Mike Gaudet on Investigations, Matter Triage and Explaining RAG

Thanks for having me. It's good to talk to you guys and always good to participate in NeLI events. The conference this year was a blast, so it was good to see you in person, Daniel.

Yeah, no, happy to. You know, as we were talking about it, Daniel, I'm from upstate New York, went to Canandaigua for high school, grew up a Syracuse fan. So by the way, Daniel, I know you're a Cuse guy. So big win over Tennessee in basketball the other day. Got to give a shout out to my orange there. We won't talk about football for a while. But yeah, I went to school at University of Buffalo, got my master's there, but I always knew I wanted to be in New York City. That was always just kind of my goal. Like, you know, coming up on high school, I was like, I got to get to the big city. And I got a job. I got really lucky and hooked up with IBM Consulting.

And so the job that moved me to New York was writing code. I used to write Java code and crystal reporting software, stuff like that on these consulting gigs. But I had fun because the consulting, you know, they were project based. So you do one and then the other. I kind of like that more than kind of sitting in a back room writing code. I thought it was more fun. And so that's the job. took me to New York. And I had that job for a little while. And then rent started to go up in New York. I actually moved there shortly after 9-11. And so the interesting thing in my journey was I was able to afford it. This is kind of strange now when you think back all those years.

But lower Manhattan had a grant program because people were a little worried and they were moving out. And so I actually got my rent a little subsidized by the city to be able to afford to move to Manhattan in the first place. So that's how far back my journey to New York went. And after a while, that grant ran out and my rent went up and I needed something else to do if I wanted to maintain my New York hood, as it were. I moved to Brooklyn for a little while, took on some roommates. That could be its own whole podcast. I took on these roommates. It was kind of a crazy situation. But I needed to find a way to get back to Manhattan.

So the long-winded answer is I needed the money, Daniel. And so I had a friend from University of Buffalo. His name is Stephen Olsen. He's been a bit of a mentor to me. We met at UB and he was working at Deloitte. And then he joined FTI Consulting, kind of in the early aughts. And he just wanted to pull in technical people to solve discovery related problems. Right. So he started reaching out to me around 05. And I said, I think everybody says this when they first hear the word eDiscovery. Everybody's like, I don't know what that is. And he goes, nobody does. We'll figure it out together. You know, come and join me. So it was around that time that he reached out and I decided to kind of take a leap.

And it was, you know, learning your world, Daniel and Brandon, the world of lawyers and how you all communicate and, you know, the types of things are important. That was kind of one of the things I had to pick up on. Right. Because there's the technology part, which I always kind of like solving problems. But then I had to kind of pick up on the, oh, you can't write that in an email. You know, you can't you can't assert completeness on this project. And I said, OK, you've got to be careful how you phrase these emails. So it's funny now when my kids ask me what I do, I go out, I write professional sounding emails to lawyers during the day. But, yeah, that's kind of how it started.

It started with my friend and mentor, Stephen, kind of pulling me in and taking a chance on me, you know, that I can apply my tech skills to this kind of emerging area. And I said, OK, I'm going to do this.

Yeah, I mean, you know, it's the I think the kind of two biggest trends. Right. And you guys probably hear this all the time when you guys are having these podcasts. I think, you know, one is AI. Right. And of course, you know, somebody had to it was going to come up sooner or later in the podcast. And the other is the you know, just the explosion of data and how much data there is out there. You know, I used to be remember back in the day, like when I started back in 05, like no projects weren't a terabyte, like a few gigabytes were or a lot. And I remember even the pricing that I think a gigabyte back then cost twenty five hundred dollars to process and process and thirty seven hundred dollars to analyze.

So over five thousand dollars a gigabyte was eDiscovery cost then. And now phones have a terabyte and phones as data sources alone. I think one of the things we're seeing is, you know, at first phones weren't even part of investigations. Right. And for several years of my career, no one would collect a phone. Then phones became sort of, oh, people say things on text and these other platforms. And now phones themselves might have 100 different data sources on each person's phone. We regularly do things that might become criminal investigations. So we take forensic copies of the phone. So there's your 20 chat apps on the phone. There's the map information. So the phones itself are large data sources. All of the traditional things, I feel like now, even in a Microsoft environment, Microsoft itself has 10 to 20 data sources, perhaps just in that one world, various links.

And we're not allowed to call them modern attachments anymore, right? But various links there. And the other thing, Brandon, that our group gets into a lot that is interesting, we do a lot of financial investigations. So. Like, in addition to our team doing the eDiscovery for the unstructured content, for looking at the communications of who knew what and when, we're also digging into the dollars and cents of, okay. So. If this person did commit the fraud, how much did they get away with? And so we're looking at back-end database systems, accounting systems. So if you think about all the structured data sources a company has, that's kind of what really takes us in lots of different directions. And of course, many of these projects span years.

You know, they'll say the fraud has been occurring since 1998. And we go, OK, what's your accounting system? And they go, well, from this period, we use this. And from this period, we use this and this period. And so you've got to go back historically and find all of the kind of past data systems. to collect from, too. So I do think that it's a big part of the conversation in these projects up front to talk data sources. And that's one thing, too, where I think the lawyers or outside counsel or inside counsel, for that matter, are really good at asking the questions and trying to figure out who are the key players, the interview things. They might not be aware of all the data sources.

Even ask about. So, you know, when we get involved early in the interview process, that's kind of some of the things where, you know, where do you keep data? And then now, in addition to all those complexities in the workplace, we have the whole, you know, kind of shadow IT working at home situation where, okay, when you go home, what do you do? Do you send it to your personal computer? Do you plug in any drives at home? Did you ever print anything out? And so all of that, I feel like adds up, especially if it's a large organization, an international organization. For anybody who's ever done a data mapping exercise, you know, for compliance, it's amazing. By the time you finish doing a data mapping within a company, it's like out of date because they acquired a new group that keeps their data somewhere else.

Yeah.

Yeah, no, when I throw that, you know, that piece of information out, people want to take comfort in the fact that AI is not going to replace humans anytime soon. We enjoy our humanity and the fact that the robots allow us to still exist on Earth for now, right? So it's funny that for now, we don't know when Skynet will be activated. But, you know, it's interesting to me that ChatGPT tried and failed to be a lawyer. So we're like, all right, elective sigh of relief. We're still the lawyers. And then a year later, it passed the bar at 90%. We're like, oh, this thing is learning pretty fast. And then, you know, since then, Daniel, ChatGPT, you know, GPT-5 has been able to pass more advanced scientific tests that have more kind of freeform application of learning and knowledge.

So not just kind of. repeating back stuff or recalling, but being able to problem solve at high levels. And so now it's basically a PhD scientist as well as of last year. So the level of that AI out there is acquiring degrees. You can perceive it as a bit scary, but there's nothing any of us can do to. stop it, right? There's no going back at this point. So I think it's probably famous by now and probably it might've even come up on your podcast, but I think maybe it was McKinsey that said, I can't remember. Somebody said, look, your job is not going to be replaced by AI, but your job may be replaced by someone who can use AI better than you can.

So I think the message is clear to anybody in the legal profession. Is, you know, you don't have to be, you know, the ultimate expert of the nuts and bolts on the technical side, but you do need to figure out how AI works. And there's been some interesting ABA papers that came out about responsibilities of lawyers, right? One of them, and I always bring this up too when I'm talking to like partners at law firms, is the duty to supervise still exists, right? It's, you can't just give it to, you know, the kid a couple of years out of college and say, you figure out the AI because I'm not learning any new tricks. Whatever output that individual has, you're ultimately responsible for. So it's, I think it's harnessing it.

I always think of AI and the way I kind of talk about it to my law firm clients is think of it as like bringing on. A really talented young lawyer who went to a good law school and passed the bar, but they don't know anything yet, right? They don't know your cases, your clients, you know, the history of, you know, maybe the industry that you guys practice law in. So it's a great kind of tool as a young associate, but you have to train it and give it the right materials, just like you would a young attorney that you hire. You have to show it the data and kind of point it in the right direction and nurture it. So that's the way I'm kind of seeing AI now is kind of, you know, a young lawyer that can write you a heck of a first draft on something or, you know, put together summaries.

And just like you would take that information from a young lawyer, you wouldn't necessarily go run to a judge without fact checking or looking it up. But you do need to know, just like you need to marshal a team of young attorneys, you need to marshal AI as part of that workforce.

Yeah, I think it's funny because the variety of cases is so broad, right? There's some, like I mentioned, kind of that might be criminal cases where we might need to do a full forensic copy of a phone and that might be appropriate where we need certain tools and to be on site. There's other things that we can do from iTunes backup if it's an iPhone. And so it can be cost effective and there's a lot of things that you can do remotely now, right? And there's providers, I'm sure you guys know. So I think the costs are coming down. We don't always need to travel. But in cases that we do, you know, all the kind of the old but good tricks are still important, right?

The chain of custody, documenting how you did it. I think… I think what we're doing a lot now is because the data volumes are so large, especially on cases where we're getting multiple phones. It doesn't make sense to just collect everything and dump it all into an eDiscovery platform, which could be quite expensive. Right. Yeah. So we've been working on a lot of off channel communications matters where we have to collect data from traders, from brokers, and it's actually investigating their personal phone communications. Right. So the SEC monitors their business line. And if they get approached by a client on their personal line and some of their clients are also friends. Right. So they'll reach out. We need to help companies regulate that because the SEC has fined billions of dollars in the last couple of years.

And so that's another nuance, Brandon, is when you're collecting a personal phone, a lot of times people rightly so are very concerned about their privacy and what we're kind of pulling off the phone and then reviewing downstream. So I think that just further underscores. You know, you know, to your question, kind of what are what are some tips or things that we're doing is that triage element of we're not just taking. somebody's phone and dumping it wholesale into a review tool, we're kind of carefully looking at what phone numbers are communicating, what sources, what date ranges, and trying to be as targeted as we can before we proceed with review. Mike, as I'm listening to your answer.

Yeah. And I think it's not only helpful to the client, Daniel, it's kind of my favorite thing about where I work. Right. It's I kind of. I like the technical stuff, but I like that I can work with talented people that are great at things that I know nothing about, right? And that we can all work together to solve a client problem. So a good example is we did a fraud investigation and J.S. Held has a number of really talented certified fraud examiners and forensic accountants. And they might be the first phone call and bring us in. We might be the first phone call and bring them in. But we did one where we had to collect the data and our forensic accountants had to go through millions of docss.

And this was a case where we didn't have an army of contract attorneys, right? We had a few people. So this was a good case for how can AI help these accountants find what they need? And then we brought in, as I touched on earlier, kind of the structured data components also, the accounting systems, the bank statements from this company. Let's dig through it. And it was great how we could feed each other. Back and forth throughout the investigation. Hey, we found this weird email from the CFO that indicates some interesting behavior. month, a couple of years ago. Let's look at the bank statements then, and then we'll pull it up in SQL and have those people go, oh yeah, there were some duplicate payments that we observed as well.

And so it's, it's kind of the synergy there. You know, we, you know, we've got a lot of different, I also say that the interesting thing is the industry focused professionals. So we have like health focused professionals here, construction focused. We have the, the largest group of IP valuation experts in the world. So we get pulled into IP cases, which as you guys know, is kind of a fun, juicy thing to sink your teeth into from a data perspective of was data exfiltrated, right? How much was that trade secret work? So I think it's good that we have those contacts to be pulled into different industries. And, you know, I always tell people within my company, look, we ought to be and we are the most collaborative group in the company because everybody has data, right?

And so we're kind of just the tactical arm that deals with that data. And, you know, and I feel like, too, there's a growing trend, Daniel, and you guys have probably seen this, too, where. I think gone are the days where clients, you know, want or will tolerate like armies of consultants being thrown at a problem. I think what they really want are, you know, a smart PM or a few smart, experienced PMs that can use the technology to steer things in the right direction. Right. And the tech in some ways, it hasn't replaced people, but it's taken a lot of the lower level tasks on so that somebody who's adept in the technology and in good project management practice can kind of steer that ship.

And so, you know, I like working on investigations that pull in all those different groups and where we get to kind of use some fun tech along the way.

Yeah, no, that's a great question, Brandon. I think, you know, a lot of people are focused on what are all the things Jen and I can do on the review side. And there's a lot going on there and it's worth talking about. But as we know, things can get quite expensive fast when you're making a lot of token calls to LLM engines and just racking up the bill. I am a huge advocate, and I think that this is low-hanging fruit for pretty much any case, in kind of a Q&A, a reg database triage where you can ask questions of the initial data. To me, that's low-hanging fruit. It's a powerful use of AI that doesn't run up the bill initially. So what you can do is, to your point, Brandon, in that article, you can grab a few key people.

We know these two, three people in the accounting department, they were important. There's 20. There's 20 to 30 people in the accounting department. Do we need all of their email for all time? Well, let's look at the two, three people we know first and see where that leads us. Maybe it's enough. Maybe they implicate one or two. And so we've taken more of a tactic where we preserve first, you know, but in the preservation step, not incurring eDiscovery hosting rates. Right. And then we'll use the kind of advanced searching in the gen AI. And it's funny because when I talk to partners at law firms, sometimes they'll roll their eyes when I said, like, we're going to use AI on this one because everybody has a preconceived notion of what that means.

So I've kind of started calling it advanced searches powered by AI. Right. And when you start demoing this, I think the one of the most powerful things about this technology is just the simple act of searching for the 20 years I've been doing this has been very geeky. Right. There's syntax, there's parentheses. Right. Right. You know, you know, and lawyers who are obviously very smart people, sometimes they shrug their shoulders and they go, I don't want to do this. You do this. Right. And one of the really empowering things is just being able to write natural language sentences into your database. So I had a call with a lawyer and I was like. what do you want to know and she was like what do you mean what do i want to know i'm like well we collected two million docss and put them in this database why'd we do it and she goes well i want to find out if you know the cfo knew this and this and i go well let's just ask it you know we type in the question and it spins and comes up with an answer and like the look on her face is like that's all i have to do um and in that particular case she goes well how much does this cost and i go there's no additional cost for each question this is built into the tool that we're using and then you could see just the light bulb go off on her head and from there.

she was off and running and the great thing is you don't need an army of tech people to use this you know you can really empower investigator you know fraud examiners lawyers and so by doing that, In that particular case, Brandon, we had been doing searches for like, I think, three, four weeks and not hitting on the things because keyword searches, as we know, are flawed. You don't know how these people are talking about the things you want to find. But the AI is able to uncover and it found, I think it returned docss that we hadn't seen within 30 seconds of using it. And 80% of those docss were net new that we hadn't found in keyword searches that were relevant. So I was kind of sold there.

And on every case now, I want to use it. And I just want to get on the phone with whoever knows the most about the case and just kind of play with it together and say, you know, and then you can rephrase the promptss. And I like the fact that you can take the cost element away for a little bit. And then, you know, I think the next step is, and you're going to hear the word hybrid a lot in the coming years, hybrid approaches of that and full on review where you go, okay. We didn't, we didn't load all of those 20 people that we preserve. We only needed three. And of those people, we really whittled it down to what the docss are that are important.

Now that we've kind of left a lot on the cutting room floor, now let's turn on maybe the more expensive elements of review and Gen AI classification and coding. And I think the trend that we're going to see is right now, a lot of those different steps are kind of managed by us or whoever the consultant may be on the case. I think the tech is going to be melding a lot of those together, right? Where there's going to be kind of a part one, reduce, then run, then test and validate. And it's all going to be built into a cycle as the technology gets more advanced.

Yeah. So a RAG is a retrieval-augmented generation.

But here's how I think of it, just to continue on the analogy of the junior associate, right? When the junior associate comes in, let's say she's trained in the law, right? So she's read the legal book. She's passed the bar. She's got a good background. on the legal world, but not necessarily your client or your case or the acronyms that your client uses, right? Because she couldn't have known that coming into the firm. So the rag is just kind of like documentation that's specific to your case, a review binder or something like. that, reference materials for how your firm works, how you build the clients. It's kind of the stuff specific to your firm that no one could pick up in law school because they wouldn't get that information there.

So in the same way, the retrieval augmented generated database built on your client's data is taking that smart LLM, that big brain open AI or Claude or Gemini that costs billions of dollars to make that no one else could make. It's taking that and now going, okay, now learn my case, read these hundred thousand docs and use that as a. reference for all the decisions you're going to make. So what that does is it grounds the client into a decision that they're going to make. So what that does is it grounds, the client into a decision that they're going to make. So what that does is it, And keeps it from, you know, the scariest word in AI hallucinations, right? We're all worried that AI is going to be making up things.

But the more it knows about the case itself, it's using the words that it finds in your client data, rather than just the stuff it found on the internet. So it no longer needs to fill in the gaps by making things up. It's learned with the reference material that you provided it.

Yeah, you've hired, you've found the smartest brain to bring into the room. Now, what does that person need to do their job? They need some reference materials. So that's why I think these RAG databases are great, because it's focused just on your client data. And then, you know, I think the emerging, another emerging thing that will happen, and we see this, I mentioned how we work with different industries like health, construction, IP. I think one of the things we're going to be doing is kind of saving from run to run, you know, okay, the next construction case looks a lot like the last construction case in terms of the documents that our construction experts need to look at. They need to look at delay schedules and all of those things.

It's going to remember the AI and some of these. Models are going to remember what was important in the last case and find the similar kinds of documents in your next case. So in other words, it's like that junior associate going through first, second year of her career. Now in year three, she sees a similar case and she's. right on all of the issues and it's not necessarily remembering all the confidential, information from the first couple. It's the concepts that are going to be the same. So I. think that's going to be a kind of a key element too in the utility of all this AI.

Yeah, well, I think it starts with me with, you know, Robin Perkins at Kutak Rock and the passion that she has for it is evident when you talk to her. And when you participate in the conference and you see the level of thought and planning that goes into it, you know, it really makes you a believer that this is this is a passionate project. And I think that's reflected in the quality of attendees and participants that NeLI has. I like that it's Midwest based where you guys are. Right. Like so many of these conferences are New York, D.C., East Coast based. I feel like a big part of the country gets overlooked. So I like connecting with colleagues at NeLI that I sometimes don't see.

And I like how there's a lot of thought that's put into weaving together these different concepts, kind of like what we've talked about here today. But I remember and you guys might recall in. In the last in-person conference that we had. In Kansas City, there was one kind of fake or sample case, right, that was woven into each step along the way of what do you do for a meet and confer? Then how do you approach the collection? Then how do you? And I had never seen anything like that in a conference that tried to weave all these concepts into a comprehensive story. And I thought the quality of the speakers did really well. And I'm, you know, I always like to have different perspectives.

And NeLI brings, you know, technologists like me, but lawyers, you know, computer scientists, judges, I think it's always helpful to have the judges, there was a great judge panel at the last NeLI. And so to me, I haven't seen a conference that better weaves together all of these important concepts through the lifecycle of a case. It's very educational. And I, you know, I met a lot of students at the last one, too. And I know they get a lot out of it. And the fact that it's kind of. not only for the in-person group, but, but kind of broadcast more widely. I think it reaches a lot of people and it's just, it's got a great message. And, and, and again, I think that all comes back to initially Robin's passion for it and how it's grown over the years.

And I know you guys have been involved too for a very long time and have helped drive that, but it's evident in the quality of the sessions and the concepts that are covered.

Thanks guys. Glad to be on.