AI for Enterprise Legal Departments – Contract Analysis and More

Daniel Faggella

Daniel Faggella is Head of Research at Emerj. Called upon by the United Nations, World Bank, INTERPOL, and leading enterprises, Daniel is a globally sought-after expert on the competitive strategy implications of AI for business and government leaders.

AI for Enterprise Legal Departments - Contract Analysis and More

AI has numerous use cases in legal, from document search to compliance and contract abstraction. This week, we speak with Lars Mahler, Chief Science Officer for LegalSifter, about what’s possible with AI for legal departments today and how AI applications for legal teams, such as natural language processing-based contract analysis, work. In addition, Mahler discusses how lawyers at companies and data scientists work together to train machine learning algorithms.

He provides some insight into how a company has to make its way into the legal space and the challenges of training an NLP system and collecting data for it.

Subscribe to our AI in Industry Podcast with your favorite podcast service:

Guest: Lars Mahler, Chief Science Officer – LegalSifter

Expertise: data science and machine learning

Brief Recognition: Mahler earned his MS in Computer Science from Carnegie Mellon University. He spent four years at Accenture as a manager of marketing transformation before becoming the Chief Science Officer at LegalSifter.

Interview Highlights

(02:30) What’s possible today with AI in legal?

Lars Mahler: We’re seeing AI being used in several areas in the legal space. One area is in the area of contract review and negotiation. So if you think about a typical contract negotiation, you’ve got two parties. Usually, you’re starting out with a template, a draft contract from one party, and then the other party is having to review that contract. So you’re reviewing it, making edits, going back and forth, editing some revisions until both parties are satisfied.

So this process can be really painful. Contracts are painful. They’re not fun to read. If you’re the counterparty and you’re viewing the other party’s paper, it’s kind of a cognitive load. You’re having to read and understand their language; you’re having to identify provisions that are in the contract, but that you do not want in there…you’re having to identify provisions that are missing but should be in the contract, and you’re having to check provisions to make sure that they are compliant with your organization’s standards.

So for a lot of lawyers and legal professionals, this is probably one of their least favorite tasks. It’s hard; there’s not much in doing it well, but if they mess up, there’s a lot of risk either for their company or their personal reputation. So that’s how it’s typically done today, how it’s been done in the past. But now with AI, we’re starting to see AI-supported contract review. These applications…read your contract, they’re going to identify all the terms and provisions, tell you what’s there, and help you find those “hidden grenades.”

So for example…in a non-disclosure agreement, there’s a type of clause called the residuals clause. It’s not very common, but it has some adverse consequences for the disposing party. So it can spot those kinds of hidden or tricky clauses.

(05:00) Which of these clauses are findable with a pre-trained system and what does that look like?

LM: To really do well in finding the clause, you typically need a lot of examples, and they’re going to perform best on sentences that look very similar to what they’ve seen in the past. So really, any concept as long as you have enough data and enough variety in your training data, you can learn that concept pretty well. But to your point, if you’ve got a long sentence and you’ve got that concept buried in the sentence or worded in a tricky way, you may see that the AI misses in that context.

(06:30) Where did you guys start with training your model at your company?

LM: We started out with NDAs, and the reason for that is NDA’s are pervasive, everybody signs them. It’s the starting point to a business relationship. So it really seems like a great place to start. And then after that, we basically went sort of one contract type at a time based on the needs of our earliest adopters.

(09:00) How did you go about training your model?

LM: So we feedback loop. It’s called Sifter Trainer. So we get a false positive or false negative, our customers can report that to us. But the thing is you don’t want to just blindly take feedback and incorporate it into the model because they may not have the same understanding of what that model is supposed to be doing or that sifter is supposed to be doing.

So we have a review process, where when we get that feedback, we’ll review it. If we agree, we just accept it and it loops back in, makes the model better. If we disagree, then there’s a communication loop to the client to help them understand the intent of the sifter.

And really with NLP, things can get very ambiguous very quickly. So that’s why we have to have that sort of internal review step. Because if we just blindly accepted feedback, there’s a risk that we could sort of let the model drift off course and turn into something that we didn’t originally intend.

(12:30) What does AI in legal mean for the future of law and legal?

LM: You know I see all of these AI tools as productivity tools. I mean they’re really cool productivity tools pretty much for all the players. Can they figure out a way to use it? Do the tools solve the problem that they actually have? And if so, how are they going to incorporate those tools into their processes?

Lawyers are [sometimes] using it almost as a spell check or a second pair of eyes. So they’re using these to double-check and make sure that they haven’t missed anything. You know if they’re reviewing late at night or under tight deadlines, this is a second pair of eyes for them.

For large corporations, a lot of times you have in-house counsels who are overwhelmed, and they’ve got many people on the front line of sales, and they need contracts reviewed stat, but those contracts are taking two weeks to get through in-house counsel. So what they’re doing is they’re using this to take the knowledge and insight from their in-house counsel and inject that knowledge into the application, and then allow anybody on the front lines to review and negotiate contracts without going to the in-house counsel.

They’re using [AI] to make people more productive. They’re not using robots to replace lawyers. 2018 seemed to be a year of awareness and education in the legal space.

So when we were having conversations with law firms, in-house counsel, other groups, a lot of the questions they were asking is, “What is AI? What can it do? What are the options?”

Toward the end of the year, it completely changed. Everybody’s aware of what it is, what it can do. Now, they’re figuring out how they can use it in their business and if so, how? So I believe I don’t think there’s going to be a “Big Bang.” I just believe it’s going to be a steady, gradual adoption.

(16:30) How will AI affect legal in five years?

LM: I would say all of the work that lawyers hate will be solved by AI. It’s that routine stuff, the drudgery combing through hundreds or thousands of documents. AI is going to take that away. It’s going to let them focus their time on the brain work, the higher order of judgments and decisions and things like that.

I think these successful lawyers, they don’t have to be AI experts, but they will need to understand what AI does and the basics functions it can perform and how to use it well. And that should be taught in law schools now. Years ago, I think lawyers were being taught how to use search techniques in LexisNexis and things like that. I think a lot of that search is going to be automated in the future, so they may not need to know that.

(19:00) Where will AI become the norm first in legal?

LM: It’s actually been kind of all over the board. We’ve seen adoption from big law firms. They’re trying to keep their fees down and grow their profits, improve their margins. Now it’s not all big law, but certain sort of visionary law firms. We’re also seeing adoption within small law firms. They’re trying to differentiate themselves and kind of stand out from the crowd.

We’re also seeing adoption from in-house legal departments. These are people with fixed budgets and they’re being asked to do more and more, so they’re looking for tools to become more efficient. And then there’s a lot of software companies, like contract lifecycle management companies and other related providers, and they are trying to compete in a very crowded market. So they’re embedding AI to try to get that sort of next step advantage.

It’s hard to make a prediction. I would predict big law firms will be the slowest the change. 

Subscribe to our AI in Industry Podcast with your favorite podcast service:


Header Image Credit: Business First Family


Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the 'AI Advantage' newsletter: