Ritavan on System Gambits and Finding Leverage

Ritavan

Thanks, Matt. Great to be here.

Ritavan

Yeah. I mean, about the moat, right? You kind of picture it as if you can sort of measure the thickness of the fort wall or something. I think that’s very often the image, and it makes it sound as if you’re standing in front of a fortress, you need to get out your measuring scale, and then you write down, “Ah, okay, it’s that many meters,” and that’s done. But it never is like that, right? And so I don’t really like the idea of having the moat as an entry on the checklist, because it’s not something you can just tick off. I think what is more important is if you want something defensible, the focus should be on figuring out if the business you have is a system that compounds in a way and at a rate that some other business you’re comparing it with just cannot. And I think that is a much more fundamental idea that is independent of, oh, okay, does it have a moat, or is it growing, or whatever it is, right?

Ritavan

Yeah. So just to take a step back, right? The word gambit is from chess, and in chess, what you do is you sacrifice a pawn, or you sacrifice some form of material, to gain a positional advantage that then allows you to either win the game or dominate the game. And the system gambit is essentially generalizing that idea to systems and hence to business and investing. And the idea is the following. You are in a system but you wanna cross into a new system, right? So that’s the difference with chess. You’re not in the same game. You’re trying to go in a new game. To get there, though, you need to sacrifice something in the current system. You want to gain a structural advantage, or you wanna build a structural advantage that then allows you to compound in this new system. And there are three properties that are super important. The first one is a self-improving loop. That means every iteration of that loop has to structurally get better. If that’s not the case, you have a nice asset, right? You just have a nice asset. You don’t have the compounding. The second important thing is path dependence, meaning if it’s something that anyone else with just more capital can buy on the market, then you don’t really—then you haven’t really executed a system gambit, because it’s something that someone with more capital can acquire later on. So it has to be something that is built loop iteration by loop iteration, right? That’s the path dependence. And I think if you have these two things already, right, you’re in a very good place. And the third thing, which is super important, is management logic antagonism. So what that means is you are doing things in a certain way that if anyone else wanted to copy you, they’d have to stop or break what they’re currently doing, right? Once you have these three things, you have executed a system gambit successfully. If any one of these is missing, then it’s not a system gambit. It’s something else, right? It can either be a good asset, it can be a wild bet, it can be a bunch of things, but for it to be a system gambit, you need all three.

Kai

So the gambit aspect, you’re saying it’s like a chess analogy, sacrifice a pawn to take a queen, but the key difference is that it’s not the same game. You’re switching game. You’re switching systems.

Ritavan

Correct.

Kai

So is the sacrifice in this case, the cost, is it simply the switching costs of moving from system A to system B, or are there material costs in addition to that friction?

Ritavan

The goal is to be willing to sacrifice into a new system. And sometimes that requires a whole bunch of things that are non-material, but it could also be material costs, right? But I think the important part is you are choosing across from the system you’re in to a new system. And the sacrifice is often very quantifiable in the existing system, right? So if you’re operating in a certain system, you have metrics to quantify how you are performing in that system. And in a very large way, the sacrifice is often seeing those metrics collapse, right? Precisely because you’re moving out of that system. That is why it hurts, right? It’s not so much, “Oh, I need to buy this new thing,” or whatever. It’s much more, “I am performing really crap in my existing system.” And I think that is what is very hard to do for people.

Kai

Interesting. So I sit here as, like, a public markets investor. So I’m not inside the walls of a company. I don’t get access to all the kind of private information and stuff under the hood. So from the standpoint of a public market investor who’s looking at companies and stocks, how can I assess which companies are, first of all, attempting to execute this sort of gambit, and second, which ones are positioned to potentially successfully do it, right? You mentioned these three prerequisites. What sorts of metrics, data, or whatnot would someone like myself look at if I’m trying to determine in real time? ‘Cause obviously we can look at historical examples and say, “Yeah, we know with the benefit of hindsight that company X successfully did this.” But in real time, how am I supposed to determine which companies are in the midst of executing this transition?

Ritavan

Yeah. So I think looking from the outside, the most important thing you wanna ask yourself is, how hard would this thing be to replicate, right? Because the core idea underlying the system gambit, I just hold it up. So this is a pre-launch copy, so it’s a bright, bright blue cover, so it wouldn’t be hard to find. But the subtitle is that it’s about finding leverage to unlock compounding value. And the idea of leverage, it’s not financial leverage, obviously. It is leverage in the old sort of Archimedes sense, right? You want to get the most impact or value with the least amount of effort or pressure you apply. And the way you can try to figure that out or sort of reverse engineer this from the outside is to ask yourself, how much of this is leverage based on asymmetric proprietary strength of the company, right? How much of this is being done in a way that others cannot just copy?

Kai

Yeah, I mean, I think what I’m thinking about is this, which is the moat—go back to that analogy—the moat is dependent on the game you’re playing, right? So in a certain setting, for example, code can be a moat, right? If you’re the only person who possesses the ability to code, then you have a moat, or you as a company have a moat. You know, the obvious analogy today is that the world has shifted to a world where AI can write code very cheaply and quickly. And so anyone can vibe code or has access to advanced coding tools. So the game has changed, and so perhaps that moat is no longer a moat in the world of AI. And so I think as we step back just from this one particular case, you have to be very thoughtful about what game are you playing. In other words, what system, to take your language, Ritavan, is currently the dominant paradigm. And then the question becoming is, what is a scarce asset? What are the moats in that setting? And I think the context matters a lot. And so I think what’s real interesting about your work is it kind of forces you as an investor to think through, you know, how is the world changing and what other systems are available to businesses, both through the technology and other factors. And then you ask the second order question, which is for each company, as you kind of go through the list, what assets do they possess today? Are they optimized for system A or system B? Because obviously as an investor, you’re looking to kind of front run change and look for companies that might be priced based on the past paradigm, but other investors in the market might be missing how their assets can actually be quite valuable in the future, right? So like we saw the other day, I think it was Getty, right? Their stock rerated because it was—they made a deal, I think, with one of the AI labs. But the idea just being that, hey, look, proprietary data could be really valuable for training, right? So that’s information that was of course available and valuable in the past, but perhaps is even more valuable in a world of AI where it can be used as an input to an AI model. So I think looking for analogies like that, situations like that, could be a helpful framework for investors.

Ritavan

Yeah. Let me actually just jump in with a historical example, because the moat word obviously comes from fortresses and protecting a physical fort. And there is an interesting historical setting where—this is during the Ottoman Empire. There’s an Albanian prince who’s been taken young, Gjergj Kastrioti, and he’s sort of, you know, trained by the Ottomans, becomes a cavalry commander. They give him a Turkish title, which is Iskender Beg. And then at some point they send him back to run that part of what was then—what is now Albania, but that region back then. And at some point he figures out, “Okay, I think I’d rather fight for my folks,” and breaks away from the Ottomans. And you know, that region and all the regions around what was the Ottoman Empire then really struggled with the Ottomans, because they just had these vast armies, and they were able to marshal these resources, and people couldn’t sort of hold out against that. And what Skanderbeg then does, interestingly, is—he’s got his main fortress. The Ottomans send a massive army. They lay siege to that fortress. And now comes the paradigm change, because it’s the same fortress, the same armies. Everything is the same, right? So if you go by your checklist, you’ve ticked all your boxes, and actually the outcome is very clear, which is the Ottomans eventually have a successful siege, and they take over the fortress. What Skanderbeg does now is suddenly invert the paradigm, right, in some ways. So he says the fortress was seen to—like, it had to have a big moat, tall walls, et cetera. And the idea was you lock yourself in the fortress, you let the army lay siege, and you try to outlast the siege. Skanderbeg instead leaves a very small garrison inside the fort, gets the hell out before the Ottomans arrive, and hides in nearby forests and hills. The Ottomans now lay siege, and now suddenly the vast Ottoman army that can actually be modularized and moved around, et cetera, on an open battlefield is fixed around the fort. And then he just keeps harassing them night after night after night. And I think this is exactly my point, which is if you have a checklist, a checklist is a list of binary items that has been abstracted out from a causal model of how reality is supposed to function. If you change that causal model, right, if you change the rules of the game, if you change the game itself, that checklist per se is useless, because the checklist is just an artifact. And I think this is sort of the core thrust and contribution of The System Gambit, is stop trying to operate and look for patterns at the level of the artifact, right? Go down—from first principles—and look at the system you’re operating in, because that is where the opportunities are. And it might not always be visible, right, from the outside what a business is doing, but I think what you can do from the outside is just ask yourself, “If you were in their shoes, what would you do? What do you see?” So do not operate at the checklist level. Go a level down. And if I go back to how the whole checklist thing exploded and became so popular with investors, I think it’s Mohnish Pabrai who, at least as far as I know, started popularizing it in my world. I’m sure a lot of people were doing that too. But it’s based on Atul Gawande’s book, right? It’s—I forget which year, somewhere around 2008, 2012, somewhere around there. So I was at the end of high school, and I remember I was excited reading it then too. And what Atul Gawande, who’s an excellent surgeon, essentially observes is, hey, surgeons, pilots, a bunch of people are using checklists successfully, right? That’s his core thesis. But then you have to ask yourself, what is the causal model on which a pilot operates or an aircraft operates? What’s the causal model on which the human body operates, right? And all of these are based on hard science, more or less, right? So the human body does not change based on someone’s social media post, right? The physiology of the human body doesn’t change, to a large extent, as far as surgeries are concerned, right? The same with aerodynamics, right? Mood of the passengers does not change, you know, the turbulence outside, et cetera. And so the causal model is what is really the underlying sort of source for that checklist. And if the causal model is fixed, then it makes sense to rely on checklists. But when you have the opportunity to either change the causal model or see it change, et cetera, then the checklist as an artifact itself should be questioned.

Ritavan

So I think the narrative question is, for me, independent of a specific structural aspect of the system, right? I think you can take anything and try to get the best possible narrative out. So I’d remove the narrative out of the picture for now. Speaking of the moat, I think the important thing is to ask yourself—like, if we go back to that example I just used with the small Albanian army and the large Ottoman Empire, and I have a bunch of these in the book, right? So I’m trying to use those that are not in the book. The book is just full of examples across history, across the world as much as possible. But the idea is the following: it’s always the same resources. That’s the same army. One is bigger, one is smaller. It’s the same David-Goliath kind of setting. But the idea is: how do you win despite not having the most resources or the largest size, et cetera? And all through history, in every aspect, whether it’s sport or business or music, whatever, right? You keep seeing this thing happen, because otherwise, logically, if there were no asymmetry, the largest guy should always win, and it’ll converge to a point where there’s only one large thing, right? So clearly, there is a mechanism in the world that allows you to leverage asymmetry, right? To go into a new system, and then that new system has a structural compounding loop, right? With self-improvement, with path dependence, and with management logic antagonism that allows you to really diverge, right? So the original system continues in the direction it was going. The new system goes in a different direction, and those two aren’t really compatible, meaning you cannot have this and that at the same time, and that’s why the sacrifice, right? Now, to go back to this Gjergj Kastrioti or Skanderbeg example, Skanderbeg could not fight in the Ottoman paradigm in which he was trained by staying in the fort and defending the fort based on the moat and the walls of the fort and win. He had to sacrifice the fort in the sense of put a much smaller garrison, let the Ottomans lay siege to it. Yeah, those guys probably had a really bad time inside the fort, the poor guys from the smaller garrison, but it allowed him to win the war, right? So you have to sacrifice something. So to the Ottomans, it looks stupid. They say, “Oh, this guy’s such a loser, he’s run away.” But he doesn’t mind being called a loser short term if at the end he wins the war. And I think that is the point. Very often, David with the small sling, right, aiming at Goliath looks like a fool, looks like a loser, but he’s precisely using that asymmetry. And that’s sort of—it’s almost like a celebration of thinking about leverage and asymmetry, and this is what The System Gambit is about.

Kai

I’m always a fan of the underdog, so I like that. And I think you make a good point that if it were always the case that the big guy always wins, then we’d have one company, right? So what you’re saying reminds me of the work by—which I’m sure you’re familiar with—of Clay Christensen and The Innovator’s Dilemma. Like, why is it that the big companies don’t do the gambit themselves? I’m sure there are things holding it back. Can you maybe talk to that a bit?

Ritavan

So I think Clay Christensen describes a very specific kind of situation, right? Where you have a large incumbent with high-end products, and then you have a new entrant with a low-end quality product. So it’s a very specific description of a similar mechanism, right? What I try to do with the system gambit is to say, okay, Clay Christensen is one example in a concrete sort of market setting. What I try to do is, if we take a step back and think from a first principle systems thinking perspective, what are all possible gambits? Like, can I think—and I think, Kai, you would resonate—always try to find some kind of eigenvector basis, right? Can I find a bunch of dimensions that are as independent as possible? And obviously, they’re not all orthogonal. They’re all independent, but they’re not all orthogonal. I would have ideally had them orthogonal. And the idea is: what are these mechanisms, right, that underlie these type of phenomena? And the book has eight such system gambits, right? And you spoke about changing the paradigm, the goals, et cetera. So the first system gambit is actually paradigm change and goal displacement, right? And so each gambit focuses on one mechanism. That does not mean that you have to execute them independently. In fact, some of them have strong relationships or causality involved. But the idea is I’m giving you eight ways to look for asymmetry and leverage asymmetry. And then obviously, based on your context, based on your setting as an investor or operator or both, you want to ask yourself, what will give me, you know, the highest, sort of the greatest, bang for the buck, right? That’s the question to ask. So it’s not one of those close-your-eyes, follow and tick three boxes type of thing, because I think that’s just trivial. That’s a farce. That’s not how things work.

Kai

In your book you tell the story of the microscope. And so over two centuries, it generated what you said, quote, “visibility without understanding,” end quote, before anyone really had built the standards around it to interpret what it was that the microscope was showing them. So talk to me about this metaphor. Like, why is this the right metaphor for thinking about how a lot of companies are now using AI and AI dashboards today?

Ritavan

Yeah. So the microscope is actually a very interesting example, right? It’s a mad exciting technology when it gets sort of invented, right? So it’s Robert Hooke in what’s England today, and Antonie van Leeuwenhoek in the Netherlands. Around the same time, they come up with pretty different devices, different looking devices that in principle are more or less the same optically. And what happens is now suddenly the entire world—not the entire world, but, let’s say, many people who could afford that—could see microscopic stuff. And so everyone is looking at these little things jiggling and structured, and they’re drawing and hallucinating, right? I mean, that’s what they’re basically doing, because what happened is you could have the same sample, you could have two people looking at it through the same microscope at two different points in time and see something completely different. And it’s a bit like with a lot of Gen AI today, like, essentially you and me can send the same prompt to the same model at two different points in time and get two different answers, because these are inherently stochastic systems. But what happened with the microscope was, in fact, that lighting wasn’t standardized, that the preparation of the sample that was being studied wasn’t standardized, et cetera, and those things had to be sort of sorted out. But even when that happened, the bigger problem was there was no causal understanding of how physiology worked, how human physiology worked. And so even though, over time, you had accurate visuals of stuff, what would you do with them? You know. So the microscope starts out as an exciting technology. For about a hundred fifty years, you have hallucinations. After about a hundred fifty years to about two hundred years, you have new techniques that come out for standardized lighting, standardized sample prep, et cetera. So at least there is objectivity in what you’re seeing. But there is still no revolution in clinical microbiology and physiology, because there is no scientific model to understand that. And this went pretty viral when I shared this article. It was before the book came out sometime last year. And a friend of mine who’s a PhD in math, I think from CMU or Duke, I forget, smart guy, et cetera. And I shared it with him, and he just shot back on WhatsApp saying, “Yeah, cool, but what about the telescope dude?” And I said, “Yeah, true, actually. Let’s dig in,” right? So I hadn’t thought about it. So I said, “Okay, let’s look up what the telescope is.” So again, now from first principles, the microscope and the telescope are essentially the same technology, meaning it’s two refractive lenses in a tube, right? Except unlike Hooke and Leeuwenhoek—Robert Hooke in England—Galileo Galilei points it at the stars. And within a super short period of time of a few months or years, you have a complete revolution in astronomy, right? Our understanding of how the universe worked as a species just changed overnight. And the question is the same technology in two different fields—let’s just go abstract and abstract it out and call them systems, right? So you have the biology system on one hand, and you have the astronomy system on the other. The same technology, essentially the same technology, in two different systems, two very different outcomes, right? One stagnates and hallucinates for two hundred—a hundred fifty, two hundred—years. The other one has an instant revolution. And so that’s—I think the point is the tool is great at whatever level, in the sense of, we’re always obsessed of, “Oh, this model could do that. That model could do something else.” I mean, I’m not saying it’s not interesting or not relevant, right? But that’s not the end of the story, right? So irrespective of what AI or any algorithm or model or technology can do, the question you wanna ask yourself is, what’s the system in which it’s operating, and with what system does it come into contact? And how does that system need to operate so that you can create value? And now, what’s interesting is astronomy, right, had a causal model, which was the geocentric model, right? So everything is turning around the Earth, et cetera. With the microscope, you have an observation that directly disproves that model. And now suddenly you have the question saying, “Okay, then what’s going on? If it’s not what we thought.” Nice thing is you had Kepler and Kepler’s mentor, right, Tycho Brahe, noting down observations, et cetera. So you had a large data set. You had a smart guy who fit a model on that data set, with Kepler’s model of the world. And now suddenly you have these quick iterations between observed reality through the tool that’s empirical, that’s verifiable, and that’s trustable. Meaning Galileo Galilei could look at the moon sitting in Italy, and an astronomer sitting in Paris could do the same thing. They would see more or less the same thing, right? They’re not seeing two completely different things. And you have everything needed to improve the causal model. And when something does not work in the causal model, you tweak and change, et cetera, until reality and the causal model essentially are as accurate depictions of each other as possible, right? And microbiology doesn’t have that. And I think that is the point in business, is you wanna ask yourself, irrespective of what the technology can do, in the sense of—or given whatever the technology can do now, let’s not wait and think what’ll happen in ten years, right? Will AI be conscious or whatever other absurd questions people ask themselves. But ask yourself, given what it can do today, right now, right, what is the system in which it’s operating, and how does it fit into that system, and how do I allow it to compound value? Because that is what you’re really interested in.

Kai

Yeah, I think that’s an apt description of the K-shaped AI economy, right? We’re seeing AI is driving pretty much all stock returns. I think I saw a chart where it was like 100% of the stock market returns have been driven by AI stocks over some trailing period. But of course it creates winners and losers. And what we’ve seen historically in these disruptive periods has been this effect, right? Where the market will quickly look around and say, “Hey, what is perceived to be a winner and a loser?” And then shoot first, ask questions later. I think what’s interesting is that when you actually look at how things have played out subsequently, yes, many of the folks who are presumed to be losers actually end up becoming bankrupt, but many actually recover, right? So I think that’s the piece that people miss. I think going back to what you were talking about with the microscope, it connects to that too. And I think a lot of people, when they think about AI, they’re obsessed with, like, the technology itself. Oh, this is such a cool model. Like, it’s got this many billion parameters. It scored this rating on this metric, like on these various tests. It’s, like, the best model. But what matters more than the model itself, I think, is often two pieces. So one is, like, the complementary assets, like the harness, they call it, right, around it. Like, why was Claude Code such a breakthrough? Well, the model was better, but it was more around, like, the instrumentation they gave it access to. Now it’s running on your local computer, on your terminal. It can look at your local files. It can do tool calls. It can run Linux commands, right? So that was really important. And then the second complementary aspect was just the setting, right? So an AI model, we know what it does. It’s kind of a stochastic parrot, right? It’s a fancy autocomplete. And so in some settings, that’s actually not that useful, right? Like, I wouldn’t use it as my therapist, for example. There’s too much downside there. But when it comes to code, software development, actually it’s the perfect setting to be using an AI tool, because it’s closed loop. Will it compile? Will it not? It can check itself, and there’s plenty of data obviously to train on. There’s very little, like, implicit knowledge. All the information by definition is codified in your, say, GitHub repo. And so I think the analogy here, just to bring this back, is it’s not just the model itself. The technology itself is obviously important in any paradigm shift, but it’s what are you building around it? The harnesses, the complementary technologies, and then what setting are you applying it to, whether it’s astronomy or biology. Sometimes certain fields are ripe for a revolution and others just are not the correct place to use the tool, right? And this goes back to the idea that to a hammer, everything’s a nail. So you have a cool AI model. Great. I’m just gonna use it on everything. That’s actually not the right approach. And I think for a lot of companies, to bring it back to the business setting, I think that’s the mistake that’s being made. A lot of companies are like, “All right, cool. We have AI now.” Like, it’s time to bump—pump up our stock price and talk about AI disruption and AI transformation. So I’m just gonna spend much, much money on buying AI. But they aren’t thinking enough about what are they actually trying—problems they’re trying to solve, and is this the right tool to solve those problems?

Ritavan

And, Kai, just to pick two words that you said. So it’s really about signal quality at the code bottleneck. I think that is the question, which is, are you measuring? Are you closing the loop, right, at the right place? And how well are you able to measure that? That is what you want, right? Because a tool in a system—you can get a better tool later on, or the capabilities of the tool might change, but the question is: what is it directed at, right? And I think that’s again the microscope versus telescope thing. Are you pointing it at, I don’t know, plant tissue, or are you pointing it at the stars, right, or at the moon? And I think that’s really the question. If you’re pointing it at the moon, it’s an objective, independent thing that you cannot change or shape, that everyone else can also point it to. The lighting is standardized, et cetera. So there’s just very high signal quality at the core bottleneck, and every time you develop a model, you can check and test a model. I mean, a causal model of the universe, you can check and test and get hard feedback of whether you’re right or wrong. And you can’t do that with a microscope initially. And I think that is—if we just sort of take that lesson to business and ask yourself: yes, okay, you deployed an AI tool successfully, or you adopted an AI tool, because I think the big dashboard KPI now is AI adoption, right? But adoption per se is meaningless, right? Because just because you’re using something—you could be using a microscope or you could be using a telescope, right? And it’s always the same technology again. So you can drive adoption to a hundred percent, except the telescope is creating value and the microscope isn’t. And so you wanna ask yourself, have you closed the loop? Do you have high quality signal at where the loop closes, and is that pointed at the core bottleneck? Because that is where you’re gonna capture value. Everything else is meaningless.

Kai

Right. This is kind of the whole pushback, the token maxing phenomenon, which was a brief flash in the pan where—

Ritavan

It’s already gone, yeah.

Kai

You had CEOs—

Ritavan

It’s gone.

Kai

I think that was—

Ritavan

It’s gone.

Kai

...like a one-month thing. But it was so, so crazy to me. It’s like, the metric should not be how much money are you spending on Anthropic. Like, that seems—

Ritavan

Yeah.

Kai

...crazy to me.

Ritavan

There’s a bit of Goodhart’s—I think it’s Goodhart’s law. What is it? Like, when a metric becomes the goal, then it stops being a good metric. And I think you have a bit of that too, right? Which is, be very, be very smart about what you pick as the metric, because if you just pick something stupid, you’re gonna get awful results. It’s just logical. There’s no—

Kai

Right. You’re incentivizing waste. Surprise.

Ritavan

Right. Exactly.

Kai

Okay. So our next question. I wanted to go to an analogy between two different historical companies. So first was Nokia, right? So they saw the smartphone coming, they documented it, and then sat on it. They did nothing. And then on the other hand, you had ASML, right? Which, as we all know, was very successful. So what was the difference in how these organizations were built that led to these widely varying outcomes?

Ritavan

So I think the most important thing is to say, again, I’m looking at this from the outside, right? So because you ask how do you see it from the outside, and so this whole thing is based on what I saw from the outside, because I was not inside either of these two companies. So let’s start with Nokia, because if you see press releases from the CEO, their kind of strategy documents and investor calls, et cetera, you see their focus is the following, which is, we want to be as close to where things are happening. We want to log, measure, document as soon as possible what’s happening, and then we will adapt. That’s broadly their pitch, which is the sort of lean, agile, whatever, just quickly change. Like, when things change, you try to change as fast, right? So you try to align your clock speed to change. And what happens there is then that they’re in Japan, they see what’s happening in Japan—and they see what’s happening later on with the iPhone. Except you can collect the best possible signal, but if you do not have a meaningful causal model of how you, in your system, meaning in your market with your customers, et cetera, et cetera—like, what is the game you’re gonna play with that signal, right? If you don’t have a causal model of that, if all you’re doing is reacting to someone else’s moves, right? Oh, the Japanese player did this, let me try to copy. Or the iPhone does that, let me try to copy. If that’s all you’re doing, you’re not really playing a game. You’re just—like, you’re aping what others are doing. That cannot be a winning strategy. And that’s kind of what Nokia did during that phase. And it’s pretty well documented. Like, you have all these interviews with the CEO, and they’re really selling this story of it’s all about speed, it’s all about agility, et cetera. And they were amazing at speed and agility. Like, if you see the way they’re reacting, commenting, and documenting what’s going on, it’s perfect. The point is you don’t have a causal model, right? It’s like if I see LeBron do something, I copy that. Then I see Mbappé do something, I copy that. What are you playing? Are you playing basketball? Are you playing soccer? Like, what are you trying to achieve? That’s, I think, the absurdity of agile, right? You can be very agile, very dynamic, very fast-moving, but it’s not going towards any meaningful goal based on some causal model of something, of a game you’re playing, right? Of a system you’re operating in. ASML, on the other hand, very interesting, because—and from the system gambit angle—because a lot of, like, you see this insane technology, right? Or, like, almost like a portfolio of technologies. Like, every machine is like hundreds of millions, and the most expensive ones around four hundred million or something like that. And you think, “Ah, okay. These guys, they’re heavy in R&D,” which they are. They’re producing a super advanced machine. Their core capability is manufacturing that machine and then selling it, right? That’s their core business model. I think what’s underappreciated is the fact that these machines are so sophisticated and subtle and sensitive that you can’t just, like, sell it to TSMC or sell it to Samsung or sell it to Intel, and then they just magically start producing. I mean, you’re not selling a kitchen knife or something, right? It’s just way more sophisticated and sensitive. And so what happens is the real capability that ASML has built is a causal model of how their particular machine in a particular fab operates on a particular design at a particular point in time with a particular set of environmental conditions. That knowledge, that institutional knowledge, that is what’s valuable, right? And that is unique. And that’s why—I mean, I think the underappreciated thing often in business is there is no easy way to put your understanding of your causal model, of your system, of the game you’re playing—you cannot put that on the balance sheet. Like, there is no clear way to financially map your understanding of the game and the system on a balance sheet or on an income statement or something. And I think that is why a lot of people aren’t incentivized to do it. And that’s unfortunate, because that is where the edge lies. It’s not in agility, it’s not in speed, it’s not just in moving fast, right? It’s about having an understanding of the game. And you see that in all games, right? I mean, there’s just this recent clip with Messi where he’s barely running, because you wanna be at the right—like, once you understand soccer well, you wanna be at the right place at the right time and do the right thing. It’s not about how much effort you put in. It’s not about how many steps you clock, right? Those are the rookie KPIs, right? And so you can ace the rookie KPIs. But the question is, do you wanna be doing that, right? And is that going to help you win?

Ritavan

No, build a causal model. I think the really big takeaway is, in every field, right? Whatever—if it’s sports, if it’s business, if you’re building an empire, if you’re trying to win a battle, whatever it is you do, right, outside of business—you’re always building a causal model, and the one with the better causal model has a better foundation to do all the other stuff. And I think this is a very underappreciated thing very often in business, both from an operating and from an investing perspective. And—

Ritavan

Yeah. And I think there’s a third angle where, if you had one of these short-sighted private equity players sort of acquire ASML, I think you could really cut a whole bunch of stuff and max out on EBIT for three months or for a quarter, perhaps a few quarters, and kill the company, right? I think that’s very much also a common playbook, is you just—the first thing when you see a golden goose is you slaughter the goose and you sell the flesh, right? That’s the first move that every pathetic operator really does, and it looks amazing for the quarter. The question is what happens after? And I think the smart move is, can you take a step back and figure out how you can get that goose to lay golden eggs and hopefully hatch some of those eggs to get more golden geese, right? That’s the game. And I think this is, to your question, right, Matt, is do not look at the financial value of what is being sold, right? Because if I sell the flesh of the golden goose, I might be able to get a much higher payoff than if I just sold one or two eggs from the golden goose, right? The question is, what is driving a certain gain or drop in financial outcomes? I think that matters so much more. Now, if we just take another example from agriculture or something, if you inherit or you buy a field which has crops on it, right? The first thing you can do is just cut all the crops and sell them right now. But then next season, you’re essentially gonna be broke. And so the question is much more, what is that compounding loop, right? Have you figured out a causal model of what your field and agriculture in that place can do? How do you max out on that? Where’s the hidden leverage, right? And what is your causal model that allows you to create value with that field, with the resources you have, but by being smart about it and doing things differently and not necessarily first just cutting and selling everything, which would be a typical one-quarter private equity play, or just doing what the farmer beside you is doing. And I think those are the two defaults where people go to, and those are problematic, because I think there is a whole range of other things that you can do, and it’s just intellectual, I would say, laziness not to explore that.

Kai

Yeah, I mean, I think that’s exactly right, that optimizing for short-term performance—whether as an operator or if you’re an investor looking for companies that have traditional quality metrics. Oh, yeah, we have high and stable EPS, right? Like, that can oftentimes miss the J curve, right? Where—so from my standpoint, intangible investments, this is kinda where a lot of my work focuses—they tend to be interesting because, and underappreciated by many investors, because they have this J curve profile where you spend the money on the R&D if you’re ASML, and it doesn’t actually lead to a payoff in the next, say, year or two. And in fact, may actually hurt your earnings because you’re now spending resources to invest. And it shows up over the next ten years and perhaps is what allows you to be such a great company. But I think for a lot of investors, they see that, they don’t like that. That’s like a negative attribute where it should really be positive. And I think what’s interesting about your work is you’re taking that one step further, and you’re saying it’s not just making an investment, it’s making an investment in finding the next—whatever the next paradigm or system is that can generate this compounding. So it’s almost like a super investment. And it all goes down to the same underlying concept, which is this J curve, or, like, trying to defer, defer the rewards—just the idea of investing.

Ritavan

Yeah. And in many ways, I think with ASML also, you would think it’s a production company, right? They’re producing these machines and selling them. In many ways, actually, what they’re doing is some form of contracted delivery, right? Basically, they’re guaranteeing your fab—like, if you’re a TSMC or an Intel, they’re gonna guarantee that you would be able to produce this many chips per hour at this level of quality, because it’s their engineers embedded on the ground to make sure their machine can deliver that. And I think that is something that—I mean, it’s irrespective of how much, let’s say, they get paid in terms of service fee to do that. I think financially, irrespective of that, just the fact that they’ve got boots on the ground and know what’s happening in the fab, that knowledge institutionally is impossible to replicate. There’s no way—

Kai

And we’re seeing a resurgence in the forward deployed engineer these days, and I feel like that’s the same model, right?

Ritavan

Correct. It’s not just a product. But ASML did it probably fifty, sixty, whatever years before Palantir, and they didn’t spin that good a story. I think that—I mean, they’re not necessarily dominating the narrative, I think. And that’s kind of the point, is, if you only look at the current financial statement number of something now, that just misses the opportunity that the system gambit, I think, gives you.

Ritavan

Yeah, not at all. So I think it started out—like, my previous book, Data Impact, had a six-step framework, and the second step in the framework was the word leverage, right? And a lot of readers came back to me and said, “You got us excited. You’ve now got us hungry. You served the starter. Now where’s the main course?” And I wasn’t expecting that kind of reaction. But that really kind of irked me. I had this itch then, and I had to scratch, because I felt I did not have a deep enough answer that I was proud of. Like, what I wrote in Data Impact in the leverage chapter was, hey, focus on your unique proprietary strengths. Focus on your physical, non-digital assets. Focus on your brand, on your reputation, on a whole bunch of things that typical business books also have. And I said, “Hey, you bring these together, you know, tangible and intangible. You combine them, you leverage them, and you create value.” But I think anyone could have written that. I felt like there wasn’t that unique take on this, and the question was, what is the mechanism? Because I can have all these assets—like, you can picture two companies who have the same strong brand, who can have the same intangible other assets, who can have the physical footprint of stores or whatever, right? So they can have all of this and yet you can get two very different outcomes over ten years, right? So the question is, what is the mechanism that allows you to take all of this and leverage it successfully? And from that starting point, I think it was really a journey of just curiosity and looking around and not shutting stuff out. So, like, not digging into and within just the sort of narrow business literature or field. But to say, “Hey, if something has a certain universal characteristic, like, can I find an underlying mechanism that appears everywhere, right? Broadly.” Then you know that you unlocked something that’s meaningful, something that’s durable, and something that has a certain truth to it, right? That it’s not just, oh, I kind of overfit these three data points type of thing. Because I think that’s easy to do, and a lot of playbooks and checklists and a lot of that is typically this, and I think you can sell and hustle those type of things short term. But I think if you want to really unlock a hidden structure, you just have to look more widely. And, you know, one thing just kept leading to another. As I traveled, met people, saw stuff, I just—every time I would take a step back and say, “What is the underlying mechanism here? What is the system? How is value being created or not? And where is the asymmetry? Where is someone able to do more with less resources?” And then you just keep seeing these patterns all over. And then you just have to kind of order your thoughts, find commonalities in those patterns, extract these principles, and then just package it into a book. And that’s broadly, I think, the work I’ve done.

Ritavan

Yes, I think the core point is the following, which is that in many ways, AI is a general purpose technology, but for the first time, it’s a general purpose technology that you can bolt onto whatever your system is, right? So whatever paradigm you’re operating in. So if you’re an industrial business, then you sort of bolt it onto your industrial paradigm, and you build dashboards, and you try to predict and measure stuff here and there, right? If you’re a digital company, meaning like a software company, then you apply it within that in terms of how can I validate a use case faster, et cetera. If you’re a platform business, you apply that to increase interactions, to tweak your recommendation algorithms and a bunch of stuff, right? And so the core idea is the following: because the data that you have is generated through your existing business, through your existing system, that data by definition is within-system, within-paradigm data. If you bolt AI onto that, you’re essentially training within paradigm, within system. And the core idea of the system gambit, and especially the first system gambit, is about paradigm change and goal displacement. That is where there’s opportunity, right? Because if you stay within the same system, you’re again playing the resource game, right? Who has more resources, and then you know who has more resources. If you want to get asymmetry, you have to see differently. It’s the same system physically perhaps, but your take on that, the paradigm you choose to play in, right, the game you choose to play in, changes the outcomes. And I think the biggest danger with AI and AI adoption today is if you do it in a naive way, if you do it in a way that anyone with a credit card and an API key can copy you, right, then where is your edge? And more importantly, you’ve just increased your cost without any knowledge or without any understanding of how you’re creating more value. And this within-paradigm optimization is exactly how you optimize yourself to irrelevance. And especially if you’re one of the big, large players, you need to be extra careful, because it’s usually the underdogs that find the leverage and that eat your lunch.

Kai

So then what is the system gambit in this case? You’re saying what it’s not. It’s not stay within this existing paradigm, slightly optimize your business using AI. If it’s not that, what is it?

Ritavan

So it is defensive, right? So if you see someone do that and you feel okay—for some, they have already optimized within paradigm a little bit and, I don’t know, increased EBIT by one percent sustainably—then you can be a follower there, right? Don’t try to lead that trend. Just see what people are doing, and then you just follow, because essentially you are letting them de-risk whether something works or not. And that’s actually why I call the system anti-gambit. So if you only do this, right, then you’re kind of doing the opposite of what the gambit should be about. In a system gambit—and again, there are eight system gambits in the book, it goes one by one through each of them—it gives you essentially, like—if we go into fitness and health, I’m not proving anything. That’s why you could always say, “Oh, you’re using mainly examples that work,” and so you’re not really proving the point, et cetera. But I’m thinking at it much more from a practical perspective. I’m a bit like a fitness trainer giving you a bunch of reps you can do and variations on the exercise. So each gambit, think of that as like a composite muscle group exercise, and I’m basically saying, “Hey, you can do the squat this way. You can split your legs a bit more. You can squat on one leg, or, like, stretch this leg or whatever.” So I’m giving you a bunch of things in each chapter. You just repeat and get the reps in. And the idea is that you start then pattern matching stuff in a way that AI cannot, because AI is trained on granular data within your paradigm. And so the idea is I’m giving you a way to concretely, practically look for patterns based on a vast and diverse range of examples across all of the eight chapters. And with that, you can then look at your specific context, right? Your asymmetric proprietary advantage, et cetera, and ask yourself, “Which of these system gambits is for me now relevant, right? And how could I execute it?” Right? Or if you’re an investor, who out there, seeing from the outside, seems to be executing it? Because the thing with the J curve is, right, that’s just a descriptive artifact of what’s going on. The question is who is seeing just a drop in performance, or, like, a value trap, I guess, in the investor jargon, right? Who is seeing just some kind of drop and it stays there, and who’s going to unlock a compounding loop, right? So who’s building a structural condition that compounds? And then that is the core contribution here, is once you go through the eight system gambits, you have eight concrete things, and you’ve done the reps by the time you’re done with the book, so you start seeing these patterns.

Kai

Yeah. So a lot of the work I’ve done as a quantitative investor has been trying to understand, like, base rates. So for example, if there is a disruption occurring in e-commerce, and you have all these brick-and-mortar retailers, and some of them are trying to become Walmart and successfully navigate this, but most fail—like, what’s the success rate in that situation? And then the second question, of course, being less about the fundamentals, more about what’s priced into the market. So as an investor, you pay the stock price that’s available to you at the time. And so if stock prices are down, say in software stocks today, what’s the chance that they recover versus not? And do we see more dispersion? And historically, the answer would be yes, where many of these companies are zeros and others end up being great generational buys, where you buy a stock at a P/E ratio of eight that goes on to become the next Walmart in its class, right? So the answer is a few things, historically, right? So obviously it’s the use of the technology itself, right? Are they adapting to the new paradigm? But it goes beyond that. I think, you mentioned actually some examples of, like, proprietary data. Obviously, AI is only as useful as the data you train it on. It’s garbage in, garbage out. And so to the extent an enterprise and incumbent has a lot of interesting customer data, or whatever is applicable to its domain, that could potentially be a very valuable asset in the age of AI. And you kind of go down the list, and it does turn out that, like, complementary assets have continued to, at least historically, hold value. And of course, things change, right? Moats, assets, brands may be valuable one day and less valuable in a different context. But on average, shaken across the thousands of stocks over decades, these complementary assets, generally intangible but also tangible as well, have tended to be decent. I mean, I guess one question I would spin back to you is around—it is—you know, you frame it as kind of binary. There’s gambits and then anti-gambits. Is there a middle category where it’s like, “Hey, look, you’re an industrial business. You have a great business. You’re a market leader in whatever category you happen to compete in, and now AI is here,” right? Like, maybe it’s fine for them to just say, “We’re gonna take incremental advantages of AI without trying to completely reorient our business.” It’s fine. It is a sustaining innovation. Like, in the internet example I just gave, if you’re a retailer, then yeah, you definitely need to figure out e-commerce, ‘cause if you don’t, you’re cooked. But there are plenty of businesses that just used email to be a little bit more productive. Is it fine to be in that final category? Like, should everyone be trying to execute a system gambit at the same time, knowing that not everyone will be successful in doing so? Or is it fine to kind of just not even play that game?

Ritavan

So since we spoke about retail and Walmart, I’ll just lay out a thesis, and I think that’ll answer the question, hopefully—perhaps not with one sentence, but with a lot of context. So if you go back in the financial press, like, about a dozen years ago or so, or two decades or so ago, you had initially the phenomenon that Amazon, like the e-commerce business, was not making any money, right? And so you have people writing, “Oh, this is one of these dot-com dinosaurs, and I don’t know why they’re not going bust. And it’s not really a business, and what are they doing? They’re selling books now. They’re selling everything under the planet. What nonsense. That’s not a business model,” et cetera. What Amazon is doing in the background is actually executing the system gambit in multiple paradigms, right? So Jeff Bezos speaks about the flywheel, and I forget who the author was, but there was this idea of the fly—

Ritavan

Jim Collins, exactly. So Jim Collins, I think, held a workshop in the late nineties, and then Jeff Bezos really liked that and said, “Everything we do has to be based on a flywheel.” And a flywheel is basically compounding, right? So it’s like every loop, every iteration gets better. And so the flywheel is essentially a compounding loop. And what Jeff Bezos did, obviously, is then to say, “Okay, everything we do has that feature,” which means you have to sacrifice short term on the metrics, metrics like profitability, to build that compounding loop. And so essentially, they executed probably the greatest system gambit of the early two thousands, and in many ways until now. But what they’re doing is operating on three paradigms. So they’re, on one hand, building out their warehousing infrastructure, which is an industrial style business. Which is you want as much predictability as possible. You’re moving physical inventory around, you need that physical footprint for warehouses. On the other hand, instead of building out the physical footprint, like stores, to sell the stuff, what they do instead is to operate in the digital paradigm, right? So it’s the Amazon online store, right? So you go and order there. That is driven on learning loops, right? So you want to collect as much data, user data, as possible to understand what users want, when they order what, et cetera. So it allows you to price things dynamically, et cetera. It allows you to offer discounts. It allows you to recommend products better. And now comes the interesting part. So you’ve built an industrial warehousing, logistics business operating on data and algorithms. You’ve built an online storefront operating on data and algorithms. That in itself is cool, because that’s two flywheels running. Now comes the cross-system reinforcing feedback loops. Because I understand user behavior really well, I can actually start shaping it, right? So now if I have a product that’s on inventory in my warehouse, I can discount it a little more aggressively, right, to sell that product off and to improve my inventory levels. If I have a shortage of a product, I can jack up prices. And so now, by building a cross-system feedback loop, right? So you’re taking two paradigms in which there is already a feedback loop running, a compounding loop, and now you’re crossing them. And then comes the platform paradigm, where you—with Fulfilled by Amazon, with FBA. So now you open it to all sellers, right? And so you’re now saying, “I already have a physical footprint. I own the customer on one hand. Now, why should I produce and sell my own stuff? Let me just offer that infrastructure to all these people who want to sell their stuff.” Except you have leverage on them, because they can’t just leave, right? Once they commit to using your stuff, they are playing the rule on your terms, and you can see what products are selling better. And then with basics, you can essentially put them out of business, which Amazon actually did. Now, let’s leave out the ethics and whatever you think about that. It’s just worth understanding the insane leverage they unlocked by building within-paradigm compounding loops and then crossing them and, you know, making them cross across paradigms and compound. And with Fulfilled by Amazon, now you understand user behavior. You’re selling someone else’s products. You’re not limited to your own products, right? So that’s the platform model. You take your existing customers, your existing physical assets, and you open it up to any seller. And then you add Prime, and Prime is the layer that kind of connects all of these, right? So with Prime, you essentially tie all these three things together. You maintain leverage over everyone. So you maintain leverage over your customers, because you know so much about them, et cetera. Like, if they go to someone else, they won’t get as good recommendations. You have leverage over your sellers. And that is why no player out there, like a digital startup that is super fast at shipping software or at building an online storefront, et cetera, they are not able to really ever attack and break Amazon, because Amazon is a three-paradigm beast. A traditional business that is a strong logistics business that says, “Okay, I’m gonna knock these guys out of the market,” they can’t do that, because it’s, again, a three-paradigm compounding beast. And I think that is something that’s underappreciated. And I think for the first time, in The System Gambit, I really felt I understood why Amazon is so unbeatable. And now comes the Walmart part, right? So as long as Walmart was trying to play the Amazon game—so from the early 2000s, late ‘90s, early 2000s to 2010, Amazon is the joker, the loser, the no-profit business. And then people finally caught on to that and realized, “Oh shit, these guys were executing a multi-paradigm system gambit, built this massive compounding machine.” And what is sacrificing a five percent EBIT for ten years, right? It’s meaningless, right? In terms of the amount of value they built, sacrificing that EBIT for ten years is chicken shit. Once they have that, then suddenly the same narrative, the same people now say, “Oh, now Walmart is dead,” right? So Walmart’s running successfully. While Amazon was not profitable, Walmart was putting out good profits, and so Walmart was the good guys, Amazon’s gonna go bust. Then they realize, oh no, Amazon has become a beast, now Walmart will go bust. Which is also an absurd story, because anyway. So Walmart now has these physical stores, right? As long as you say Walmart now has to become the new Amazon, meaning they have to sell everything online, that’s an absurd proposition, because Walmart needs to focus on their asymmetric proprietary advantages, right? What is it that they can leverage? And obviously what they will leverage will be different from Amazon. And so, over about ten years, Walmart is kind of aping on the outside what Amazon does. “Okay, we will use some digital systems here and there. We’ll get a better CRM. We’ll do this, that”—and these kind of small defensive tweaks, what I call the system anti-gambit. It’s not a bad thing necessarily. It’s just the rational thing to do in the existing system. And then about half a decade or so ago, you have Seth Dallaire, who’s an ex-Amazon exec, moved to Walmart. I think he joined as chief growth officer. I don’t remember. And he was, before that, for a while at Instacart. And now, over the last half a dozen years or so, what Walmart has done is they said, “What is the one thing we have that Amazon will never have despite having bought Whole Foods?” And that’s the physical presence. We know how our customer not just shops online on walmart.com or whatever. We see them physically shop. That’s a proprietary strength that we have that Amazon can never have by design, structurally. Unless Amazon suddenly starts opening physical stores, which would completely destroy them. So that’s management logic antagonism, right? And the second thing here is path dependence. A shopper that has shopped in Walmart over two or three generations, that’s a level of path dependence that you’ve built, a level of trust, a level of brand, a level of customer understanding that you’ve built about a customer and a community, a local community, that Amazon just cannot have by design. So you have path dependence, right? And you have the self-improvement loop, which now needs to be built, right? So two of these were already there. And then what they did is, okay, we will now take what we understand of our customer online. We’ll take what we understand of our customers uniquely physically. We’ll put this together, and now we try to delight our customer. And with that, they are now on such an amazing trajectory, and they went multi-paradigm, meaning you can buy online, you buy in store. We have a loyalty program. That’s the platform model, right? So you’re taking an industrial business with the physical stores. You’re leveraging that to build your digital capabilities, and then you open it up through partnerships, et cetera, with a loyalty program. And now, again, you’ve built an insane compounding beast. So it does not matter where you start. You could be a pre-dot-com bubble tech startup. You could be an established family-owned, a hundred or whatever, or half a century old business. It does not matter who you are and where you come from. What matters is, are you willing to think from first principles, from a systems thinking perspective? And are you willing to unlock leverage and do something that is different from the crowd, that is outside the scope of the paradigm in which you’re operating, in a smart way? So it’s not just, oh, let’s just wake up in the morning and brainstorm ten ideas and then randomly do three. It’s, can we rigorously think from a systems perspective, understand the mechanisms, and really map it out logically as a causal model of what we’re trying to do? And if we go wrong, or if a specific thing doesn’t work out, then you update the causal model. So you’re not just madly shooting—you’re not just throwing stuff at the wall and seeing what sticks. You have a causal model, and then you’re updating or modifying it based on empirical feedback, and so you’re closing the loop.

Ritavan

It’s always hard to distill it into one lesson, but I think the real opportunity at any given level of assets, at any given level of size, is the question: what game are you choosing to play? And The System Gambit allows you to think that through from first principles in a way that allows you to unlock that leverage. And it could also be on a time axis, right? So if everyone is trying to optimize the next quarter and play that game, the question is, do you want to play on a different time axis, right? And so concretely, you can map it on a bunch of things. It can be based on the assets you already own, physical or intangible. It can be based on time. It can be based on a bunch of stuff, and that’s why it’s contextual to you. And The System Gambit, the book, is really just like a bunch of workouts and reps that, once you get in, you can choose to play the game you want, right? You’re just a healthier, better, smarter person, and then you play the game you are uniquely capable of playing.

Ritavan

So the book is The System Gambit, so that’s the system crossover from blue to black. The book is on Amazon, and you can go to thesystemgambit.substack.com. That’s where I share stuff. Or go to the YouTube channel called The System Gambit for more. And otherwise, yeah, just follow on LinkedIn and engage on LinkedIn.