On Thursday, February 27th, LFJ hosted a virtual town hall on AI and legal technology. The panel discussion featured Erik Bomans (EB), CEO of Deminor Recovery Services, Stewart Ackerly (SA), Director at Statera Capital, David Harper (DH), co-founder and CEO of Legal Intelligence, and Patrick Ip (PI), co-founder of Theo AI. The panel was hosted by Ted Farrell, founder of Litigation Funding Advisers.

Below are some key takeaways from the discussion:

Everyone reads about AI every day and how it’s disrupting this industry, being used here and being used there. So what I wanted to ask you all to talk about what is the use case for AI, specific to the litigation finance business?

PI: There are a couple of core use cases on our end that we hear folks use it for. One is a complementary approach to underwriting. So initial gut take as to what are potentially the case killers. So should I actually invest time in human underwriting to look at this case?

The second use case is a last check. So before we’re actually going into fund, obviously cases are fluid. They’re ever-evolving. They’re changing. So between the first pass and the last check, has anything changed that would stop us from actually doing the funding? And then the third more novel approach that we’ve gotten a lot of feedback

There are 270,000 new lawsuits filed a day. Generally speaking, in order to understand if this lawsuit has any merit, you have to read through all the cases. It’s very time consuming to do. Directionally, as an application, as an AI application, We can comb through all those documents. We can read all those emails. We can look through social and digest public information to say, hey, these are the cases that actually are most relevant to your fund. Instead of looking through 50 or 100 of these, these are the top 10 most relevant ones. And we send those to clients on a weekly basis. Interesting.

I don’t want you to give up your proprietary special sauce, but how are you all trying to leverage these tools to aid you and deliver the kind of returns that LPs want to see?

SA: We can make the most effective use of AI or other technologies – whether it’s at the very top of the funnel and what’s coming into the funnel, or whether it’s deeper down into the funnel of a case that we like – is that we try to find a way to leverage AI to complement our underwriting. We think about it a lot on the origination side just making us more efficient, letting us be able to sift through a larger number of cases more quickly and as effectively as if we had bodies to look through them all, but also to help us just find more cases that may be a potential fit.

In terms of kind of the data sources that you rely on. I think a question we always think about, especially for kind of early stage cases is, is there enough data available? For example, if there’s just a complaint on file, is that going to give you enough for AI to give you a meaningful result?

I think most of the people on this call would tell you duration is in a lot of ways the biggest risk that funders take. So what specific pieces of these cases is AI helping you drill down into, and how are you harnessing the leverage you can access with these tools?

DH: We, 18 months ago or so, in the beginning of our journey on this use case in law, were asked by a very, very big and very well respected personal injury business in the UK to help them make sense of 37,000 client files that they’d settled with insurers on non-fault motor accident.

And we ran some modeling. We created some data scientist assets, which were AI assets. And their view was, if we had more resources, we would do more of the following things. But we’re limited by the amount of people we’ve got and the amount we get per file to spend on delivering that file. So we developed some AI assets to investigate the nearly 40,000 cases, what the insurers across different jurisdictions and different circumstances settled on.

And we, in partnership with them, improved their settlement value by 8%. The impact that had on their EBITDA, etc. That’s on a firm level, right? That’s on a user case where a firm is actually using AI to perform a science task on their data to give them better predictive analysis. Because lawyers were erring on the side of caution. they would go on a lowball offer because of the impact of getting that wrong if it went to court after settlement. So I think for us, our conversations with financiers and law firms, alignment is key, right? So a funder wants to protect their capital and time – the longer things take, the longer your capital’s out, the potential lower returns.

AI can offer a lot of solutions for very specific problems and can be very useful and can reduce the cost of analyzing these cases, but predictive outcome analysis requires a lot of data. And so the problem is, where do you get the data from and how good is the data? How unstructured or structured are the data sets?

I think getting access to the data is one issue. The other one is the quality of the data, of course, that you put into the machine. If you put bad data in a machine, you might get some correlations, but what’s the relevance, right? And that’s the problem that we are facing.

So many cases are settled, you don’t know the outcome. And that’s why you still need the human component. We need doctors to train computers to analyze medical images. We need lawyers and people with litigation experience who can tell a computer whether this is a good case, whether this is a good settlement or a bad settlement. And in the end, if you don’t know it because it’s confidential, someone has to make a call on that. I’m afraid that’s what we have to do, right? Even one litigation fund or several litigation funders are not going to have enough data with settlements on the same type of claim to build a predictive analytical model on it.

And so you need to get massive amounts of data where some human elements, some coding is still going to be required, manual coding. And I think that’s a process that we’re going to have to go through.

You can view the full panel discussion here.