After decades of being labelled risk-averse technology laggards, corporate legal departments are emerging as enthusiastic early adopters of artificial intelligence.
Legal services — often information-dense, analytical and procedural — are fertile ground for the AI-powered automation of tasks including drafting contracts and providing initial legal opinions, experts say.
For in-house legal teams considering AI legal tools, an early and important question is whether to build tools in-house or buy them from a software supplier. What are the pros and cons of each approach?
The builders
Given the complexity and cost of designing, testing and rolling out proprietary software, many companies building their own legal AI tools are in the technology sector.
Hewlett Packard Enterprise (HPE), for instance, encourages its legal team to act as “citizen developers” by building AI applications themselves rather than relying solely on third-party products.
This initiative began in 2023 when HPE held an company-wide “hackathon” for staff to propose ideas for technology projects. The legal department won with its “proof of concept” for using generative AI to help monitor regulatory compliance, which led to the team developing tools exploiting the technology for various tasks, including contract analysis.
HPE trialled a third-party software supplier’s AI software to use for its legal department but was “underwhelmed” by its performance, says Jeff Fougere, HPE’s associate general counsel specialising in cyber security and IT.
There were also concerns about sharing sensitive information such as contracts and negotiating strategies with software suppliers. “What we decided at the time was, OK . . . this technology is still pretty new. We’re not super confident sharing all of that information . . . maybe we can build something ourselves,” he adds.
Now the tools are used by between 50 and 100 of HPE’s legal team worldwide, he estimates.
Building the tools internally has several benefits, Fougere says, including giving HPE more control over its data, therefore minimising the risk of leaking sensitive information.
It has allowed the company to revise and fix problems — such as the AI producing incorrect data or “hallucinations” — faster than if it were using a third-party product, says Fougere. Building the AI software has proved “significantly” cheaper than buying third-party tools, he says, which can cost “thousands of dollars per user, per year”.
Some legal departments start by building relatively basic tools and wait for them to prove their value before proceeding to develop more complex ones.
“AI initially is just dealing with the low complexity, high volume stuff,” says Andy Cooke, chief legal officer at TravelPerk, an online business travel management platform.
The Barcelona-headquartered company, which in January raised $200mn to fund its expansion, began building AI tools to answer simple legal queries in 2023. Now they are used for tasks including monitoring whether new features in the company’s platform could breach industry regulations and to extract legal information from company emails.
To build the software, Cooke says the team starts by replicating features of the commercially available third-party tools they like, connecting them together “like Lego blocks”.
The initial legal AI tools were relatively simple, the aim being to build “useful products that meet business needs”. More complex tools will be built in future iterations of the technology, he adds.
The buyers
For other corporate legal departments — especially ones with limited IT resources and smaller legal teams — building bespoke tools in a fast-changing technology, in which there is a growing third-party market, is hard to justify.
Stationary supply chain Staples Canada uses software from legal AI start-up Luminance for managing and negotiating several hundreds of thousands of contracts, says Adrian Lang, chief legal and privacy officer.
The software uses a traffic-light system to identify contract clauses in need of closer review — with red signalling high concern, yellow for matters to negotiate and green for clauses that are ready to approve. Automating part of the review process has saved between a couple of hours and a day or more per contract, she estimates.
Even if it were possible to justify the resources required to build such a tool internally, keeping up with the standards of legal software suppliers could prove challenging, says Lang, who leads a legal team of six.
“[Third party] technology quickly overtakes [in-house IT] and unless you’re actually committed to continually investing in that build, then it quickly becomes obsolete,” says Lang. “Or at least . . . certainly not nearly as cutting edge as other new products on the market.”
Spanish energy group Repsol initially built its own legal AI software for contract drafting and negotiation. The software was at first only designed for one type of contract — for exploration and production — in three countries. The in-house software worked “quite well”, according to Repsol general counsel Pablo Blanco Pérez. But the company switched to legal AI start-up Harvey after testing its software and finding that it was more versatile and cost-effective.
Harvey was “much better” at handling different contract types, he says. To reach the level of versatility that Harvey had, Blanco knew they would have to invest a lot more money and resources. “And we think that Harvey is going to invest a lot of money in improving the tool,” he says.
Despite the growing interest in legal AI software, many in-house legal teams are short of staff and in some cases lack even basic legal IT such as conventional document management systems, says Weston Wicks, an expert in legal and compliance technology at research company Gartner. That can make it difficult to commit to building their own AI tools, he says.
Legal departments with limited budgets can cut the cost of building their own tools by using “low code” and “no code” development platforms, he says. These allow users to build applications with little to no knowledge of coding by using intuitive drag-and-drop tools.
But in-house teams willing to spend significant sums on building advanced tools are still in the minority, says Wicks: “I don’t think [many] law departments are going to sink millions of dollars yet into their legal [AI] tech . . . building it from scratch.”