The Exchanges

Q And a few years in, where, where do you land in the balance between services and, and technology? I mean, are those problems solvable via technology brute force?

A Well, that's, that's, that's the fundamental question of the business model for Palantir, is your goal is always to productize something, and to be, to be clear, I'm, I'm an advisor, I guess, but I haven't been there for four years now, so I'm just more like a friend of the company and a shareholder at this point. But their, but their current strategy is always to identify new areas where you can productize, but before you productize the first time, you're going to have a services component while you're figuring it out and hooking it up and stuff, and then you learn it really well, and then you make the product something where if you do it next time, every time it's easier, it eventually, eventually gets to the point where you just productize. So the early areas Palantir works in, you can just install it, hook it up, it works right away. Whereas all the new areas on the, on the outside is always going to be some services component while they're learning it. And that, that's, that's how you think of it.

AI assessment note: “before you productize the first time, you're going to have a services component”

Q there's a bunch of entrepreneurs who are starting companies or running companies, a lot of which, or a number of which, target Wall Street. So you're selling to, you know, great industries, but which are known to be particularly hard to, you know, to start a relationship with. Any lessons learned there in terms of how you get a big government contract or how you start working with those banks?

A It took, it took quite a long time to get them to take us seriously. Um, it was a really good thing that we had, first of all, older people. I don't think there's a, my roommate and I, his 21, twenty-two-year-olds had a product that they would have actually trusted us with any of these big things. You had to have older people who were involved, who were able to bring in, and then we had some mentors, guys like George Tenet, who, thanks to, thanks to our CEO, were able to convince him to be inspired by us and to help us out, and then, and then he was really smart about how to get in and get in and get access and how to come across the right way and whatnot. So we had, We had quite a few mentors in both finance and defense that were really critical. Um, I, I, I think with a consumer company it's really easy for really young people to build something that takes off and, and takes over the world virally. I, I don't think that's, I think you have to, you do have to include like older people probably for these, for these institutional sales and who actually have a big role in the company. Otherwise they don't, they don't take you seriously. And there's probably a lot of lessons too about how you interact with them in a way where you come across correctly. Um, I, I, I think, I, I think, I think also it's like You can never really get the giant contract at first. Whenever you're starti…

AI assessment note: “It's always, like, a little pilot, then a little bit bigger pilot”

Q Okay, great. Um, at a par, what, what kind of data problems have you been solving there?

A So at a par is a lot more deeper esoteric data problem. There's, There's about 6000 custodians in the world. A custodian is like a Schwab, or a Fidelity, or Bank of New York, and they each have their own thousand or so transaction codes, plus a lot of little, little extra things, and, and, and basically to get all that data to talk to each other was, it was, it was similar to Palantir's problem, but it's a very specific esoteric thing, because you have things like partnership accounting, and tax slots, and multi-currency, and all this other stuff, and so to be able to actually go into like a ten billion dollar RIA, And run and solve the technology problem. I thought this was going to be actually relatively easy after the Palantir stuff. It turns out it took us about three years with a team of, like, you know, 30 or 40 really talented, you know, product engineer people to get, to get, like, the basic systems to work at these places. And then, and it just, it just, again, it's a, there's so many hard data problems in the world, and there's so much mess out there. I was, I was shocked to discover a lot of the custodians were giving us bad data, custodians, we've all heard of their names, and, like, their systems are that broken that they're, you know, the, the, the, the world of data is just so messy still in big corporations. It's just really shocking.

AI assessment note: “to get all that data to talk to each other was, it was”

Q So, so it's, it's, it's, it's a network effect in terms of, uh, knowledge within the organization, or in terms of sharing data?

A It's the fact that they're actually, they're sharing data, they're, they have a, It's like, it's like, there's a history of like, of like, like how, you keep track of how they're doing the analysis, and then, and then how you're sharing it with each other. There's like, methods and policies they put in place where this is the only way they could do certain things, and then, and then because you have this network effect and you spread, you get to be dominant, and then everyone wants to build on your platform, so you open up the platform. I think there's over a thousand partners now, actually, building on top of Palance here, so being able to be, have a network, and then being able to be a platform, I think is absolutely critical to be the really, really big companies, and I think, I think that that's definitely, that's definitely a lot of what Adapar is doing, and a lot of our investments we're making right now, it's not just about, it's not just about solving these hard problems in big industries, it's about building it in a way where you do create, you do create these networks and platforms, so I think those are going to be the biggest companies that are, that are coming up.

AI assessment note: “It's the fact that they're actually, they're sharing data”

Q Another one? Ok, I'll ask the last question then. Um, and sort of very generically, but what, what, what do you see in Asia in terms of innovation around data? Um, anything specific or different or?

A Well, I think there's not as many really hard data problems being solved yet. If you're gonna go outside of the U.S., I think probably Tel Aviv and Seoul are the two strongest, uh, innovative areas in my opinion from what I've seen. And so you are seeing some new things there, but I think, I think most of the places outside of Silicon Valley Are still more focused on the consumer wave than anything else, and right now in Asia, the big thing that's happening, the secular trend, is the creation of this giant middle class, and so I think you're seeing some really clever mobile innovation in China and Southeast Asia, and, and really clever ways of, of targeting these consumers, because that's definitely a big thing right now. You don't get as many people buying as much software there yet, and, and it's, it's, it's, it's really hard to do the smart enterprise stuff there, so I, I, I think, I think we're just starting to see them become consumers of a lot of the data workflow companies, and that's, that's a new thing, And I, and I don't think you're seeing as much really hard data innovation, though, yet over there.

AI assessment note: “there's not as many really hard data problems being solved yet.”

Q Does the ontology keep learning? I mean, how is it a data?

A Yes, yes, so the ontology, so the idea is there's, even though we're against using artificial intelligence to figure out, like, Fundamental truths about the world. You constantly want to use machine learning to, to, to figure out, to, to keep learning about ways to update the ontology, and then to suggest it to the analyst, suggest to the administrator, you know, here's, here's, here's a new way the ontology should work. Here's how we're going to set it up. And then the, the really interesting problem becomes, let's say, I'm not giving away any silly secrets here. Let's say, like, the CIA is using it, and MI-VI is using it, like, who knows if they are, but let's say, let's say they want to talk to each other during a crisis, but of course, their ontologies are very different, and so they're, they're calling it with different things, so now how do you, how do you enable that, and how do you put, and then how do you deal with the collaboration rules for who's allowed to share what, depending what database it came from, and then, and it gets really messy, because you might have a hundred databases, and maybe 62 of them within, say, group A, you're able to, Ingest right now, but maybe a lot of those databases you can't ever ingest. They're, they're going to exist independently, so you have to have something sitting next to it, talking to it, and updating it. And so you have this, l…

AI assessment note: “You constantly want to use machine learning to, to, to figure out, to, to keep learning”

Q And the typical problem when you try to sell analytics is that it's sort of the ROI, right? It's like, how do you demonstrate Success when you do a pilot with 60 people, but how do you, you know, how do you get them from doing the 60 person pilot to something much larger? How do you convince them to spend money?

A Yeah, I think, I think the business relationships are really key. I think there's a lot of really talented people who work on the technology side in these big companies, but the incentives and the culture is not set up to work well with bringing in outside technologies and bringing in new technologies whatsoever. You need, you need somehow to have a really pressing business need I mean, I guess that's another thing I'd say is it has to be something that they desperately need, and it has to be an order of magnitude better if you're a small company. You can't just come in and have a slightly better solution than IBM and then somehow win an RFP. It just, it never works. You have to be, like, massively, massively better to the point where it's literally, like, like I was saying, like, something's gonna cost a billion dollars, and you're gonna charge them a few million, and it's gonna get a better result. And if it's that big, because, because, because, because, because they're not predisposed to work with these crazy kids who are coming and who they'd never heard of before. Like, they'd much rather work with the people they're working with. And so, so you have to have, like, the top business guy's buy-in who he knows is critical for his business. He's desperate for something. Something's not working that you're fixing, and then it has to be much, much better than the alternative. U…

AI assessment note: “we'd start with these really small contracts... once we succeeded, we'd see another bigger thing”

Q And, ah, so, you know, stop me when, ah, it'd be awesome to hear some details about that, obviously, whatever you can talk about, but let's take the, I don't know, data integration piece, for example, which is like the holy grail that everybody's been trying to solve for such a long time. What's the, what's the approach there?

A Sure, and I don't know if there's any one secret. I think I was, ah, I probably attribute a lot of my success to the fact that I have a lot of smart friends from Stanford and MIT who were able to convince us to work on this, so I think I, I, I think right now there's an interesting thing in the culture where everyone wants to do their own company, and it actually makes it really hard. I think there were five of us, my roommate from Stanford, Peter Thiel, Alex Karp, and Nathan from PayPal, who, who founded Palantir together. So first of all, there's already a lot of us, but then like the next 20 people, I think any of them these days would have gone to like Y Combinator maybe instead, given the culture, but we were able to convince them all to instead come together. So I, I think, I, I think the real secret to Palantir's success was we just got a lot of awesome people, we gave them all a lot of upside, and we all worked together on it. Um, and in terms of the actual technology problems for, for productizing data integration, um, you know, we kind of built it out in information theoretical way, so we, we have what we call dynamic ontology, where, ah, any instance of Palantir works with its own ontology of the world, its own schema of the world, so if you're in a certain part of, ah, defense analysis, there's gonna be 50 different types of missiles, and that really matters, wherea…

AI assessment note: “we have what we call dynamic ontology, where, ah, any instance of Palantir works”