Howard Kennedy has selected and begun to roll out GenAI document review and marketplace builder Jylo, with adoption at the UK top 100 law firm already said to be high. The selection came after what director of IT and change Tony McKenna told Legal IT Insider was an extensive review and proof of value, where the firm looked at opportunities and use cases across various practice groups.
Jylo has multiple potential use cases but what is fairly unique is that practice groups can leverage the most appropriate large language models for the specific use case, and are able to create bespoke playbooks to automate tasks such as identifying key clauses, extracting information, and flagging potential risks. Jylo also provides secure document and team centric chat.
Speaking to Legal IT Insider at the end of 2024 about the selection, McKenna said that Howard Kennedy started to pilot Jylo a couple of months back, although the firm has been speaking with Jylo founder/CEO Shawn Curran and his team for longer than that. Jylo was spun out of UK law firm Travers Smith in May 2024 and McKenna said: “We know the track record of the guys who run Jylo and we started to talk to Shawn and the team pretty early on. The critical difference is that they are looking at how to leverage the broader investment in large language models, not just a specific large language model.
“As we’ve seen in 2024, other major providers have started to develop or enhance their LLM capabilities and when you look at Microsoft, with its Phi-3 small language models, the developments through 2024 have been significant. Tying yourself into products that tie you to certain language models seems commercially to present a challenge when we’re still trying to work out the use cases and what value they give you.”
McKenna said that adoption has been easy, with users able to engage with a document in natural language and receive similarly clear answers, all within a firm’s own tenancy, so no data is being exchanged with the outside world.
“One of the other advantages of Jylo is templafied prompting,” he said. “So, within the solution you can have someone who’s very naturally curious, understands the law, and understands their specific requirements who can then templafy a series of prompts or questions. They can create that template for real estate or litigation or construction, for example, then allow the rest of their team to use and access that functionality, that capability, without having to understand prompt engineering or how to get the best value out of a large language model.” In doing so, users are able to create a ‘marketplace’ that has significant benefits in terms of educating people on ways to work and encouraging more junior lawyers to work in a specific way.
In terms of selecting the right model for the right type of work, McKenna said: “Rather than doing straightforward volume work with an expensive large language model, you want to be doing that with a lower cost language model because it’s high volume, low complexity. But where you have high complexity, you might want to use a different and possibly more expensive language model that’s more trained and that will give you more detailed results.
“There’s also a lot of work being done at Stanford University around large language models having personas and responding differently depending on the type of question that it’s been asked. So for example, the view was that if you want a litigation type question answered in a litigator’s way, you would choose large language model X. Having the ability to work with different language models based on the department and the type of work you’re doing makes good commercial sense.”
Howard Kennedy was given access to Microsoft Copilot in the Spring of 2024 but McKenna observes that with Copilot and other LLMs, the whole premise is around developing a use case. “It’s a bit like ‘here’s a technology, find a problem to solve.’ The problem tends to be person-specific, not necessarily the same for a whole department or the firm,” he says.
Jylo, meanwhile, goes a long way in helping firms to productise or operationalise their knowledge.
McKenna expects to deploy Jylo across Howard Kennedy’s legal teams and potentially make it available to anyone involved in reviewing contracts or large document sets, such as risk teams. He observed that Jylo’s commercial model enables organisations to enter at a level they are comfortable with and expand from there, as opposed to insisting on an enterprise license, which is attractive. Jylo also provides transparency over token charges in order to help maintain control of costs.
To watch a demo of Jylo with Curran, click here: Jylo – Product Overview