Now Is the Best Time to Become a Junior Analyst - Ex-Citadel and D. E. Shaw PM Brett Caughran

What makes a great, invested process? The core part of investment process is really about driving differentiation on those ski drivers. My general priors that alpha lives in the tails, 80% of stocks are mostly fairly priced, but there's a 10% tail on the right, 10% tail on the left where there's meaningful mispricing. How do you figure out how names will move as a reaction to potential news that you forecast coming out? Stock picking is step one, analyze the business and step two, analyze the stock. Public market investing is a deeply busy and approach in the sense that I form my initial belief and then I'm updating that prior every day. I think about AI as an intellectual power tool. The speed to developing a consensus view is massively sped up by using these tools, but I would point out that the consensus view in public market investing is a losing view. Do you think that junior analysts will still be needed in the future? There's almost no better time to be starting a career as a fundamental investor because you have the confidence of these tools for junior investors that allow them to get to the juicy part of investment process more quickly. Developing differentiated perspective is very very hard. There's a key distinction in being a great analyst and being a great investor. The great investment firms have inescapably a marriage of good process and good judgment. I'd say in general applying the outputs of that process are where the real alpha lies. Brett, thank you so much for coming on the pod. Thanks for having me. I'm excited to be here. What makes a great investing process? It's a deep question. I'd say sort of invert it and say a great investment process is something that works. So sort of start at the end and I think that matters sort of a different question for different investment strategies. What I would say in general is that investment process has a couple big buckets to it. One is a process to develop a comprehensive understanding of the business. What does a business do? How does it make money? What are the key drivers? I sort of think about this as a we use acronym of E-TIC. Everything there is to know. The mandate when I was a junior analyst at a hedge fund was to know the companies I covered better than any other non-insider. So this level of deep comprehensive understanding of really all of the elements of the business. Now that doesn't mean you want to get stuck missing the force for the trees. I'd say the second part is the critical pivot of the key drivers. I think most investors, most great investors, a tangle with that complexity, tangle with that comprehensive understanding of a business, but are able to pivot to identifying the reality that there are two or three variables that ultimately will determine investment success or failure, right? Those key drivers. And then the core part of investment process is really about driving differentiation on those key drivers, right? How do I develop a very perception, a differentiated view on those key drivers? And that's where the game gets fun. That's where creativity matters. It's not just reading the 10K or listening to the earnings call. It could be a host of different tools that align with the investment questions. Stocks ask different questions at different prices. And so our job is to sort of tease out and understand the question that a stock is asking, calibrate an investment process around those specific debates, and then go develop a differentiated perspective. That's sort of one other framework I quite like is sort of thinking about the concept of the alphas in the tails. Like markets are sort of the question of market efficiency is a complicated one. My general priors that alpha lives in the tails, 80% of stocks are mostly fairly priced, but that there's a 10% tail on the right, 10% tail on the left where there's meaningful mispricing. So how do we understand identify that mispricing and then go invalidate or invalidate that hypothesis with investment process? So if you sort of decompose alpha or the outperformance of markets versus the market, sort of the decomposition of generating alpha is finding a differentiated perspective and investing behind the differentiated perspective with conviction. So there are three things you mentioned there. Understanding the list extremely well, key drivers, and then developing the differentiated view. For the second point, determining key drivers, I guess I have two questions about that and forgive me if it's difficult to answer both. But firstly, how do you determine what are truly the important drivers of those names? And then I guess as an extension from that, how do you determine the drivers that other people aren't seeing? Yeah, it's a great question. And the challenge in identifying key drivers is the debates shift over time. Often key drivers can be almost purely financial, revenues, margins, profits. Many times they can be a function of the narrative cycle. If you think about decomposing the present value of a stock, if you sort of play around with the DCF, you can sort of understand that your average 15 to 25 P.E. stock has a duration of 30 to 50 years. What does that mean? It means to get to today's current price. I need to discount 40 years of free cash flows back to justify current pricing. And so what matters this quarter or next quarter on fundamentals is informative of that. But narrative cycles matter a lot, right? How will Uber fare in an autonomous vehicle future? That's a key question. How will SASTOX fare in an agentic future? And so key drivers can both be purely financial. I think about a frame where we call Focus 5, where it's organic revenue, growth, margin, sort of trajectory, margins, capital intensity, capital deployment, and terminal value visibility. Those sort of five buttons, if you will, are the key drivers of value creation, sort of value assessment of stocks. I think in general, it's a combination of how is the business performing? What's happening on organic revenue, growth rates, trajectory of profitability, capital deployment, capital intensity. But also the trickier and more elusive key driver can be how will news flow impact the collective market reaction on these stocks? You've seen this sort of recently in SASTOX. Many of these stocks are 50, 60, 70 times earnings. They get reprised to 30 times earnings. And it feels like that's a big move down. But if you sort of decompose those now lower multiples, you're still in the market is still still implying these businesses not only maintain their cash flows, but grow their cash flows for decades. I'm not sort of saying one way or the other, but I think understanding the market price mechanism is an important step one in calibrating your investment process around that. If you're given a situation where, let's say this has never happened before, how do you figure out how names will move as a reaction to potential news that you forecast coming out? Yeah, that's a challenge. There's a little bit of sorcery associated in that. Paul N. Wright is a big fan of a framework which I've borrowed from him of stock picking is step one, analyze the business and step two, analyze the stock. And so when you analyze the business, it's going through the financial model, looking the revenue profits, profits, cash flows. But stock, the stock is a piece of paper effectively, or three digits on a screen, ticker on a screen. The price of a stock is what market participants determine it to be. And so the challenge, the challenge in analyzing a stock is understanding the second order of fact of how investors will behave based on news flow. And that will shift, that will shift over time. Markets go through regimes, investors become optimistic, they become pessimistic, they become concerned about a certain variable, they become unconcerned about a certain variable. So some of that is sort of this constant handicapping and awareness of how markets are behaving right now. Number two, it's having a deep lattice work of priors, understanding how different stocks have traded at different times on different competitive threats. And then it's sort of building your own foggy crystal ball of how you think certain sectors and companies will evolve and having your own view on these different debates and being right and being early. That's the sort of challenge in investing. Investing is not just a one shot, create an investment thesis, tuck it away for three years. Public market investing is a deeply Bayesian approach in the sense that I form my initial belief and then I'm updating that prior every day, every day, week, month with the blocking and tackling of analyzing market behavior, analyzing what the conveyances out of companies, analyzing data. And so public market investors will change their, the best investors will change their mind as a facts change and try to see how stocks behave in different regimes. There's various times, for example, when concerns about blockchain were leading to sell-offs and visa mastercard, those were viable dips. There were times when I was a healthcare investor where Amazon entering news, healthcare sub-sectors led to pretty viable dips. And so the trick is understanding when to freak out and when not to freak out, when, when to buy the dip and when to run. There are also, you know, a number of, you know, sectors that have had real technological destruction, right, where the first miss wasn't the last miss, the first dip wasn't the last dip. And so running into a burning building can, can be great if the fire goes out, but if the house burns down, that's a problem. And so that's the trick. That's the judgment that is really built over, over years and decades of, of approach and listen, investing is a deeply probabilistic game. So you're not going to get every one of those, right? I think, I mean, you touched on this theory. I mean, you mentioned it. And I think the most important thing that you were referring, what you spoke about was building up your own magic crystal ball. That is your own personal judgment. I find that in the age of AI, right, there is such an ease of access to information. And it's so easy to outsource judgment and to outsource knowledge to, to the models. And then going back to trying to build up a differentiated view. I think that is probably the worst ways, one of the worst ways to build up a differentiated view because it's the illusion of knowledge and of deep research. Today, given that we have all these tools, how do you think investors should build up that crystal ball, that knowledge, that expertise that great investors have? Yeah, no, it's a great and very topical, great and very topical question. And I would say, to sort of frame it at a high level, I think about AI as an intellectual power tool, right? It can really accelerate what you're doing. I also wouldn't give my 12 year old son a table saw, right, for fear he may cut off his cut off his finger. And so I think the careful deployment of AI is a very important skill to build for today's investors. I'll give you a couple of couple thoughts. Like understanding the foundation of what artificial intelligence is matters. And when people say AI today, they're primarily talking about large language models. Large language models are next token prediction engines trained on primarily a web scrape and academic articles and books, which is great. They're sort of accessing almost all of written knowledge. But in general, when applied to investment context, you're capturing the consensus view quite, quite, quite, quite quickly. So the speed to developing a consensus view is massively sped up by using these tools. But I would point out that the consensus view in public market investing is a losing view. If you look at the percentage of long only investors after fees that I'll perform the market over time, it's not a great story, 70% underperform. If you look at the T plus one price performance of sell side recommendations buys and sells, there's no discernible alpha based on various academic reports. And so if I can get a very fast thesis from AI, that is mediocre or worse, AI slides a lot. I can't sort of, there's no intellectual grounding for me to think that that has alpha embedded in the view. So I'd say we sort of break good investing down to two buckets. We break it down to investment process, which has some creativity, but it's a little bit more of a scientific approach. Here's 120 hours. This is how you spend your time, read the model, read the filings, build the model, make the calls, do the research, do your checks, et cetera. The second and more challenging part is judgment. Judgment sits on the top of that process. Process feeds up to judgment. Judgment's a little bit more handicapping the situation. It's a little bit more of analyzing the stock. Based on what we've learned about the data is this situation, a positive risk reward. I would say that investing is such a competitive game that there are no layups in this business. At best, we're trying to flip 60/40 coins. We're trying to find stocks that will go up 30% in a bull case and down 20% in a bear case. If the odds are 50/50, I found some sort of misprice edge in that investment thesis. I can populate a portfolio of these sort of weighted coins. Over time, that should drive alpha-forward in the portfolio. Can AI do that natively, not in my opinion? AI can accelerate parts of the process, but the ultimate judgment is a deeply human endeavor, in my opinion. AI can mask judgment, but AI has no innate sense of the world, has no innate sense of causal relationships, has no ability to pattern recognize unique N of 1 situations, which sort of the world is dominated by N of 1 situations. There's sort of big gaps that I think are best left to humans. I think the evidence of quantum-earned investing over decades has pointed to an irreducible benefit of humans in the investment process. What parts of the investment process do you think AI can currently do well, and what parts do you think AI will be able to do well going forward? One obvious place that's a couple of these places that are available today. I think about the sort of life cycle of an idea in sort of three buckets. I think about it as one hunch, which is basically a low stakes, wondering whether something is interesting. If that's sort of sufficiently interesting, I'll upgrade that to a hypothesis. Let me sort of build a testable proposition on this idea. Then ultimately you upgrade that to a thesis. It's like, "Okay, I've done the work. I've identified my 6040 coin, my 2-to-1 risk reward here. This goes into the portfolio." To me, AI is a wonderful tool for accelerating the hunch and hypothesis part of the investment process. When I'm looking at a new idea, you might take sort of classically 6, 8, 10, 12 hours of just sort of like sniff test workflow. Let me smell this thing and see if something's interesting. Let me read the 10K, a couple notes, a couple earnings calls. Maybe talk to a sales side analyst. Just to sort of look around motion to see if I should spend my time here. Sort of filler-kill mentality. I don't want to jump right into building a model for 12 hours that I, if something's not interesting. Particularly true in a generalist mindset. If that's interesting, let me sort of frame the bold base, bear. Let me frame the key drivers. Let me frame the core debate. Let me sort of sketch out a process workflow that I can follow to go develop that differentiation. So classically, that might be the first 5, 10, 15 hours of an investment process. Then thesis work is like, let me go do the work. Let me go do the desktop work, the primary research, and really drive that insight formation with a mosaic approach. AI is really, really great at that hypothesis formation. Why? Well, AI can process at incredible speed with questionable accuracy. But I don't need 100% accuracy for a hypothesis. I will go downstream and verify and validate all of those outputs. And so think about this way. If I want to find three new ideas for my portfolio, I can go look at 60 ideas rapidly with an AI augmented sniff test. Get that down to 10 ideas. Sort of disqualify 50 out with this process. And then go turn over the stones in an analog way. So it's really the marrying of the AI augmented and the analog approach that matters. So I think that's where the tools have been. I think that's where for me, the primary value has been in the chatbot world because of these fundamental weaknesses like hallucination and data integrations. Now I would say that's changing quite rapidly. Agents have taken a huge step forward just. in the last three or four months, the biggest change is that LLMs can now orchestrate tools in a very effective way. And so the capabilities are expanding in a very rapid pace. The ability to build an update model is like an emergent skill that's like happened in the last few weeks in my tests. The ability to create agentic workflows and ultimately deploy a swarm of agents to the investment process, I think is something that even at the end of last year was very conceptual, but is now becoming more of a potential reality in an investment process. So that's very exciting and happy to elaborate on other ideas with that. Yeah, I guess, as you mentioned, they're progressing quite fast. One can only speculate whether they're going to be a year from now, two years, five years. And I guess if we're imagining that situation, do you think that junior analysts will still be needed in the future? It's a really good and a really timely question. I tend to think, yes, while sort of mentally considering a range of sort of array of possibilities. I'll sort of give you a few prior examples. One that comes in mind, when I started as a junior analyst, I spent a lot of my time spreading consensus. So if I covered a name, I would have to go identify the 18 models from the 18 cell side firms, compile those cell side firms, KPI by KPI, and put all that data into one spreadsheet. So that we could see gross margin consensus, segment level consensus. That might take me a couple hours per name, and I did a lot of those. But there was value to sort of seeing where we were different in our models relative to consensus. That's classic junior, grunt work. The fast forward five years, visible alpha came out and that visible alpha is a tool that does that on the back end. And so no one's hand spreads consensus anymore. That's great news for junior analysts, because that was incredibly sort of mind numbing work that arguably didn't really provide any training, training value. That's sort of a prior that I apply to my mind in terms of grunt level work that can now be automated. I think agentic tools accelerate that ability. So what does that mean? If you're a junior analyst and you're spending 40% of your time helping your PM solve those bottlenecks of model updates and data analysis and meeting preparation and earnings preps. And these tools can be automated. Now what does that mean? Does that mean that your PM can let the analysts go or hire fewer analysts? I think potentially yes. Now I think ultimately you'll see a forking here where we call the long tail of sort of under-resourced funds, smaller funds, smaller long-only family offices, etc. We'll be able to do more with less. One or two people on investment team will be able to cover more stocks with more rigor with these tools. And that's happening today. My sense or my belief from maybe my hypothesis is that the institutional level sort of billion dollar plus AUM funds is that it won't be a head count reduction play. And AI will come in as a tool to deepen rigor. Right? Ultimately we're playing a giant game of poker where there's sort of this competitive adaptation of alpha pools. And the pool has always been similar. If my competitors are adapting with new tools and new information and I'm not, I'm slower to that move. And so we can sort of walk through the economics. And the economics of letting a junior analyst go at a large institutional manager, like generally not a great ROI if that junior analyst can help drive alpha. Now I think, and this is a good news for junior analysts, I think that the role of the junior analyst will shift a bit. It will shift from updating models and doing data work to being that human in the loop in the investment process. If you go back to the, we sort of think about three layers of the investment process, there's desktop research, anything I can do in front of a screen, right? Read a 10K, update a model, do data work. That sort of commoditized like no, no, no self-respecting PM will run a portfolio just on desktop research. Because ultimately as we talked about earlier, alpha is a functional variant perspective. It's very hard for me to read a 10K and have a differentiated view that I can invest behind with conviction, right? So a lot of conviction comes from primary research. It comes from the relationships I build with management and IR and consultants and competitors. It comes from this process of connecting the dots from desktop research to calls, to conferences, trade shows, and building a mosaic of insight. So I think what you will see with junior analysts is that the junior analyst resource will be shifted more towards a primary research role. That's how I would do it if I was building an investment organization right now. I would work to aggressively automate more of the vote tasks. Now validation is incredibly important. The one shot approach of building a model or doing data work is going to be about 85% accurate. Right? 85% accurate on Wall Street will get you 100% fired. And so building and validation systems, which are both systematic AI, validation systems, which are shockingly good now, but also some human validation at the end. And AI is actually good at building human validation checklists. So I say one is building the system to free up hours. We all have the same 3,000 hours of time to work in a given year. If I can free up 600 hours for myself, 1500 hours for my junior analyst by automating more of these processes, collectively we have over 2,000 hours to deploy. To other value added parts of the investment process. All of a sudden I can justify encouraging the analyst to fly to Chicago to a hospital CFO conference for three days rather than spending his or her time back in the office updating models because earnings is three weeks away. And so the redeploying of that save time I think is the ultimate sort of, you know, sort of big key opportunity for these tools to go deeper and build more, build, build more, build more rigor. I think something that's often spoken about is how all these tools will help the markets will translate in the markets becoming more efficient. Are there any inefficiencies that you can see that will come up as a result of these tools? I would very much challenge the premise that these tools will lead to more efficiency for a couple of reasons. Number one, I think sort of, you know, we've seen a long slow death of informational edge in markets. When I started in the business, I would go and call 30 Wendy's franchisees and ask them systematically every month how same source sales were evolving, sort of build that into a forecast and generate an edge on the print. So there was this informational edge associated and investing. Well today alternative data tools are effectively now casting same source sales via credit card panels and that edge is shifted. And so the democratization of information and analysis that has been supported for decades by tools like Bloomberg and Excel and alternative data and real time information has led to sort of the compression of this idea that I know something that others don't know about this company. That's sort of very rare. Sort of if you go back to Rage FD in the late 90s and sort of the sort of incredibly welcome crackdown on insider trading etc. Like most investors aren't going to know something with certainty about a KPI for a company. Does that mean that alpha has died? In fact, the opposite is true. Many of the investors I respect sort of push the view that markets are more inefficient than ever. Meaning that the alpha and the tape is sort of wider than ever. I think there's a way to empirically measure this by looking at what the multi-manager alpha machines are generating in dollar P and L. these funds have gone from 10 billion. to 60 billion plus or minus an AUM and they've sustained double digit returns. So they've scaled dollar alpha into these markets. And so I think the prior is that more information is not compressed alpha. Why is that? Well, behavior alpha sort of continued to be a sort of meaningful pool. And this pool of sort of integrated perception, like I'm going to do all this mosaic work and have a handicap of view on the business that is different than the market's employees are alive and well. What's supporting that is the shifting of the players of the poker table, market micro structure changes, the death of sort of the hollowing out of the long-term value investor community, the rise of quants, pods, indexers, and factor investors, have led to real observed constraints on market price discovery, such that stocks whip around in ways that seem irrational. Now as a fundamental investor looking for an X-anti alpha opportunity, an expectations gap, that's wonderful news. If I think a stock is worth 20 today going to 40 in three years and it whips around like crazy, that's an opportunity for me. If I can buy it at 15 or 10 or 8, right? Now I need to have the backbone, the patience, the capital base to do that. But volatile markets are a gift to fundamental investors. So my belief is that there's almost no better time to be starting a career as a fundamental investor because you have the confluence of these tools for junior investors that allow them to get to the juicy part of investment process more quickly, not be just data monkeys in the first year. With markets that from a setup perspective, I think are highly inefficient, dominated by non-fundamental investors, dominated by indexes. And at some point the index will have a sort of a financial gravity problem. The math will make seven compound to get 20% for another 20 years. It's just almost mathematically impossible. So I think the setup is wonderful. And if you sort of go back to the specific question you had, will AI kill alpha? No, because of those sort of market microstructure changes. Also, the mosaic and ultimately the insight that drives conviction is not a function of just desktop research alone. It's a function of this mosaic of public information, but also hidden information. And it's a function of the three-dimensional complexity that is inherent in investing that a good thesis sort of first reflects what does the market believe, second reflects what do I believe, and then sort of this constant updating approach of where do I have my widest gaps, my widest rewards. So everything I can tell about AI as a tool is one, it's incredibly useful to the sort of recent shift to agentic tools is mind-blowingly useful, in my opinion. Three, what's important is this exoskeleton approach. Not the outsourcing of judgment, but the sort of supporting of investment process with AI that frees up time that surfaces more signals that I can apply my investment judgment to a broader set of higher quality investment cases. What skills do you think increase in value going forward? I mean, with all these tools around? You know, I think to some degree it's the old skills or the new skills. In one way, I think there are some changes as well too. I'll give you an example. Like, you junior analysts ask me, is it a waste of time to learn to model in Excel? And I say, most likely not, and I sort of give my experience on learning to code with coding agents. I can build something very simple with coding agents, but not anything complex because I don't know how to debug code. My sort of sort of brain gets fried when I look at a term, look at a coding term, terminal. My experience with AI and Excel now is exactly the case. The use case of AI and Excel, even in the last two or three weeks has taken a big inflection and improvement, but there's still a lot of debugging that is required. And so there's a level of mastery comprehension around the underlying process of modeling that is required to require to debug. I think to some extent, however, hedge funds have classically been very focused on hiring former investment bankers, private equity investors, and cell site equity research analysts because the ability to crank a model from scratch was table stakes to be a good investor. I think perhaps that could shift a little bit to the new value of a junior analyst on a team could be outthinking the market, could be creativity, could be imagination in tools, could be tenacity and going after key drivers, could be fluency in using these tools. The quantitative bar to be an investor at a large asset manager, I think, has gone down. If you can sort of handle the quantitative elements with some of these tools, then it's a question of how good a thinker are you? Do you have the ability to go in, understand consensus, understand why consensus is wrong, and calibrate an investment process to build conviction, build conviction around that. I'd say classically that's a skill that takes investors three, four, five years to learn and not all get there, right? Many hedge funds will hire ten investors and they can all build the models, but maybe three or four can get to the point where they can build that varying perception engine, that expectations gap muscle relative to the market. So I think the exciting point is that three, four, five year learning curve may be two or three years, one, two or three years now, which is a really exciting time to be a junior analyst. Some advice that I've been getting a lot recently, and advice I agree with is to read more, because with all these tools, there's often a temptation to outsource one's thinking, and we're seeing it happen with honestly, like, everyone in high school, middle school, writing all the essays, the chat should be tea or with all these other LLMs, when the point of writing the essay is not to get the essay done, the point is to think better. I guess for training of junior analysts, given that deep thinking is going to become an extreme, you know, an even more valuable skill given that all the other stuff becomes increasingly tablesticks, should they be allowed to use AI in the first couple of years, or should they have to do the deep grunt analog training, slash work that older investors had to do. I saw your your your pod with my friend, Aliq's, Pasquette, who is not allowing his investment team to use AI, and I'd say I'd directly agree with that. I think that AI is a power tool, cognitive power tool, that investors should earn the right to use. I'll give you a personal anecdote. My 12-year-old son was doing his algebra homework, and he was taking pictures of the questions, getting the answer and just writing down the answer. He got a few points off for not showing his work, but he did well in the homework, and then test time came and he had issues, right? And so that's an example of not a good use of AI. He's bypassing comprehension of the question. Now where I think it's nuanced and sort of the lecture I gave him is, I want you to be AI fluent, but not to bypass comprehension, but to expand comprehension, right? And so if you get stuck, ask for support, ask chat GBD for support, if you want, create a test prep based on all of the homework skills you've done, right? And so like any power tool, that tool can be used well or poorly. And so I think that's sort of one framework that I think about that the junior analyst generating AI slop work, mediocre or submedioche work, that's not useful to investment process, right? Identifying an idea that has alpha embedded in the idea is very, very difficult. You alpha lies in the edges, you know, developing differentiated perspective is very, very hard. You have tens of thousands of investors looking at the same stocks. And so it's very hubrisistic to say, "I and I alone have an out of consensus and correct view on this stock." To think that I can go and generate a thesis like that that hits that bar in a chat interface powered by a token completion engine on a train on a web scrape just sort of doesn't ring, ring, ring as true. Now, I agree with you about the reading and one of the things we're encouraging people to do is is, is, is do the same, but create more high signal elements of the, of the reading stack, right? For example, if I'm looking at a new, you know, healthcare company, I can do a deep research report about the history of the Medicare Advantage industry and I can get a very succinct targeted up to speed document around what, what's happening there. Find looking at draft kings and I want to understand what's happening with prediction markets. I can create a great custom predictions market primer that relates back to how draft kings and fandom will be, will be impacted. I can also create an agent to tracking system for news flow and other variables to sort of track how those things, those things are evolving. So I'd say the number, the number one piece of advice we give really all investors when adopting AI is don't optimize for speed necessarily, optimize for rigor. And so carefully think about your investment process. One thing about the sort of the pure wrote exercise of okay, you're updating, you know, six, six quarters of cash flow statements that I just need to see in the model. That's a low value process. You know, going to all the restaurant companies and putting same source sales into one Excel spreadsheet so I can see how you know, QSRs and casual diners are evolving over quarters. It's a dashboard that it's helpful to see, but not that useful to build by hand. And so find those areas where you can automate those low value signals. But in the high value elements of the process, be very careful at bypassing that, bypassing that comprehension. Ultimately our industry is a decision making industry and it's a discernment industry. And those insights spring from the foundation of the daily blocking and tackling of you know, reading the notes and listening to the earnings calls and reading the 10 Ks. And even this concept of summarizing a 10 K. Well a lot of insights come from sitting in your office with a pen and paper and a 10 K, you know, sort of slowly reading through elements of the 10 K, taking notes in the margin, sort of contemplating and absorbing what the business does. And after an hour, hour and a half of reading a 10 K, it's not just the information that I process, but it's sort of the contemplation, the dancing with that information that, that sort of the insight formation comes from. And so if you go back to three layer cake of investment process, you know, desktop primary and insight, the insight formation is really, it happens in unpredictable ways, right? And so if you short circuit the desktop research, do you short circuit insight formation? I tend to think yes, I tend to think yes. And so I think we just have to be more careful on things we sort of work flows to which we apply AI. One of the most powerful mental models I've seen, I worked in the Tiger world, I worked in the multi manager world. There are certain workflows that we would do in the Tiger, Tiger Cup world that you just don't have the time to do in the multi manager world, right? One of those would be deep research reports on CEOs and CFOs. I'm going to go spend two or three days doing all sorts of research about the history of this executive, where they worked in the past, talking to, talking to people they worked for, doing a deep analysis of the proxy and compensation alignment and going in if they were to pass company, go into the transcripts and models of what happened there, sort of building this mosaic of who this person is running, running, running the business, right? Long duration investors tend to be very much bet the jockey, bet the jockey sort of investor management is very, very important. If you're a, you know, lower duration higher velocity trader, it's not, it's not as important, but it is important. Even simple things like a guidance credibility analysis, right? This is one I was trained to do when you're looking at a TFO, go back and study the 10 years of initial to final guidance, right? Have they set guidance in a way that is beatable in the past, right? If they cut guidance, are there any tendencies on what quarter that guidance cut came? Often for healthcare, there would be Q3. You have a couple of back quarters. You can't, you can't sort of continue to hide the, the business weakness by Q3. So that's the meat the maker quarter where guidance comes down and you see a lot of sort of guidance cuts on Q3. That's a lot of work to go back into the prior press releases across 300 companies and do that systematic guidance credibility analysis, but it's really valuable. You might capture a couple of those that you would have missed otherwise. That's one example of a workflow that is obviously automatable with AI. Across 300 companies, let's do that, go into the past press releases, see it, and I can sort of see and systematize what of these companies are most likely at risk for a guidance cut. It doesn't mean that I'm going to take that feed and go make the traits, right? It means that, okay, team, Q3 earnings is four weeks away. Let's look at this report of AI where AI has indicated hockey stick guidance risk, right? Or beat and raise risk. Sometimes companies will be ahead, ahead, ahead, ahead, and they only raise a little bit. And if they've made comments at a conference that, hey, yeah, we, you know, we're feeling pretty good, but we want to wait till Q3 to revise it. Those little signals in the unstructured data can give you great sort of kernels around where a beaten raise or missing lower outcome may be. Again, it's a signal, not a thesis. And I could give you 64 other signals like that that are now possible to build in a dashboard with AI to, I think, an acceptable degree with validation systems in place. He touched, I mean, you just touched on signals from that unstructured data. And it's making me think back to what you said at the start of the conversation about using the new tools, AI for generating hypotheses. And then you yourself going from that point and doing the deep research. Can you lay out, I guess, specifically, or can you give an example on how you would translate these types of signals or just how would you generate the hypothesis with AI? AI does have sort of a creative ability to think through different scenarios. One of my favorite questions with AI is to say, create me a research plan. Who should I talk to? What work should I do to go and validate this hypothesis? In a way, the AI systems, I'm building become an orchestration tool of the humans. As a PM who's managed analysts, this could be really valuable as well too. I can have my AI augmentation process sort of help with hypothesis and say, hey, these are the 12 analyses we should do. These are the eight people we should speak with. AI can't do that. They can't go have a conversation, sold the sole conversation. So go and have those conversations now. And by the way, here are question lists, question lists to ask. So in a way, the LLM can orchestrate the human behavior, the downstream behavior. We can then take that context back in the decision making process. So using the tool as not something that has the answers because again, these are chatbots or agents trained on a web scrape. That best case will give you sort of a consensus view and consensus view will lead to a submedioca result in investing. But it can give you the range of outcomes. It can give you ideas, sort of stimulate ideas on which path to follow. You can then sort of accept and reject which of those paths and then apply the Bayesian approach of I'm going to go do research and figure out on this probability tree, this three node or five node probability tree, which of those nodes can I support with research to build investible conviction? That's the sort of piece that I don't think AI can do and arguably will never be able to do is sort of build that investible conviction. How do you build investible conviction that SaaS will or will not be killed by AI agents or Uber will or will not be killed by autonomous vehicles? These are unknowable questions, but these are questions and investors have to handicap to ultimately invest in these companies. And so AI can give me the range of outcomes. You could say, "Hey." help me imagine a future where autonomous driving is a tailwind to Uber. What does that look like? Help me size the tam and think about competitive dynamics and help me think about the value of the aggregation platform. And on the other side, help me think about Tesla Robotaxi. If they to develop sort of roll out that fleet and develop their own app-based ecosystem that obviates the need for the Uber aggregation, what does that look like? And then give me an investment process. What should I be monitoring to sort of identify what's path we're on? Because stocks reflect probabilities. A situation like Uber, I don't sort of give you round numbers, but Uber is a $75 stock. If AV is a tailwind, it's probably $200 stock in three years. If autonomous driving, autonomous vehicles are a big headwind, it's probably $25 stock. And so at $75, the stock right now is reflecting some sort of implied probability of the weighting of those outcomes. So AI can help frame that probability tree. But ultimately, if I'm a PM walking into my CIO's office and recommending a long or a short on Uber, I need to have done the work myself to support that idea. Particularly because almost never have I bought a stock and everything just go swimmingly. Right? You buy a stock and the fun is only begun. Right? There's drawdowns and sales side downgrades. So that sort of Bayesian updating of a thesis, you can't borrow conviction. Right? It's like what sort of saying what people, you know, traders will say, it's part of the reason why it's very tough to just go ask a friend like, "Hey, oh, you like this idea? I'm going to buy it myself." And the stock goes down 30, then what do you do? You call your friend. And so good investors develop their own conviction. You find signals. You develop. Your friend can give you a hunch or hypotheses. But a thesis is built on the sort of back of your own work. I think that's the same with AI. It's developing your own conviction. And so I think you go back to the pie. The great news is of the 3000 hours I have in a year, maybe a thousand hours was dedicated to sort of deep thinking, deep thesis work before. Now if that's 1500 hours, right? And I'm deploying these tools to develop deeper rigor around data tracking and thesis creep and, you know, idea pushbacks, etc. I can deepen the rigor of that work. I can generate more signals around the business. And I can also free up more time to go do sort of the meaty high value part of the work. I can go do more primary primary primary research. So that's to me why these tools are so exciting now in terms of what can be what can be done, what can be done today. Brett, you're super fascinating that you've Tiger Cup World, multi-manager world now teaching process to as a young analyst's young investors. What percentage of the coming a great analyst or more broadly a great investor do you think is attributable to talent versus just I guess work ethic wanting to learn about the business. So I think there's there's a key distinction in being a great analyst and being a great investor, right? So I think that's important. I'd say, and I sort of think go back to process process, process, and judgment. Being a great analyst is a bit more of a scientific, scientific process, right? Building models, doing research, meeting, management. That's a much more teachable, teachable skill where there's some degree of talent, but I think most funds evaluate that for evaluate that on the front end really well. Judgment and being a good investor is much more difficult. It's much more difficult to evaluate. It's hard for even allocators to evaluate are their funds good? Are their funds good? Even the process of evaluating is this alpha you've captured for the last three years? Is that just a lucky bet? Can you generate that alpha over a cycle? I think that's harder. The great investment firms have an inescapably a marriage of good process and good process and good judgment. In general, talent is revealed and investment firms over the years more than trained, in my opinion. Why do I say that? If you look at a typical 30-person fund, for example, maybe 25 of those investment professionals are analysts. They're doing the work. They're covering the companies, generating the expertise, building the models, doing the due diligence, feeding that process up to the decision makers often. Now having an influence on that decision, but those decision makers that rise through the ranks of that hypothetical fund are the ones who have that skill, that real talent, to take that investment process and translate that investment process into investment results. Some of that is training, but a lot of that are just the revealed intellectual wiring being a good investor over time. It's the independence of thought, the ability to go with the flow, when the flow makes sense. Consensus pays a lot of times, but having the ability to have that research driven conviction to go against consensus when the facts matter. That's more of a little bit more of an innate nose for money is something that's harder to teach and train. Do you think that the type of person that thrives at a single manager is the same as a multi-manager, is the same as Tiger Cub? Yeah, I think more similar than different, I would say. One insight I had having been at five different funds is that I looked around the halls of those funds and their investment process was more similar than different. I think individually we all had a sense that we were doing super unique things from an investment process. I think that was quickly dispeled. That's a founding inside of Fundamental Edge. Let's take all these things that are like, you're going to say best practices, but it's like, they take the best practices that I've observed through 13 years of doing this, that aren't in a textbook, but are widely practiced. But I think it's also a recognition that process is just table stakes. That judgment is sort of applying the outputs of that process or where the real alpha lies. That's a trickier thing to that's a trickier thing to train. You can't have judgment without process, but process alone without judgment and I think will not give you the proper results. So circling back to AI, AI can be great at sort of mediocre process, but with no judgment at the back end, what are you left with? You're left with an AI slot portfolio, which you may get lucky and out perform the market, but it's luck not skill in my opinion. Last question. Are there any skills that you think are underrated that are often present in the best investors? I think the biggest one that comes to mind is just curiosity, which is very hard to sort of innate in some sense. My first interview process, first fun I worked at, the last interview with the CIO is turning the tables and the interview E has to ask questions of the CIO the entire time. And so in that sort of ecosystem, there's a sense that the art of curiosity, the art of asking good questions is a really important metascale for investors. And so that was one of the biggest red flags for me. I adopted that, not that whole process when I was interviewing analysts. If they had no questions at the end of the interview, that was almost the biggest giant, disqualifying red flag. You want to spend your career working for me at this firm and you have no questions like, I'm sorry. So I think a genuine curiosity and passion is a huge metascale. I think this even compounds more in an AI world where process becomes more commoditized in a way. The barrier of sort of exploring your creativity goes down. This sort of hunch of, hey, I wonder if this Asian business will be impacted by oil. prices, blah, blah, blah, you may not have the time to go and do that analysis in an analog way, but I can spin up an agent to do that analysis, right? And I can sort of chase my curiosity in a much more detailed way. So intellectual curiosity, the art of asking the right question is a very important skill. And then a big part of using AI tools is just being very clear. We sort of shifted from the art of prompting to the art of context. And the biggest unlock in I've had in using AI tools is just a building rigorous context documents, like being very, very clear on who I am, what I'm trying to do. Being in many cases 30, 40, 50, 60, 80, 90 page context documents that can sort of wrap the process around that deep context. So the ability to sort of break down, be a system thinker, articulate what you're doing and why and what your objective is is a new meta skill that I find to be very critical. I love that. I think, yeah, curiosity, clear thinking, in an age of AI that will lead to victory. Thank you so much for coming on the podcast. Brett, this was wonderful. Awesome. Thanks so much for having me, Ethan.