Welcome back to the Big Law Business column on the changing legal marketplace written by me, Roy Strom. Today, we look at a legal tech startup that says contracts are not as complex as Wikipedia articles and is building a machine that learns how lawyers negotiate.
Lawyers do a lot of things, but a lot of times what they get paid for boils down to what they do with words. How can they convince a judge to focus on this word and not that other word? How can they eliminate words they don’t like from a contract?
There is something that feels distinctly human about fighting over language. But what if you’re a lawyer who just argues about the same words over and over again? Well, it turns out there is a company that thinks there are a lot of lawyers like that. And that a lot of what they do can be replicated by artificial intelligence.
Daniel Broderick explains what his startup, BlackBoiler, does in just a couple sentences. You send a contract in a Word document to an email address. The Word document comes back edited with the changes tracked a few minutes later.
“It feels like you’re working with someone on [your] team,” Broderick said. “We just do the work.”
One of the company’s clients is Jim Michalowicz , a veteran of law department efficiency efforts dating back to the DuPont Legal Model.
Michalowicz has most recently set his sights on reducing the time it takes for publicly traded electrical component manufacturer TE Connectivity to draft, review and sign contracts.
TE Connectivity’s roughly 180-employee legal department had made some progress on that front by creating a standard NDA template. That cut in half the time it took to turn around those often simple agreements. But that was over a year ago, and Michalowicz, whose title is senior manager for legal operations business performance, was still hungry for more speed. He thought AI might provide a kick-start.
“There wasn’t a whole lot of guidance available in the marketplace,” Michalowicz said. “There are a lot of AI providers, but their explanation about what they do is not necessarily clear.”
Michalowicz tried a pilot of BlackBoiler this summer, which led to a 66% reduction in turnaround time for NDAs. The company purchased the software and began a broader rollout in June, which so far has helped standardize its lawyers’ NDA review process, Michalowicz said. The product needs to see similar edits to function best.
BlackBoiler learns from a company’s history of editing documents. It asks clients to send in 200 or more sample agreements that have already been marked up. Its software is designed to understand the relationship between words and mimic those edits.
Broderick founded the company because he was sick of making similar mark-ups over and over as an associate at Thompson Hine and Kilpatrick Townsend & Stockton reviewing clients’ contracts. He estimates that Fortune Global 2000 and Fortune 1000 companies spend about $35 billion annually to review and negotiate contracts — most of that in salaries to lawyers. He thinks his product can automate at least $7 billion of that, which is spent on lawyers making verbatim, repeat changes to contracts.
“It’s just bad business to be paying people to do this repetitively,” he said.
BlackBoiler’s business is backed up by research into contract language that the company’s three co-founders conducted in pursuit of a $225,000 grant they received in 2017 from the National Science Foundation.
In its grant proposal, BlackBoiler says it intends to “render the tedious, time-consuming and expensive manual process of contract review and negotiation as archaic.”
The company’s research compared the complexity of language found in various types of contracts to the complexity found in Wikipedia articles and the Brown Corpus, a sampling of common English-language texts like newspaper articles.
The results found that contracts contain far fewer unique words than the other texts. For instance, in a sampling of Wikipedia articles one out of every 24 words appeared only once. For NDAs, there was one unique word for every 135. For contracts between a general contractor and a service provider (what the study called “prime contracts”) unique words were just a bit more common: 1/120.
The company’s research found that about 12% of sentences in a sample of 200 NDAs were identical. In a sample of 200 Wikipedia articles, about 1.6% of sentences were identical.
That research counters a common critique Broderick said he faced when attempting to apply natural language processing to automate the edits of contracts. Critics said contracts were “too complex” to be learned by a computer and edited in similar patterns.
“What we’re saying is you’re right, you can’t do this with newspaper language or Wikipedia [articles],” Broderick said. “But because what you have in contracts is recycled language, you can automate the editing of contracts. The language is so constrained.”
At TE Connectivity, the tool is first being deployed on NDAs. But Michalowicz said the company hopes to move up the value chain to smaller purchase orders, due diligence for onboarding a new business partner, or legal reviews of marketing materials.
The tool has so far delivered insight into how different lawyers on TE Connectivity’s team review NDAs, Michalowicz said. And it is helping lawyers standardize their approach to reviewing and drafting these agreements. Because the machine needs to learn on “clean data,” the team is having discussions about where in contracts they need to be more consistent and where they need more flexibility.
“Once we make those decisions based on that knowledge, then I think the machine will really start cranking,” he said.
In an effort to further speed up the pace of contract turnaround, Michalowicz said he could one day envision the AI tool being part of a customer-facing portal. A customer would submit a contract to the portal and BlackBoiler’s tool would take a first crack at applying its lawyers’ historical edits.
“We believe at least initially there would be some human review,” Michalowicz said. “But over time as confidence builds, some of the lower risk lower dollar contracts could be pretty instantaneous.”
Worth Your Time
On Partner Runs: And not a charity 5K. My colleague Meghan Tribe writes that when partners start racing for the exits it might already be too late to save a struggling firm. A Yale Law School professor blames the classic partnership structure.
On Nat Sec in Big Law: Big Law firms are increasingly hiring partners with national security backgrounds at the National Security Council, the Department of Homeland Security, the White House, the Treasury Department and other agencies, Elizabeth Olson writes for Bloomberg Law.
On Am Law 100 Collections: Thomson Reuters Peer Monitor Index came out with new first half data this week that showed Am Law 100 firms were the only segment of firms that are failing to grow revenue at a faster pace than they were a year ago. That is despite Am Law 100 firms raising rates faster than their Second Hundred and boutique competitors. One reason might be that Am Law 100 firms have the worst realization rates among all segments. They have trailed Second Hundred and boutique firms by at least 2.5% throughout 2019.
On Last Week: Declining Am Law 100 realization was the subject of last week’s column, where I wrote that Thomson Reuters’ Peer Monitor Index said demand for Am Law 100 firms grew 1.6% quarter-to-quarter. The figure was year-over-year. I also wrote that Thomson Reuters said firms collected 89.2% of the bills clients had agreed to pay. The figure for Am Law 100 firms was 80.4%. I regret the errors.
That’s it for this week! Thanks for reading and please send me your thoughts, critiques, and tips.