legal.