How Evan Schnidman Went From Professor to AI Founder to Leading Fidelity Labs (Part 1)

Rob Sarnie: Foreign, um, Sarny. And you're listening to what the heck? Fintech, where you'll experience and learn that fintech is fin life. Let's get started. Welcome everybody. We have quite a guest today. We have Evan Shinman today, and he might be the most interesting person in fintech. The things he's done, what he's done, how he's done it, and, and uh, he's, he's got some unbelievable stories they can be able to share with us. He's been in academia, he's done startups, he's done, um, he's at, he's the head of Fidelity Labs right now. He's always helping the fintech ecosystem. And a little interesting story here. When I first met Evan. Evan, I think it was about, I think a year ago now. And I looked him up and, and, and I, I had like 50 connections already with him. Like, wow, I go, that's crazy. He must be at, uh, Fidelity. And he was like, okay. But only like one or two of those connections. Fidelity. He's been so deep in the fintech ecosystem, we've been probably walking by each other in events and didn't even know it. So very cool guy. And what he's doing and what he's done and how he's done is unbelievable. So, so, Evan, I won't do it justice if you could do a little player profile of, um, how you even got to this area. Because you've been, you had like the opposite. Like, I was an industry guy and then went academia. You were academia, founder, startup. And then, and then look at you now. Head of Fidelity Labs. I'll let you tell the story. How'd you get to this place? Because that's always the best part. Everyone comes from different directions. And you're the most interesting man in fintech, so I can't wait to hear this story.

Evan Schnidman: Well, Rob, first of all, thank you for having me on. Um, and I guess probably first and foremost, uh, I think I probably should dispel that I definitely am not the most interesting man in fintech. But, uh, I'll try to live up to that. For the purposes of this episode anyway. Um, uh, I guess quick, um, background on me. Um, as you alluded to, I started my career as an academic. Uh, I'm actually a game theorist by training. So, um, I was modeling small group decision dynamics and during the financial crisis decided the most interesting small group of decision makers in the world is the Federal Reserve. So I set out to model how the Fed makes decisions and how those decisions affect financial markets. Um, within, I don't know, probably a few months, I figured out that super interesting area I could spend the rest of my career just focusing in on. And there are approximately two dozen other people in the world not really cared about it. And I uh, had this crazy notion that as an academic my research should have an impact and um, that a wider swath of people should actually read what I'm producing. Um, so I took a step back and looked at the world and said, all right, if only a few dozen people actually care about the internal decision dynamics of the most important central bank in the world, the most important uh, economic policymaking body in the world, right. Maybe I should take a step back and just think about what are markets paying attention to, what are um, really the top investors in the world looking at. And the answer was not decision dynamics. It wasn't even policy, it was actually communication. That's what moves the markets. Um, I totally uh, changed my research agenda. Um, I decided I was going to start analyzing so called Fed speak. And um, what I realized pretty quickly was that there was no standardized methodology by which people were doing that. Um, and so it was totally subjective. It was, it's, it feels like this or it seems like that or they're implying this. Um, and so the burgeoning field of natural language processing appealed to me because I could actually use my math background to apply a much more quantitative understanding to uh, um, what was going on at the Fed and central banks more broadly. And so I ended up developing a novel methodology to mathematically map linguistic patterns and tie them directly to price action. This was early days of NLP. So uh, this is kind of 2010, 1112 developing this methodology, um, and looked around at the world and said hey, this is something that I might be able to commercialize. And so um, I ultimately I finished my PhD at Harvard. I was faculty at Brown for a bit. But I left academia to commercialize that exact technology, um, the ability to analyze central bank communications. And so uh, I um, walked out of academia, uh, as a truly terrible entrepreneur. I was uh, totally an if I build it, they will come guy. Um, I had done no market sizing, I did no work to figure out what is likely pricing of my product. I done very little demand testing, um, but I knew I had a really novel product idea. Um, and frankly I got lucky. Um, I got lucky that there were a few quantitative macro hedge funds that really liked what we were doing. Um, it was enough to get the flywheel in motion when I first started the company. And so prattle Took off uh, really in 2014, um, with a little bit of investor backing and a few quant, uh, hedge fund clients or say quant macro hedge fund clients. And um, that led me down this whole journey of really going into the entrepreneurial ecosystem, the fintech ecosystem, and more broadly into the data and AI worlds. Um, and so built that business through venture and strategic backing uh, for five plus years, sold it to Liquid Net in 2019 and then spent uh, my year there, um, really helping build out an investment analytics division and left uh, liquid in 2020 kind of peak pandemic, um, and looked around at the world and said you know, I'd love to do some fractional executive work, um, but working with the big companies, while lucrative, was not as gratifying as I had hoped. And um, and I found there were so few exits in asset management oriented fintech. I had a lot of companies coming to me and asking for advice, asking for help. And I love working with early and growth stage companies. I love frankly teaching them to avoid all the stupid mistakes I made along the way. Um, uh, and so I ended up uh, launching a fractional executive consulting business called uh, Outrigger Group. And so we built a whole business really uh, slotting fractional executives into Fintech, data and AI companies uh, for the last five plus years prior to joining Fidelity in 2025. And so um, it set me up wonderfully to step into managing Fidelity Labs where we've got a portfolio of early and growth stage companies. And so working with those entrepreneurs is very similar to what I've been doing for the last half decade. Wow.

Rob Sarnie: Now you must be in your heyday with LLMs and everything. I mean you were thinking about this. I mean obviously machine learning has been around for 50 plus years, but the LLM stuff and the uh, NLP that you're so gifted in, you must have been like, here I go, look at me, I'm waiting, wait for this big day. What was going through your head when I was on this big push for all that?

Evan Schnidman: Yeah, I mean it's sort of an interesting question because if you look at the evolution of AI from your point, it's actually been over 70 years now, right? Um, you talk about AI really emerging in the 1950s and um, it's had, depending on how you count how many boom and bust cycles. Um, but um, but I think you know where, where I saw uh, I think a pretty obvious evolution was I saw LLMs coming kind of 2018, 2019, um, the, the now famous Google Transformers paper, um, saw that paper come out and realized we were actually already using some of that technology within, within weeks of that paper becoming, um, publicly available. Um, we had actually deployed some of that technology within Prattle to solve a named entity recognition problem. So we were building what was effectively an LLM to solve this NER problem. And we were just racking up bills on, uh, uh, really just because we needed so much compute to run it. And that was part of the realization I had when we, when we decided to sell the company, was we were on a product treadmill that we were going to have to eat a lot of dilution, uh, to continue growing the business, to keep pace with the, with the evolution of AI and the technology change and presented a lot of execution risk for that business. And so, uh, I had to make the call of, do we keep raising and eating dilution and, and, and hope that the market comes to us over a period of a couple of years, or do we take the bird in hand and we've already gotten several acquisition offers and, and do we, do we go down that path? And so I chose the latter path. I have lots of questions in my own mind about whether that was the. But, uh, I also had projected that it was going to be, uh, just a couple of years before ubiquitous and would be able to kind of at least supplement, if not supplant what we had built, um, and be able to democratize the use of AI. And so I was worried about our clients being able to have something that's 80 or 90% as good as the signals we were selling. And, um, we were either going to have to keep pace from a compute standpoint and, and therefore eat dilution, or we were going to have to, uh, uh, you know, really pivot our entire product stack into something new. And so the, uh, the ultimate answer was we sold the company. Um, I was right directionally. I was wrong about the timeline. It took a couple of years longer than I expected. Um, and I'll be completely honest, I never thought a chat bot was going to be the answer, that that was what was going to get mass uptake on this wave of AI. Um, but that's certainly what we saw in 2023, and we've been off to the races ever since.

Rob Sarnie: Yeah, yeah. So I'm going to take a little step back in your story too. Now. Now you came from academia. I mean, because I was a transition for me coming from, you know, industry, uh, and, you know, the pace at Fidelity then going academia. And I was like, come on, let's go, let's Go, let's go, let's go. Now you went the opposite. You went to the startups and now that's even faster. What was that transition like? Just personally, how did. Because I went through a transition going the other way and tried to drag people with me. What was it like? Was the speed or like you already built that way from day one and you just, you know, even though you. And you adapted?

Evan Schnidman: Well, Yeah, I mean, maybe I was always an odd fit in academia because I was such a, uh, um, kind of hard charging person. I don't take days off, I don't sleep a lot. I'm constantly looking at what's next. Um, that's always been true. Didn't matter whether I was graduate student, faculty, or running a startup. It didn't really matter. Um, and frankly, even when I was an academic, I was doing consulting work on the side the whole time. Um, so I had always had one foot in industry. Um, really in part just to pay the bills, but in part to keep my feet grounded in what was reality. Right. Um, it's pretty easy to get stuck in the theoretical when you're an academic and it's much harder to see how that translates to, uh, the day in, day out adoption of the technology you're talking about. Day in, day out, actual utilization of the theory that you're referencing. Um, and so I'd always kept my, myself kind of straddling both worlds. Um, so maybe it was less of a rough transition for me than some, um. I would also say that just innately, um, I'm, you know, I'm a person who's most comfortable when I'm going 1,000 miles an hour. Uh, I don't, I don't really do slowing down well. So, uh, maybe the startup world was just meant for me.

Rob Sarnie: That's right. Right now you're fidelity. And that's definitely meant for you for what you just said too. Right. So let's go, let's make it happen. Right? Because. Yeah. Yeah. Fidelity moves fast. When, when, when it, it moves, it moves great. I love that.

Evan Schnidman: Well, I would say not only does fidelity move fast, I would say Fidelity Labs moves exceptionally fast. As I, you know, as I often say to my team, it's our job to move fast and break nothing. Right. Um, it's fidelity. We don't get to break things. We, you know, we manage people's retirement accounts, so we're very, very careful. Um, but the flip side is like we're, we're an innovation division. We better be moving faster than the rest of the organization. That's the point. And so uh, um, the beauty of what I get to do day in and day out is that I get to use the fact that my innate bias is to move quickly and I get to now impose that uh, on others, uh, who then uh, I get the great joy of recruiting like minded people who get to move as quickly as I do.

Rob Sarnie: Yeah, yeah. I mean one thing that um, I bring when I teach and I don't teach PhD students but grad and undergrad students, I still do them and I junk but um, for fin, and always fintech. Right. Um, so but the thing I always say about fintech and you hit on it with your Fed speak is that you know, there's so many technical terms, everyone's like what are you talking about? Then you get all the finance terms. What are you talking about? Especially when you turn on the news, 99% of the, of the country doesn't, probably doesn't even know what they're talking about and why they're talking about it and what do they really mean. And you, but you went after that and I think that's one of the hardest. So that was like very, very cool Fintech, you know, to relate that language to. What does it mean? What's the impact of the market? What's that ripple effect? Right. And then having that data, you know, get all reamed into it and figure out that that's such a, that's a huge challenge. And, and I love what you talked about on your research with Fed Speak. How is that that's still so important today. Right? Because it's not like it's, it's reading to read those tea leaves and you did it. And now is Liquid Net still doing it. Where is that in the stage now? Because that's huge. That's huge.

Evan Schnidman: Yeah. Uh, I should probably back up. What we did was we actually built a methodology to analyze central bank communications uh, around the world. So we covered the top 20 central banks around the world, um, top 20 based on currency trading. Um, and um, and so obviously that meant not only the Federal Reserve, but European Central bank, bank of England, bank of Japan, Canada, et cetera. Um, what we, what that meant was actually building a unique model for each central bank. And then we would actually residualize down to the speaker level so you could build a baseline for each individual speaker. And the beauty of that is that you start to identify patterns really quickly. Um, one of my favorite ones is that um, Janet Yellen, uh, had been president of the San Francisco Fed, Vice Chair of the Federal Reserve Board, Chair of the Federal Reserve Board, and then subsequently was Treasury Secretary. Across all four of those roles, she actually exhibited completely different speech patterns and in fact was more like her predecessor in that role than like herself in her prior role. In terms of her speech, her linguistic patterns. What that really means is you can see that role matters more than individual. And some of that they have speech writers and things like that, but some of it's actually literally the subject matter. They're talking about changes. Right? Um, you know, regional Federal Reserve bank presidents have to talk about their region and they frankly talk less typically about regulatory issues than board members do. Um, whereas, you know, when you're, when you're sitting as the, as the vice chair might be in a supervisory position or something like that, you might talk more about regulatory issues and less about macro or less about regional issues. Um, and so we were able to develop these unique baselines for each speaker and identify deviations from baseline very accurately. So the beauty of that system was not that we had actually developed a system to be all knowing about so called Fed speak, It was actually that we built a system where we could identify trend and uh, were they becoming more hawkish or more dovish? So we can very accurately identify what was the likely next policy move and when was there the possibility of surprise relative to what the markets were expecting. And so, um, doing that for the Federal Reserve System was tough enough because their language is so closely watched. Um, doing that for kind of secondary central banks that are less closely watched posed a completely different challenge, which is that the markets kind of mispriced what um, what they were likely to be doing. And we uh, were able to identify some really unique trends, um, also identify some, some challenges like when there was an attempted coup in Turkey, um, that presented some really challenging data because, um, completely changed how the Turkish central bank operated. Um, and then we ultimately applied that core technology to analyzing much more, a much broader swath of, of market moving language which is m. Corporate communication. So earnings calls, regulatory filings, speeches by corporate officers, press releases, et cetera. Um, so as far as I know, we were the first ones to use AI to analyze earnings calls. And um, you know, we launched that product in 2017 and within a few months the market was rife with competition, including several major investment banks. Launched computing product. Um, turns out we've kind of struck gold on that. And uh, um, to my comment before, uh, about being a terrible entrepreneur, when I started, um, I was totally, you know, if I build it, they will come purely product oriented. And then it took a couple of years before I really talked to my clients, listened to the market, understood where the demand was, and said, oh, there's a much more profound problem over here, which is keeping track of thousands of companies and their communications, not a couple dozen central banks. Right. Um, and so the market for analyzing corporate communications is so much larger and so much more diverse. Um, and frankly, was a much better business to be in. Um, and so, um, it both taught me a lesson as an entrepreneur about really sizing your market and understanding where the customer demand is. It also taught me a lot about product development and how to be sort of discreetly focused on the thing that will actually sell, not, um, the thing that you're fascinated by. Right. It's pretty easy to fall in love with your own solution. Um, it's much more important to fall in love with your customer's problem. And so that's where I've looked around and realized my own shortcomings and learned a lot in that process. But then I turned around and also realized that, uh, our earnings call product in particular, but kind of the broader corporate communications product, um, spawned a whole nest of competitors. Because when you do hit it right, the market really does identify that very quickly. And then you finally find yourself in a competitive situation which maybe you hadn't planned for either. So, um, that was kind of the evolution there. And then to your question about sort of where is it today? Um, a lot of that technology, uh, still sitting within Liquid Net, which is owned by TPI Cap. Um, they retained the ownership of that after we were acquired. Um, tpi, uh, Cap acquired Liquid Net shortly after I left. And so, um, they've gone off and done their own things with it. To my knowledge, it's not a standalone product anymore. Um, I think there's some interesting ways you can use current AI tools to get very close to what we did there. Um, but it wouldn't be a direct replication.

Rob Sarnie: Now, do you think your tool will be able to handle my hard Boston accent? I don't know. I don't know what this guy's talking about.

Evan Schnidman: You know what, Rob? Actually, this is actually one of my favorite problems. Uh, um, so we would analyze the transcripts from the communications, and the most common question we would get was, what about tone of voice? And I used to jokingly refer to this as the Italian CEO versus German CEO question. Uh, you get a CEO who's Italian or a CEO who's German. Like, they might be reading the same words off the page, but their inflection is going to be wildly different for cultural reasons. And so to your point about your Boston accent, right, you're going to have certain language you just use slightly differently. And, um, back when I was running prattle, we didn't have an ability to build up the baseline for that person's language and efficiently enough to understand what the deviations were. So we actually couldn't analyze tone of voice very efficiently. And so we just never built it into the product. That's now a solved problem. That's like, AI has evolved to the point where you can actually analyze tone of voice very effectively. And, um, in fact, you can even layer on facial microexpression analysis. Uh, every, every human being's, uh, face has the same, roughly the same musculature. So we all have the same seven core facial, facial microexpressions. So you can identify markers of deception, joy, fear, et cetera. Um, and so the end result is you can actually get a pretty good composite view just from AI based on the language they're using, their tone of voice and inflection and their facial expressions. So you give me a videotape of somebody, uh, and you get a pretty good shot of knowing, one, are they being truthful? Two, what's the sort of implications of, of what they're saying?

Rob Sarnie: Hm. Wow. Yeah, I mean, like, if I said, uh, hey, that's a, that's a wicked good stock and I'll meet you at the Paki after, like, would it pick up that. Okay,

Evan Schnidman: well, maybe.

Rob Sarnie: Let's go, let's go.

Evan Schnidman: 70% chance if you don't want to straight down. Well, I think we probably get there though, as, uh, as a lifelong New Englander, I can say, uh, you know, nothing lost in translation here, but on the AI system,

Rob Sarnie: wow. So let's pivot a little bit. I'd love to hear, because you meet so many startups, you advise so many startups, you're on board that whole thing. Uh, what, uh, in today's world right now, right, because everything's so converging, right? Like fintech's a perfect example, right? You have to understand the business, you have to understand the tech, you have to understand the finance. What are you seeing, especially in the fintech space for startups right now? Where are they falling down, but where are they excelling too? I'd love to hear that take because you see so many.

Evan Schnidman: Yeah, I think, I mean, let's go back to basics here, right? What is it that startups are really good at? And the answer, frankly, has always been product and engineering Right, right. Like large incumbents excel at distribution. That's how this works. Right. They've built out the Rails typically on last gen technology, what the next gen technology comes in. It's very hard for you to rip and replace if you're operating at scale, especially if you're in a regulated industry. Um, financial services. Right. So um, in his life, if you think about what, what's been working for a generation or more, I mean you can go back to the kind of dawn of the Internet 30 years ago. Right. Um, and realistically product and engineering have always been where startups kind of win. Um, and now that's upended. We're in an environment where AI has made product and engineering easier than ever. And yeah, customer discovery is still challenging and there's a lot of turmoil in the market about new things coming out all the time. Um, but I think that you know, we're still, we're still in a, in a pattern right now where startups are adopting AI faster than large enterprise and therefore they're iterating faster. And some say, you know, some will say, oh well that means, you know, the companies that are adopting AI are going to run circles around everybody else because they're moving so much quicker, etc. Etc. I don't buy that one because it negates m, the inherent advantage, uh, of large enterprise, which is distribution. And distribution is the hard part. I've been a part of, in one way or another, I've been a part of more than 70 businesses in the last decade, um, almost all of which in FinTech, data and AI. And in every situation they ultimately get it right with product and engineering and in most of those situations they struggle with distribution. And so AI hasn't solved the distribution problem, at least not yet. Um, we'll see. Um, but I think it has shortcut product and engineering. And I think there's an interesting question about does that actually strengthen large incumbents who might be slower to adopt AI, but once they do, they've already got the same built in distribution they've always had. And can any of the startups come in and move fast enough on product and engineering and build out their own distribution? Rails. And I think you've seen that with some of the AI companies, pure AI companies. So the anthropic is a good example of this. Um, but I don't think you've seen that a lot with anything in heavily regulated industries like financial services or healthcare.

Rob Sarnie: Please come back to part two where with Evan Shinman. As Evan explains how Fidelity Labs is in the sweet spot, working across enterprise wide fidelity and partnering with startups to make great innovation happen. Thank you for listening to what the Heck Fintech. Make sure you tune into our next episode as we continue to to interview fintech leaders and share real world stories from across the fintech ecosystem so you can fully understand that fintech is fin life.