Episode Transcript
[00:00:00] Speaker A: Welcome to the Relay, the legal show for personal injury law firm owners, presented by Lexamica. I'm your host, Gabriel Stiritz. Today I'm talking with Ethan Hilton, founder of Caseflood.ai, an AI powered intake system most recently from Y Combinator's winter 2025 batch. At 18 years old and yes, that makes me feel freaking old, Ethan represents a new generation of founders entering the legal tech space with different assumptions about how technology should and can work. So we're going to examine his approach to the intake problem, what his early results tell us about automation and legal services, and whether his model addresses the real challenges that are facing law firms today.
Just to take a step back, most law firms treat intake like it's 1995. We got call centers, baby. We got manual processes, inconsistent follow up and moderate data on conversion rates. It's okay. AI is providing a layer of insight that hasn't existed in the past, but still largely being run by people. And they're spending a lot of money on marketing campaigns. You know, because you're a personal injury law firm owner. Listening to this, the question here isn't whether technology can improve the process. We know that it can, will and is. The question is how soon and how long will it be until human call centers are replaced by AI? I've talked about this a lot with a lot of people. I think that Ethan may be at the very, very forefront of the, this piece of technology. And I do think that it's happening in 2025. And just to frame it out, the bar that we're talking about is the ability to sign 92 to 90% of MQL or marketed qualified leads on one call with an E sign document. That's the standard for personal injury. If your team's not doing that, they should be doing that. And the, a lot of the listeners on this call know that that's what the bar is, which is a higher bar than a lot of the other pieces of legal where you're trying to book to a meeting or just get an initial bit of interest done. The bar, I think is actually higher in personal injury than anywhere else. So, Ethan, welcome to the show. Really excited to talk to you.
[00:02:04] Speaker B: Thank you so much, Gabe. That was an incredible introduction.
I'm super excited to get this thing running.
[00:02:10] Speaker A: Absolutely. So I'd love to hear like this is the hot topic. I mean, I'm at masterminds, I'm at conferences, AI is the hot topic. And the hot topic in the hot topic is the voice, because this represents a large cost center for personal injury law firms. Anyone that's working at scale has an entire team of people. We've been able to reduce costs across the industry by offshoring, but not everyone wants to do that because there are trade offs associated with that. But like I said before, I think AI is going to replace people. It's going to get to parity. So I'd love to just hear from you as an expert, as someone who's building this every single day. Where are we as of the end of July, beginning of August 2025? The question is for you, Ethan, what we are. What is state of the art with AI? Voice. What's the state of the industry as of August of 2025?
[00:02:57] Speaker B: So the thing is, with Voice AI right now, it's moving so fast.
I like that you mentioned August 2025, because I think, you know, even December 2025, the answer will be very different.
There was a lot of projections even like a year ago that maybe this would be something that people would be seeing. A year, two years, three years, four years.
I heard some people say 10 years.
But we're at the point now where I think most firms have realized that this is coming and this is coming in the next few months, if not already came. And so, so you're, we're based out of here.
[00:03:32] Speaker A: Just to clarify, you're saying that the, the technology is actually getting better, faster than it was predicted last year?
[00:03:38] Speaker B: Constantly. Yeah. There's, there's a phrase here in Silicon Valley which is in AI, six months is a decade. And that's absolutely true. Like every single week there's new things coming out and there's advancements.
[00:03:48] Speaker A: So. Yeah, so where are we right now? What can Voice AI do? Like, how long can a conversation be? How natural is it? Tell me, give me the ins and outs.
[00:03:58] Speaker B: So like the history of where this actually started. And I'll kind of like go from like where the very first ever voice AIs were to like where we are now. So in the beginning, Voice AI was simply like one of the types of things was a preset script. And then they say like, you know, what's your name? And you say Ethan. And they say, you know, what's your last name? You go, Hilton. Spell that out. H I T. Nope, nope. L. Right. You have to. It was, it was this very, very upsetting process. And it's what people, it's what most people are used to. Right? It's what the banks have been using for decades. And then in recent years, you know, we've, we've been able to have sort of this, this voice, have a conversation with you and it actually is able to respond to things you're saying somewhat intelligently. Right. It's no longer like a super preset script if it's actually able to have a little bit dynamic in here. Right. We know this, the quality still hasn't been great for a while and so you still get a very robotic sounding voice and you get kind of low quality responses and it wouldn't be able to actually intuit too much, but we were able to get somebody to like qualify themselves and then maybe leave a message. And then like in, in maybe the past year AI has been really, really good, specifically like for banks, healthcare, et cetera. And actually having people navigate through trees that are sort of crossing treated out by the people that made the agent to actually like get to a certain place. Right. Like firms were able to actually slot in a voice AI to help them navigate to like get to reception, leave a message, navigate to their certain person, maybe even put in their information, say their existing client, and then get mounted to like their case manager automatically and even like get qualified and potentially book a consultation. And so we've been, we've been there for at least like the past year in the sense that like AI could, could reliably book a consultation with a person that was willing to talk to an AI.
The goal is definitely not to have somebody who is willing to speak to an AI, be able to do something, but rather give the experience that's better than a human and actually be convinced and persuaded to do something, ideally sign a retainer and add a very, very high percentage, just like 20% of the time.
And so what's been proven today? What is proven today is that AI is now better in a state of the art. When I, when I say things like I'm talking about state of the art AI is now better than the average human at persuading them to book a consultation or book a paid consultation. And that has been proven at least within our own lab. We've been able to see that booking a consultation is something that's pretty low commitment. It's not like deciding on an attorney. And so people can actually come in. And now with the state of AI, it's not like this preset list of questions. You're sort of like moving around with them, you're ducking and leaving. You're trying to figure out what is their situation, what are their pain points. And then if they came in and sometimes they'll say, I don't Want to commit to a consultation?
I'm just kind of getting information and AI is now able to say things that are a little sales, a little persuasive, like okay, what are you thinking about?
And they can respond and they can be like well I just want to know the difference between this and this. I'm just exploring my options right now. And the AI, without giving legal advice, can give just like information and then say does that answer your question? And they'll be like yeah it does. And they say, great, do you want to book a consultation? And you can get them to book a consultation.
And so we're there now and so that's exciting. And what I think is still remaining to be done is the ability for AI to actually sign retainers, um, at a, a super high rate.
[00:07:18] Speaker A: Just, just to recap, cause we had to cut a piece out here. You've got human in the loop now filling in like the most difficult piece, which is actually, you know, getting a case signed up, which in some level is still good, cuz you want those guardrails on it. What's the biggest challenge right now? Is it, is it getting kind of that natural conversational flow where hey, I might overlap with you a little bit and then I don't want an awkward pause. Is it just the range of crazy and kind of off the wall responses that people can give or questions they can ask, like what are the big challenges that you're, that you' seeing that are being solved for right now in voice? AI and that are like that are going to get solved through to get us to the next phase.
[00:07:54] Speaker B: So right now with state of the art models, like the average person cannot tell that they're speaking to an AI. So like that is no longer the issue. The ability for something to feel conversational enough to where the vast majority of like the American population doesn't realize that they're speaking to an AI? Like that has been solved actually with state of the art models. The tough part with actually like getting people to sign like a 90% plus conversion rate with purely AI is how much of a wrestle it actually is to get somebody to sign a retainer. You know, very rarely is it a matter of like you go through this sort of like 20 minute back and forth conversation understanding like what's the situation? Like can we help you? And then you know, you send the retainer and they're like okay, I love signing things. And they just kind of sign right there on the spot.
Sometimes it's this, it's this commitment they have to make right, like they're, they're, they're hiring an attorney that's going to, you know, hiring this is going to help them get to the situation. So that's kind of a big commitment.
And so usually we're seeing people, they'll need some wrestling in the sense they're like, well like I need to think about it or let me take a look at it and I'll make my decision or whatever. I'll call you guys back.
And intake teams that I guess aren't trained like, like salespeople. They're, they're not actually going ahead and saying like well hold on, like what are you going to think about? And kind of like helping them through the decision making process. Because most people don't need more time, they just need more information.
And so like they actually need to understand like every single clause of the, the retainer agreement. And it's like this is stuff that AI can do. There's a lot of back and forth because then like you know, as, as the, you know, the, the human or the AI would be explaining things, the caller would be interjecting asking like well wait, but you know, and, and so like sometimes they, they might need, even need to speak to like their wife or something like that. And so like a human can actually like loop and be like okay, let's, let's loop in the wife, let's have a three way call and I'll explain it to her. Right. Um, and so this type of stuff is something that will be solved for but at the like the time being what I know to be like best practices to make sure that every single retainer that like enters a call is closed is, is not something that AI can be reliably doing.
[00:09:53] Speaker A: I mean that's really interesting. Like I had no idea how much there was a last mile problem with this. I kind of assumed out of ignorance that the issue is more about being able to like tie an E sign provider into like an AI thing.
Not that you get these scenarios and questions that are actually like really complex. That's, that's really interesting. Like you're not saying. I, I thought you would say something very different than you did and it's surprising like, but I, I love it because I, you're obviously like actually trying to solve for the problem and not just what the idea might be.
[00:10:25] Speaker B: We've gotten quite deep into optimizing the.
[00:10:27] Speaker A: Process, which is great because eventually you'll figure it out. Do you view it as like waiting for technology with the voice AI to get to a certain place or is it really like there's a number of problems. Let's say there's a hundred different things that crop up. Like I need to do a call, I need to be able to answer questions about a specific retainer. Like do you have a list of here are all the specific things that we need to solve for? Or is it more of a generalized. The technology is just not there yet. That's what's going to be the breakthrough that gets you to that sign rate.
[00:10:58] Speaker B: I think it's a mixture of both. But honestly I think that the technology might already be there and it requires an actual team like ours to, to be the ones that go ahead and like account for every single nuance and every single edge case, compartmentalize the agent into the exact tiny little steps that's necessary to make sure that's persuasive and closes, you know, all the way down to like, you know, being able to adjust the retainer agreement for like super small things. Maybe it's just like either changing a language, right? Like maybe the language is, you know, they speak Spanish or Portuguese and you know, we're getting English and that's a problem.
Or maybe it's a matter of like they have a really complex or like a really, really high value case and they know what their case is worth and they're saying I'm not going to do this for 45%. Like I know I can do.
And, and so there's, there's some edge cases around, like, well, how do you handle that? Right?
And, and so I think that like we can actually get there. But there's a thousand, a million tiny little things that need to be tweaked, adjusted features need to be added that like it will only happen if there's like a research team like the guys like that we have or anybody else that's working on it, that they dedicate their lives, they dedicate their entire sort of human purpose into solving this one specific part.
[00:12:08] Speaker A: Yeah, well, I mean, I think it's a smart approach because if you look at any emergent technology, a lot of what Google developed over time with Google Maps or you look at Waymo now, what's happening, you have escalation centers, right? Technology fails, fail over to a human being who can make a human judgment. And in law, I think similar to driving driven cars, like the stakes are similarly high. One obviously would result in a bodily injury. But in law you have like huge legal stakes that are involved. And to some degree you can't have AI making a Decision about can I change the language on a retainer, can we negotiate the fee on the contract? Because we don't have AI negotiating demands yet between insurance companies and law firms. And so it would make like there are just things that AI can't be trusted to do yet. So that's actually, that's super interesting.
Do you see? Because like every law firm is going to have a different position on all these things. Do you see trying to standardize these decision making rubrics for everyone who comes to you and says, hey, like here's the way that this is going to have to be done? Or are you able to customize those parameters on a law firm by law firm basis and say like, here's how this law firm's policy exists and are there 10 different policies that they have to communicate? Or do you think that this will standardize into basically like a universal set of protocols?
[00:13:25] Speaker B: I think that the universal set of protocols probably won't happen, but I think that similar to how someone might adjust the risk tolerance in like a portfolio, a firm might do the same thing. Maybe some firms are just super, super hard set on uncertain things, don't negotiate, et cetera.
And so like that could be a setting in which like we sort of, they softly say like, you know, we, we want to keep the same and we'll, we'll make some leeway there with like a predetermined sort of like waterfall decision tree. But it's something that they, they adjust marginally. We have had firms though that have like told us specifically if these specific use cases or if these specific circumstances then this changed to the fee. Right. Um, and so like that's already happening in a, in a sense, but I think quite honestly most firms don't even know. It's kind of like however the attorney is feeling that day. Um, and so like I, I think that it's sort of our job to be the ones to like guide them a little bit on like what we actually see works best. Right. Cause everything is like an expected value function. Um, and, and so like if maybe there's like a 20% chance of like a, you know, 100x increase, it's a lot better than like a, you know, a 60 chance of like a X increase or whatever. Right.
[00:14:29] Speaker A: And I, I, no, I, I totally agree. And one of the things that I think. Well, two thoughts. One is I think you're in a really interesting position because you're going to be able to start, get to a place where you're like, if you, if you flex on this parameter, your sign rate's going to go from 92 to 95% or from 95 to 97. Eventually you'll be able to say like, here's the ideal set of decision parameters that if you go toward these, you're optimizing for signups or you can optimize for the percentage fee rate, right? But like you'll be able to know which, which is which because you're tracking it at that level, which is really, really interesting and creates, I think, really powerful decision making tools. One of the things that I generally think is hard for a lot of attorneys to see and understand because it's maybe less intuitive, is that kind of risk adjusted mindset because they've been trained that like everything is a scenario that should be treated individually. It depends. It's sometimes hard to present them with. Taking a risk adjusted approach to a large number of individual claims, eventually, like it will have an outcome that is predictable, like everything is not a unique scenario. And so I think like, looking like risk adjusted thinking can be really difficult for attorneys. I've met some that can do it and I met others where even when I present them, you know, one of the cla, some, you know, some of the classics are like, you know, a referral fee. It's like, well, if you want to have a referral fee that is 50% across the board, you have to understand that that's going to eliminate your ability to refer a lot of cases out successfully, which basic economics would say you're going to make less money when the price is too high than if the price is set by the market. And to some degree that's what you're talking about. Like, so like this kind of like economics, principled think thinking is, you know, sometimes challenging and I think it's really powerful. But at the same time, like you're, you've probably already seen this. Like some people get it and some people were like, no, you're crazy. And you're like, well no, it's economics, but so it can be a challenge.
[00:16:25] Speaker B: That was some of the things that I was, I was, I was noticing in recent times we sort of default to the gold standards or the things that we've learned through just in relation, we're seeing, you know, as you can imagine, quite a few calls and, and the less technical firms that kind of just trust, like, hey, just do your thing, that tends to perform a lot better than the really picky ones. You know, occasionally you'll get the picky firm that wants every single thing that they already do to be perfectly replicated. But like, and, and it does well. But oftentimes like those types of firms, they actually end up putting in their own practices that are already flawed.
And, and so I think like, there's this really interesting opportunity that we get to have, which is we get to see so many iterations and so much volume of the exact same thing basically over and over that like, we could probably come out in a few months and definitively say the best practices for literally every single segment of this entire, like this. Is this the same conversation over and over basically, which is, which is really interesting.
[00:17:19] Speaker A: And then you'll be able to optimize.
[00:17:20] Speaker B: It at a level that.
[00:17:22] Speaker A: Yeah, oh yeah, for sure. And, and you'll be able to optimize in a way that small individual firms just won't be able to. In part, even if they were, were large, you're going to have more data, you're going to have the ability to process that data better and talk about what those standards are. And, and some law firms, like you say they'll never be able to overcome the biases that exist or potentially like what they view as ethical policies, which is more of a principled approach. I think the more, the problems that I see a lot more are, you know, this is the way that we've decided to do it. It's like, okay, that's fine, but once you have better data, you should change your position to the better one and not just say, well, this is how it's going to be done. So I think that's a big thing that a lot of law firms on the cutting edge, the bigger ones, like they're, they're coming to terms with this and saying like, yes, like, tell me how this is done. But that can be a real struggle for an industry that's been legacy for a very long time. But Ethan, we could talk about this for ages. It's really interesting. I've learned a lot even in the 20 minutes that we've talked. Caseflood.ai, you are at the forefront of voice AI for intake and also for some case management. Really exciting what you've built. Really impressed and, and not just a little bit jealous that you're the third youngest co founder in Y Combinator history, but really cool product that you built here. People who are interested in learning more or getting involved, how should they reach out to you?
[00:18:45] Speaker B: Go to the website, book a demo and we'd be happy to show them a tattoo.
[00:18:50] Speaker A: Fantastic. Ethan, thanks so much for being on the show.
[00:18:53] Speaker B: Thank you so much. Gabe.
