The Center for Data Innovation spoke with Lars Mahler, chief science officer and co-founder of LegalSifter, a Pittsburgh-based company that combines human expertise with artificial intelligence (AI) to deliver affordable legal services. Mahler discussed how AI can help automate contract review and provide expert advice to both lawyers and non-lawyers alike.
Daniel Castro: What was your inspiration for creating LegalSifter?
Lars Mahler: The sinking feeling of dread that I had every time I needed to review a contract. For most of us, signing a contract can be intimidating. What am I signing? Is there anything in here that might hurt me? And most people reviewing an employment agreement or lease typically can’t afford to pay a lawyer to review it. And even if they can, they don’t want to wait two to three weeks to hear what the lawyer thinks.
So our original idea was to help the average person, so they understand what they’re signing, what it means, and what they should try to negotiate. Now we focus on the B2B market. Most of our clients are small and medium businesses, enterprise clients, and law firms. But our long-term vision is still the same. We still intend to make LegalSifter Review—that’s the name of our contract-review product—directly available to consumers. There’s a huge, underserved market around the world. People need fast, affordable legal advice. It’s an access-to-justice issue.
Castro: What are some of the advantages of using AI to automate, or partially automate, contract review?
Mahler: Speed, cost, and quality. When people use AI, they negotiate faster, cheaper, and with better results.
At LegalSifter, we call our AI algorithms “Sifters.” They’re trained to read documents and look for a specific concept, such as “governing law,” or “force majeure.” They’re faster than people, they’re more accurate, and they don’t get tired or careless or forget stuff.
When a user uploads a contract to LegalSifter Review, our Sifters flag provisions that need to be reviewed, or provisions that are missing. And they provide advice, with sample or recommended language. This advice can come from us—we have a network of experts who specialize in different areas of law and different kinds of transactions. Or users can supply their own advice, based on their playbooks. Or the advice can come from their outside counsel. This allows our users to get expert, in-context advice as they review their contracts. LegalSifter Review does this all in one or two minutes. Then, users edit the contract based on the advice and their understanding of the deal.
This review allows users to rely less on in-house or outside counsel. And many of our users are lawyers, who use LegalSifter Review so they can make better use of their time. Either way, LegalSifter allows you to save significant amounts of time and money and allows you to make better-informed decisions.
Castro: How do you ensure reviews meet your quality standards?
Mahler: To assess the quality of our Sifters, we measure a metric called F1. In effect, we’re measuring how accurate they are. On average, our Sifters achieve 95-97 percent F1. On average, a trained lawyer would achieve 85 percent F1. This is based on a study from 2018. This doesn’t mean that humans are bad at contract review. It just means that AI is very thorough. AI doesn’t lose focus, and AI can pick up on subtle nuances that a human might otherwise miss.
Once the Sifters find provisions and provide advice, it is up to users to read the advice and use it as they review and negotiate. We don’t have a way to measure whether they are negotiating well or poorly. But in general, our clients tell us that LegalSifter Review helps them negotiate deals faster and more consistently, with fewer deviations from their in-house playbooks.
Castro: What are some of the challenges in providing a tool like this in multiple languages?
Mahler: Using LegalSifter Review with multiple languages raises two issues. First, do Sifters only recognize concepts in English, or can they recognize the same concepts in other languages? And second, is the advice only in English, or is offered in other languages as well?
Currently, our algorithms work only in English. We’ve been asked to support other languages, and there are two ways we could do that. In the first approach, we’d rebuild from the ground up in another language. This would give us high accuracy, but it’s expensive. In the second approach, we could use machine translation to take our English-language Sifters (or datasets) and automatically map them over into another language. This approach would be much less expensive, but it would result in lower accuracy—and our users need high accuracy to be able to trust the AI. So currently, we have to trade off between accuracy and cost. Having said that, machine translation techniques are already strong, and they are getting better each year. So we might soon be able to use them to offer Sifters in multiple languages.
As for our advice, most of it is in English, but several clients are providing it in different languages, including Chinese and Croatian. This allows non-native English speakers to read English-language contracts, understand the implications of the text, and get negotiating advice in the language of their choice. So the advice layer can be written in any language. It is already provided in multiple languages, and we expect this to continue to more and more languages.
Castro: How do you expect technological advances around AI to change how law firms serve clients?
Mahler: AI allows law firms to offer their clients better and faster services more cost-effectively. We’ve seen law firms use LegalSifter Review in different ways.
First, they use it to check their work and make sure they didn’t miss anything during their own contract reviews.
They also use it to drive revenue. They resell LegalSifter Review to their clients, adding their own advice. Their clients can upload contracts and immediately get the law firm’s feedback. The law firm gets new revenue and the client gets cost-effective advice—on demand.
And recently, law firms are starting to outsource some of their high-volume, low-risk contract reviews to us. Clients send them contracts for review. The law firms send those contracts to us. Our review team uses LegalSifter Review to do an initial analysis, and we send the results back to the law firm. It’s a variation on standard techniques—outsourcing or offshoring.
So law firms are using AI in different ways to improve their margins and better serve their clients. I anticipate that this trend will continue and accelerate over the next few years.
