AI’s Role in Revolutionizing Technology with Perplexity’s Dmitry Shevelenko & FactSet’s Kate Stepp

Landscape AI's role in revolutionizing technology and workflows. We're joined by Kate Stepp, she's the CTO at FactSet, and watch this: Dan Demetri Chalenko, the Chief Business Officer at Perplexity. I did. And by the way, for you folks playing our home game, Demetri is coming off a great season with the Detroit Red Wings where he led the team in scoring. Kate, how are you doing?

I'm excited for this. Thanks for having us.

So am I. So Denis as well. By the way, everyone registered for this webinar will get three free months of Perplexity Enterprise Pro for up to 50 seats within their organization. You can start to do that math. You'll get an email about all that stuff. So I think Demetri, maybe we can just jump right into things. I want to make sure we cover all the things that we want to cover.

Yeah, and if it's okay with you, I'm gonna ask the first question. Well here, I just want to take one step back here. I mean, Guy and I are really fired up about this. FactSet has been great partners of ours on social media now going on four years or so. Guy and I are longtime users of your products. We're also, we're also early adopters. Maybe Guy you can just say, nod and say you're an early adopter. I've been using Perplexity.

That happens to be true.

Listen, it does happen to be true. Okay. I've been using Perplexity for quite a while. We had the CEO, co-founder of Perplexity on our podcast back in February. He's been on Fast Money with us. So it is truly an honor to speak with Demitri because he is at the forefront of all the deals that they're doing. And then we're going to talk a little bit about their partnership with FactSet.

But these new iterations are obviously going to make things that much better. And when I say that much better, exponentially better. So Kate, I'm going to start with you. As we mentioned, we're longtime FactSet users. We've seen some of the most impactful improvements to using the platform most probably in the last decade or so, specifically with the launch of Transcript Assistant earlier this year. So tell us about that, and tell us how groundbreaking it is.

Well, it's interesting. We were down at the FactSet Focus event a few, I guess it was in the spring, and you guys had just launched this Transcript Assistant. And you guys had a bunch of breakout rooms where you were doing sort of trials. And it was amazing because it was kind of like standing room only. And to Guy's point, we've been using your products for 20 years. To have this sort of step up in capabilities for a product like that, when you think about four times a year if you were a market participant, a professional market participant, you're getting these earnings calls and you're really leaning on that.

Because I'm sure you're making tweaks along the way.

No doubt. And so around the time of that launch, Demitri, you guys launched your Enterprise Pro product. We had Aravind Srinivas, CEO of your company, on Fast Money, kind of mapping out some of the things that you were doing away from that initial consumer experience on the show with us. So talk to us a little bit about the launch, who you wanted to partner with, what sort of verticals, because obviously FactSet was one of your first select channel partners. And speak to us a little bit about why you went into the financial services vertical right out of the gate.

Yeah, so when we think about our most important consumer demographic of Perplexity, it's knowledge workers. And that's who we want to be a best-in-class solution for. And what we heard from those knowledge workers was that they were using Perplexity at home but couldn't use it at work because their IT departments were worried about data security and other policies. We want to make that easier for folks. We wanted people to be able to use Perplexity 24/7 and not have to resort to their personal device at their work desk just because somebody in IT wasn't happy with a given solution. And then as we thought about all of the verticals in enterprise, financial services has been our breakout star. And it makes sense, because people that work in financial services do a lot of research, they have a lot of questions, there's a lot of novelty every day. You have to keep up to date with a fast-changing world, and Perplexity obviously gives you superpowers there. And so when we think about why a partnership with FactSet was such a natural fit, FactSet has this incredible data, both structured and unstructured. And when you combine that with Perplexity as a synthesis tool, we're able to combine the best of FactSet, the best of the open web, and the best of your internal knowledge sources, and give you those concise answers. So we're incredibly excited about the partnership and just getting started.

Yeah, so when you think about that, Kate, you're obviously at the forefront, you're always checking out new technologies, seeing how to integrate them into your workflows. How did you think about integrating a company like Perplexity and their technology? Because as far as we're concerned, we saw the uptake on the consumer side. And I can give some examples how I use it for my business—not just my business, but when Guy and I are being goofy on TV, it's something that I've learned to rely on. Last night on the show, I used Perplexity to try to understand a Chinese EV maker. And you know, for folks in the trading seats or portfolio managers or analysts, when you need to be quick, quick, quick, you're starting to rely on these solutions. So you guys obviously like to go—and I'm just going to say this because, Chalenko, what did you say, Guy? He was the leading scorer for the Wings last year, right?

Yeah, we'll fact check that.

But I'm going to say this and Guy usually cringes: you guys are going where the puck is going, right? And so speak to us a little bit about how you think of this technology and where it's going, and whether it's being driven by you or a combination by you and your users.

Yeah, it's an interesting point you make because Demetri, we were talking about you know—I think when we think about the use cases here, there's a lot of people spending a lot of time thinking about, how do I just sort of get my arms around the technology? How do I explain to other individuals how they can help leverage the capabilities of AI?

I think it's—I think in some ways we've over-exoticized AI. And the phrase itself isn't that helpful. Like, I think of what we do is we save people time, right? And so, working backwards from sitting down with your analysts and just breaking out, like, okay, in a given 60-hour week, what were you actually doing? And what ended up taking up a lot of time? And then I think, you know, what's magical about Perplexity is there's a number of questions you can ask on Perplexity—again, answer that doesn't do your job for you, but it gets you 80-90% of the way there. There's that uniquely human judgement that that's what AI is all about, right? It's kind of finding these productivity unlocks and letting people do the things that they are uniquely positioned to do, where you want that human judgment and intuition driving work outcomes. So working backwards from again, just thinking about how we spend our time.

Yeah, you know, Demetri, when Guy and I think about this, it's almost like the first pitch in the first inning, is that fair to say? Because we're spending so much time, to your point, going over what the verbiage is and what are the use cases and what's the return on investment. And we're just kind of scratching the surface. And so Kate, picking up a little bit on what Demetri just said, I mean, I've heard folks use this term 'copilot' and 'agent' and I don't know if that's the same or if they're different. So like, why don't you give us sort of a working definition of how you see the spectrum from copilot to agent?

Yeah, so it's a great segue here, Demitri, because I don't know if you guys caught Marc Benioff last week. And I think every time he opens his mouth with a camera in front of him and a mic, he's talking about this. Obviously the CEO of Salesforce, he's basically talking about AI agents versus copilots. And he basically—listen, I'm sure you guys, well maybe more FactSet—you guys are partnered with Microsoft on certain areas. But I mean, what does that mean to you that he keeps saying that copilots are like the equivalent of Microsoft Clippy from 25 years ago, and AI agents are the way forward? Speak to like our viewer, our listener, what the difference is.

I think there's a lot of kind of marketing jargon rather than substantive indicators of different capabilities. The spectrum that we see is, right now you use a product like Perplexity, you use other AI tools, it's user-initiated queries. You're asking questions, you're getting answers back, asking more questions. I think the transition we're going to see is as the system understands you better, understands your needs, your preferences, it can begin initiating its own queries. It is basically asking the question that you are going to ask and delivering the answer to you in a way that's useful without you needing to do that work. So I think that is the spectrum that we're going to see acceleration on.

In some cases, you can think of it already happening. Perplexity Shopping, for instance: if you're asking about products, you can just buy the product right from the answer with one click. And then that third phase is again that what we would call agents, where the system knows enough about you and your needs that we're just initiating those queries, taking those actions, and you're just approving them rather than having to initiate them.

Yeah, that's logical now and it makes a lot of sense. But it's obviously something internally you considered. So speak to us sort of the road map to get here.

Yeah, hey Demitri, so Kate just mentioned trust. And that's something that I know that you guys spend a lot of time on. Obviously, a company like FactSet, where they have mission-critical data, you think about the users of that data—they can't afford to have bad data inform decisions. So when you think about—and we hear this again and again—hallucinations, right? And some of the early reporting that we saw, I think it was that reporter from The New York Times had that long conversation that went in some weird places. And I think that was the thing that, you know, for me, I've used Perplexity—well, I had a question about a Chinese EV maker recently, and I was asking Perplexity, and the answer seemed a little goofy, but I don't understand the Chinese language, so this is a Chinese EV maker. And I had to click through to the citation, which brought me directly to BYD's website. So I felt pretty confident about that. Speak a little to this idea of hallucinations, how you guys have kind of tried to get better and better at getting the right answers to people, and then they feel as confident as they do to use those answers to make decisions in real time.

Yeah, so one of the founding insights of Perplexity was you're not going to usefully answer people's questions just with large language models on their own. And it's just really this merger of search and sources with large language models. And that's our secret sauce. We show you the sources that we're using to produce the answer before we show you the answer, because that grounds your confidence in what you're about to see. And then you obviously see on a line-by-line basis where the information is coming from. So there are several aspects here. One is transparency. If you know where the information is coming from, you can make your own assessment: is that a trustworthy source or not? And you obviously can then click through and dive in further and get to your own judgment. And then, just on a core capabilities basis, if you're not relying on the large language model as a store of knowledge and you're just using it for the skill of synthesis and summary, that alone materially reduces the spectrum of where hallucinations can occur.

Let's talk about workflows. Because obviously, there's some very tangible things here. So Kate, what do you think the most impactful ways you've seen AI integrated into the existing financial workflows, from the front office to the back office operations?

Yeah, and I want to follow up on this, Demetri, a little bit. Because you know, there's a lot of verticals that you guys could have launched this Enterprise in, and I know you're in other ones, right? But when we think about markets—and Guy and I have been sitting in front of FactSet machines Monday through Friday, 8 in the morning till 5, 6 o'clock at night for, you know, Guy for 50 years, me for almost 30 years—and when we think about just the markets, they're moving all day long. And in that seat, you have to be constantly reading, synthesizing, adjusting. It's non-stop. How do you guys think about serving an industry like that? Because there are far simpler industries where you don't have things moving every day, with really unpredictable things. You need to have great tools. So talk to us a little bit about this industry and what makes it interesting but also challenging.

Yeah, I mean, what makes it interesting is that there's a really big problem: there's just too much information in the world. Just a single earnings season for a single ticker, there's 20 sell-side reports you have to read and digest, and then act on that. The source of alpha can't just be somebody consuming the information faster. The source of alpha is knowing what to ask, and that's where I think the new alpha is. And so I think because of just the vast volumes of information and all the different questions you can ask to interrogate that information, it's a great platform for what Perplexity can do. It's challenging as a vendor. There's a lot of security requirements that financial services companies have. We've gotten very good at filling out security questionnaires, but that was definitely an acquired skill. And I think the newness of AI is causing concerns and questions from a regulatory perspective.

Yeah, and I want to hit regulation in a second. But Kate, I wanted to bring it back to you. So when you mention, Demitri just mentioned security, right? We have compliance, we have regulation. These are like the three pillars I think of financial services companies when they think about how to manage their data, how they interact with clients. All like, how do you put all of that together? Because again, this is a new technology. A lot of consumers are just kind of figuring it out. Enterprises are the ones who are going to unlock, I think, a lot of the benefits near term. You guys all, it probably makes your product that much stickier, and it probably is bringing people over from other platforms that don't have this kind of technology integrated into their processes and services. So think about just give us a sense of how you're thinking about compliance and regulation in this new era.

Yeah, Demetri, there's a new administration coming in as we know. I mean, regulations are probably going to change. My sense is, you know, for the more lenient, let's put it that way. But how do you see it evolving regulation in this new incoming administration?

Listen, I think we should regulate the use cases of the technology, not the development of it. And there's no reason to make LLMs less smart than they could be. But like any new technology, people can use them for not great reasons. And that's where we should be putting the regulatory leverage. But there's just so much good that can come from the increased productivity that this new generation of AI is enabling for all of us. So we need to be open to that while being sober to understanding those nefarious use cases and having strong laws around that.

Yeah, and I think where Guy was going with that was, Demetri, let me ask you this. Because your background, you were at pre-IPO Facebook, you were at LinkedIn—these companies all had really good outcomes. And then you think about your time at Uber. Uber was something that you had to navigate a lot of regulation. So when people look at a technology like this, and most people aren't even using it yet when you think about it. So I guess, like, how do regulations get framed from your past experience in a way that keeps some guard rails on? I can't think of a better way than a partnership with enterprises like FactSet, because like, to our earlier conversation, they are so focused on compliance and security and privacy and regulation. So how are you guys thinking about—is it more mission-critical to get these partnerships to kind of help you navigate that regulatory world?

We're trying to design our product in a way that's future-proof to where the world is going. So for example, for enterprise users, we don't train any of our internal systems based off of prompts from enterprise users. Because that's obviously a concern for them, but that's also something that could get enshrined in a legislative or regulatory way down the road. And so I think we're trying to anticipate where some of these modalities are going. We don't resell data from consumers. We just use prompts to improve our own products on the consumer side, but that's it. And so I think being aware of the future privacy implications, kind of seeing what regulators are talking about and how they're thinking about it, and building in safeguards from day one is really important.

Yeah, and Demitri, Kate just mentioned automation. So the idea that you put in all these prompts, you engaged with the platform, maybe it was on transcripts, maybe it realized that you're only focused on technology and specifically the price action of stocks after a beat and raise. And it's going to be learning, it's going to be all the prompts, all the other users. So what do you—how far away are we from automation, where basically the system learns and realizes your behavior and what you're looking at, and maybe starts pushing that information to you? Like, you want to see Tesla's earnings or whatever. At what point is this going to get a lot easier for users? And are we years away, or is it some sort of automation that's going to happen in the not too distant future?

Yeah, we were talking earlier about Perplexity's daily podcast, the Discover Daily. And we're going to soon release a version of that that's personalized, where you'll get your Perplexity podcast that's just about what you're interested in and the topics that you care about. And that'll be different than Guy's and it'll be different than mine. So we're already moving in that direction. I think when we were talking about that spectrum earlier, you can almost think of it right now we're very much in the pull model. You're really asking questions and getting answers. I think the next phase is push—a Perplexity that is anticipating, 'Hey, you might be interested in this based off of that.' And the challenge is giving somebody something they didn't ask for and being wrong is worse than nothing at all. So it's a very high bar. And I think you want to be very thoughtful as you're building a product like this not to annoy people, because the last thing you want is more information overload that's not actually simplifying your life. It's going to require a lot of experimentation. I don't think it's necessarily about pure technology—it's about the user experience and nailing it. I think the technology exists. I think it's just a matter of iterative product development.

Well, we'll be here. And speaking of the personalized, we wanted to mention we absolutely love Discover Daily. It's 10 of the most impactful stories, things that are trending, and they basically synthesize it, they have a voice bot reading it. And it's just really great. And it kind of spurs some of the things that you guys built on: the idea of curiosity and finding the answers. Guy and I were so enthralled with it. We do a similar thing. Our team curates about 10 market stories, business stories of the day. Now this is my question about automation: we put those 10 stories in a chat for our team, they put them into Perplexity with a prompt to summarize it, they take the text, and then we put it into a voice bot and it reads it. Now our name, I think, is a bit cooler. Ours is called 'The Halftime Report Podcast.' But I'm just curious, Demitri—how close are we to a fully automated version of something like that?

Absolutely. I think in some ways, this gets into some issues around creators and artists. When people consume that podcast, they consume it because you're affiliated with it. So they want to feel there's something special that you did to that prompt that they're relating to you. I think where it gets into full automation is something that is not possible at human scale, which is a completely individualized personalized podcast for every one of your listeners. That's something that you need the full automation to enable. And so I think we're getting close to it, but it's not quite there yet.

I want to give you the last word here, Kate. Guy, I'm sorry. I apologize. My final question to you, Kate, was a statement. You should feel incredibly proud with your team and FactSet that you guys were so ahead of the curve on the AI front. But my question to you is: how are you going to leverage this to have this sustained competitive advantage that you're enjoying right now at FactSet?

Well, listen, I think we're at time here, folks. Guy and I really appreciate this. I know the folks at FactSet understand that we are honored to be a steward of your brand. You guys have been great partners of ours. We use your products, your services, we rely on your data. Our listeners, our viewers, have become accustomed not just to us articulating how we use your services, but we've had the benefit of having so many of your thought leaders come on with us and kind of educate our users. So Kate, thank you so much for your time. And Demetri, thank you. We've been incredibly impressed. You guys have come on our podcast. We're really excited about what you're doing. It brings me back to the first time—and I've said this on our show—that I remember using Google, getting off a search engine. This goes back 25 years ago. I haven't had this sort of web experience since then. It just changed the way I think about interacting with the internet. So we're super excited about what you guys are doing. We are amazed at the timeframe you are shipping product, and not just shipping product, but product that works, and is really changing the way folks like us have done things. So Perplexity, we're rooting for you guys. We're users. And we can't say it enough: this combination, this partnership, is something that we're benefiting from in many different ways. So we hope your users, Kate, and your users, Demitri—and by the way, those that are on this call get a three-month trial. You'll have all the ways to do that. So Demitri, Kate, we really appreciate your time here today. Thanks guys.