KLDiscovery’s Eric Mandel Talks AI in Information Governance and Legal Operations

In this interview, KLDiscovery’s Eric Mandel discusses the role of AI in Information Governance and Legal Operations. To hear more from Eric and other experts on this topic, register for this Today’s General Counsel panel discussion.

Information governance (IG) brings together a lot of stakeholders: IT, legal, risk and compliance, data security, etc. What are some of the challenges these teams face when implementing an IG program and how might AI help with that?

Eric Mandel: Information governance is a holistic approach bringing in all these different stakeholders (the IGRM Version 4 does a nice job of showing who all the stakeholders are) and they’re facing a massive volume of information, an incredible variety of information, and the velocity of change is shocking.

The core challenges are listening, working together, understanding, and making reasoned rational decisions about finding a holistic, optimized balance between those responsible for reducing and managing the risks of the enterprise, and those whose job it is to extract as much value as possible from the enterprise’s data.

How can AI help with these competing forces? At its core, AI can help identify what information they have. We live in this world of, “I don’t know what data I have, I don’t know where it is, and I don’t know who has access to it.” Governance is all about creating a unified approach and having AI provide insights to summarize and synthesize these vast volumes of information allowing stakeholders to take more reasoned, stepped, and logical approaches toward achieving the ultimate balance of information governance.

A key element of legal ops is data analytics, not just with eDiscovery and internal investigations, but also looking at the larger business functions related to that data. How can AI help make this task more streamlined and intuitive?

Eric Mandel: The legal field has been using machine learning focused on classification for a while. Three years ago, no one knew the term GenAI unless you were in data science. Now it’s everywhere and we are trying to achieve results in a different way. Instead of saying, “Sort all this stuff into buckets,” I’m saying, “I want to know about X, teach me. Give me materials, summarize this, synthesize that, organize the next thing, and give it to me in a way where it’s in a report, it’s in a function, or it’s something I can read so that it’s the equivalent of research output.” I want to start to use it for decision-making purposes.

Large Language Models (LLMs) require a vast amount of information for training. We are starting to see some GenAI tools focused on legal, but specialized tools will take time to train, and it’s not going to be cheap. To be on the leading edge, you have to invest in the technology and people who develop processes. This isn’t going to happen overnight, but I do think we’ll see more innovations coming.

We’ve been talking about exponential data growth for the better part of a decade, and how finding relevant ESI within these large data sets is no longer tenable for human review alone. How can AI level the playing field against large data volumes with an ongoing IG program operating whether or not there’s ongoing litigation and then move over to discovery or investigation should there be a triggering event?

Eric Mandel: I do think LLMs, particularly once they’re trained, are going to be a real game changer in terms of investigation and compliance functionality. Right now, most enterprises with a compliance program are doing random pulls of data and searching those random sets, and there are challenges with that. A random pull of data might lead an investigator to say, “I unrolled this one haystack and there were no needles in it.” But there are 5,000 haystacks out there.

With investigations, we’ll have a greater ability to locate specific types of information as we get better at training these models and with our prompts. We’ll be able to cut through these larger volumes more efficiently with less hassle.

But when it comes to producing documents, where’s the line going to be on a production-side review of materials with an obligation to find all non-duplicative information within the scope of discovery? With an investigation, you just need to have an understanding of what’s going on with the data. But when I have a production obligation under the federal rules or local state equivalent to producing everything that’s responsive to the discovery request, will I actually be able to use an LLM to sort this stuff out and put it into buckets when that’s not the function of the algorithm?

I understand the desire for an easy button, but I don’t think we’re going to get it on the production side, at least not anytime soon. I do think GenAI will be a help in organizing datasets, in getting some early understanding, in creating searches, and in training a classifier for active learning functions. There may be workflows where LLMs are used to generate information that can be used to train a classifier. But, for now, you’re going to have to manually move between the LLM and an active learning system.

One last question to finish things up. The mantra for in-house legal teams is “do more with less,” and everyone is talking about how AI will empower legal ops to do just that. What are some specific ways you’re seeing or anticipate seeing AI help?

Eric Mandel: I think once we start applying LLMs within an enterprise environment, they’re going to start helping us identify patterns that would be completely dark without doing some massive data science exercise. We’ll find cost savings, process efficiencies, and see which attorneys do better than others. We’ll find where our legal spend is going with eDiscovery, and we’ll have the data to back it up instead of just assuming. The ability to synthesize large volumes of information will be valuable and could drastically change the spend for legal, freeing up capital to be applied to other places within the enterprise.