“My Other Lawyer is a Robot” – LawGeex Automates Contract Review

Artificial Intelligence is already helping doctors examine MRI scans, travel agents book flight tickets, and accountants keep books. But the legal industry has been somewhat stubborn when it comes to AI. “The assumption is that as long as it’s read by humans it’s fine, but there’s no ground for this in any other industry in the world,” says Shmuli Goldberg, VP of Marketing at LawGeex.

Founded in 2014 by international business lawyer Noory Bechor and machine learning guru Ilan Admon, LawGeex is a Tel Aviv-based legal tech startup focused on contract review. It closed US$7 million in Series A funding last March, principally from Indeed.com owner Recruit Holdings, Lool Ventures, and LionBird. The company has a team of close to 50 in its Tel Aviv R&D units and New York sales office.

This February, LawGeex created a stir in the legal industry when its algorithms outperformed human lawyers at contract review in a study conducted by Stanford, Duke University, and the University of South California.

Researchers asked 20 experienced lawyers and the AI to examine five previously unseen contracts composed of 153 legal clauses copied from standard Non-Disclosure Agreements (NDA). The AI used just 26 seconds to complete its analysis with an accuracy rate of 94 percent. Human lawyers meanwhile required nearly 92 minutes to achieve an average accuracy rate of 85 percent.

A Fortune 1000 company can deal with up to 40,000 contracts annually. Many companies are dissatisfied with the existing organizational contract review process, which is an essential but laborious task. Confidentiality agreements, in particular, take a week or even longer to review.

Goldberg says the LawGeex AI can cut the contract review cycle from a week or a month down to one hour, and reduce the time spent by a lawyer reviewing the contract from hours to minutes — adding that law firms which bill by the hour are not as interested in automated efficiency. LawGeex can also review NDAs, purchase orders, service agreements, software license, sales and other types of contracts with a consistency, says Goldberg, that tired office-trapped lawyers on a Friday evening cannot match.

The ideal LawGeex client is an in-house legal departments of medium to large-sized enterprises with over a thousand contracts per year. The client will first making clear their precedents and definition of terms. The AI, pre-trained to understand what they mean, will then identify pertinent elements in the contract and make sure these align with the client’s specifications.

The default AI application for document review is natural language processing (NLP). However, LawGeex discovered that NLP performed poorly in contract review due to legalese. Although natural language can have multiple interpretations, legalese is specific and rigid.

To cope with this problem LawGeex developed two new AI algorithms. A Legalese Language Processing (LLP) algorithm first trains the neural network with as many contracts as possible. The AI learns to understand terms such as “non-compete” and “disclosure,” and identify terminology-relevant clauses in the contract. A Legalese Language Understanding (LLU) algorithm works on top of the LLP, translating the legalese into legal concepts to help the AI understand unfamiliar clauses.

The hardest part of building the process, says Goldberg, was finding half a million contracts as training datasets, as there are no standard learning or training sets in the legal world — such as ImageNet for computer vision or MNIST for handwritten digits. LawGeex enlisted the help of experienced lawyers from top US law firms and spent three years hunting for shared and disclosed contracts.

Goldberg says what separates LawGeex from automated contract service providers such as Legalsifter and Kiva Systems is that “they are all based on the same premise, if I have thousands of contracts and want to find one thing, they are the best for that. Whereas our tool looks at one contract and answers the question ‘can I sign this?’ When someone comes to the lawyer, hands them the NDA and asks if he can sign this, our tool opens the NDA up, looks the potential hundreds of issues, highlights what’s relevant and irrelevant and passes it onto the lawyer.”

“This isn’t a new technology,” says Goldberg, “and we are not saying that AI is taking over the legal world, we are saying the opposite: AI has taken over the legal world.”