Grab has developed an AI-powered legal review systemthat automates the first-pass screening of non-disclosure agreements (NDAs) while ensuring human experts handle final decisions.
AI-powered legal review system
Developed through a collaboration between Grab’s Legal team and Grabber Technology Solutions (GTS), the company’s internal corporate IT and enterprise technology department, its pilot project focused on automating the initial review of NDAs. The company initially experimented with a Large Language Model (LLM) capable of reading contracts, extracting clauses, and recommending whether to approve, flag, or escalate agreements for further review.
However, based on early tests, the tool struggled to apply policy checks consistently across contracts, missed some non-standard wording, and occasionally generated recommendations that lacked the accuracy needed for legal reviews.
AI-human hybrid system
Because of this, Grab developed a hybrid decision engine combining AI, rule-based automation, and human oversight. Multiple AI models extract facts and citations from contracts, while a rulebook by Grab’s Legal team evaluates those facts against company policies. Human reviewers then validate the results before approvals are finalised.
The engine was tested against hundreds of contract variations, including compliant, low-risk, and high-risk scenarios, achieving 100% rule-evaluation accuracy with zero false positives and reducing NDA review time from about 60 minutes to less than 2 minutes, showcasing its effectiveness.
The company reported that the system achieved 100% rule-evaluation accuracy with zero false positives, while reducing NDA review time from about 60 minutes to less than 2 minutes.
The company plans to move the NDA review engine from sandbox testing into production before expanding the technology to commercial contract reviews and other knowledge-intensive business functions.
