00:27:06 · Raj Shah, Brett Caughran
[00:27:06] S0: I love that. [00:27:07] S2: And one way I've seen people do is actually record a handful of those. It's like, obviously, our different clients record a few where you're just ripping apart an analyst pitch and the agents can infer that pretty effectively turn that into a sparring partner. [00:27:26] S0: I like that a lot. Grounded [00:27:28] S2: and like you create that a a Right, right. Talk to me about Excel, which has been sort of a challenging one. I mean, of today, even still, Microsoft Excel is the workbench of the fundamental investors. Sort of like what we do is so heterogeneous as investors, there's no front office software. So like Excel's been it. [00:27:48] S0: Yes. [00:27:49] S2: Sort of the translation layer from research to decision. What have you done with AI and Excel and sort of where do you see that going?
00:27:59 · Raj Shah
[00:27:59] S0: I probably of everything we've talked about so far, it's probably been the most... It's probably been the biggest positive surprise of everything we've used in terms of AI applied to hedge fund investing and managing business and all of that. We started using out, like, I'm sure many people did, a lot of the startups and a lot of the guys who are early to doing AI integrations into Excel to help sort of turbocharge model building, analysis, things like that. And then very quickly just sort of gravitated towards Claude, Codex, those integrations with Excel. And they're they're fantastic. They're amazing. And they've sped up... They... On the on the good side, they've sped up our ability to just cut data different ways to create output. A very simple prompt everybody should use, it actually speaks to what you were just saying, Brett, about the pitches, is you can just ask an agent. You can ask Claude when you have one of your models open to test this model. Like, where is the structural flaw in the logic of my model? Yeah. And it'll... It's actually quite insightful. Oh, wow. The way you're, you know, doing free cash flow conversion makes no sense at all. Or you're building a bunch of cash on the balance sheet. Is it gonna be used to buy back stock? What are you doing here? Sort of a lot of times, it's funny going back to, like, when I used to look at sell side models and like, oh, that's dumb to sell side model it this way. Well, they do it for a reason that way. But oftentimes, like, it points out those dumb things for your own models. And it's a great sort of check privately. You know, it doesn't embarrass you in front of anyone in terms of where a flaw in your logic might be. You could then take it a next step and you can say, hey, here's the last three transcripts. Here's some appearances at conferences. Here's what peers are saying. Is there anything about what the company's articulating about its near term and medium term to long term that's inconsistent with the way I've modeled it? And the the the combination of a smart LLM with the ability to navigate and contextualize Excel is amazing. And so you can use it that way on financial models. I've loved doing that because the modeling is... Look. As I've gotten older and I've become a portfolio manager and a founder of Fund, I spend less and less of my time in Excel doing model building.
00:29:59 · Raj Shah
And so it's been helpful to help me just remember, look, here here are the things that are important. But there's all sorts of other ways in which I've been using it. We've... I mean, very simply, we had a... One of our largest holdings. We were trying to help them understand the difference between talking about low and growing at a lower rate versus truck talking about growing at a higher rate and the valuation gap between those lower growth companies and the higher growth ones. And very simply, we were able to create a scatter. We said, look, dump a 100 companies in it. Dump a 150 SaaS names in there and spread apart all of them, by market cap, spread apart all of them by, growth rate, cut out the low gross margin ones, a ton of different analysis. You just keep firing into Claude and Excel. It's... And it's just repeatedly doing a bunch of analysis in real time for you. Two years ago, I started to become a more power user of Power BI. We were originally doing Tableau and things like that to try to do data visualizations. And I found myself, like, drawn right back into Excel. Excel with Claude, you know, next to you is kinda better than doing a lot of these data visualization things because it's newer. I hope maybe they can get, you know, a good LLM into Tableau or Power BI, and then maybe then you're, like, really off to the races. But it's been super instrumental in that. And then the last thing is, like, just tell me a story. Tell me a story about this set of data that I've... That I have here. I'm... It's funny. I'm... This is totally unrelated, but I just I just bought a new home, and I'm in the process of selling my old home. And I had a meeting with my real estate agent who's selling my my old home, and she had a bunch of, you know, insights and, like, you know, that's, like, more art than science. I'm like, hey. Here's where I think you should price your home, and here's a bunch of comps, and here's bunch of listings. Immediately threw all that stuff into Excel, and I was like, hey. What do you call it? What do you infer about home prices relative to other characteristics about these homes in the town in which I live here in Northern California, which is Northern California is like an insane place for home buying and selling right now. And it very clearly drew a correlation between the...
00:31:58 · Raj Shah, Khe Hy
In in this specific town between the age of the home and where homes were being listed at. And it was basically inferring that, look, people tend to... In this town, people tend to bias towards homes that are newer, and they don't have to be fixed up as much. I'm like, that's an amazing insight for Claude in Excel to just do. Now apply that to all sorts of other sets of data, not just models, not just the boring stuff, credit card data, other kinds of alt data. Inferring patterns is, I mean, like such a great use of this. [00:32:24] S1: Hey, Raj, on that point, tying that to the intern and the junior question, you're deliberately not trying to hire young, you know, more junior folks given trying to stay lean. What happens though, if larger firms do what you just described and you're like, well, we don't need the juniors to do the models because everything you just described. [00:32:45] S0: Yeah. [00:32:46] S1: What do you... What happened? Like, do you do you believe that there's a a challenge for, like, the next layer of talent? Or or is that also kind of a narrative that, you know, that that that we're falsely buying into?
00:32:57 · Raj Shah
[00:32:57] S0: I think it changes the progression of a junior analyst and not necessarily in a bad way. So I think the junior analyst now has to be sort of AI native and AI fluent as soon as they come on the job. Whereas maybe before, you just had to kinda be Excel fluent. And you still don't have to know a lot about what makes a good business or a bad business or how to read a CEO and how to think about consensus. You learn all of that stuff on the job. I still think you're gonna need to do that. And so I think the junior analyst will still grow into those types of things. I still think it's gonna necessitate those types of conversations. Like, hey. I created this output. I've done this analysis. I've done this screening work, but let's also have the conversation PM or sector head about why you think this idea is interesting, what makes a good business, what time horizon we're thinking about, what are red flags we should look for. The analyst is still gonna need to know all those things. And I think that goes back to this point that a few people have made, but I I don't think it can be said enough that AI will almost always lack that judgment factor. And that's the part that junior analysts will grow to learn. You learn through experience. You learn through exposure on the job. I I kinda hope for my own sake because, you know, I hope my competitors just lean too hard in the AI stuff that the judgment stuff sort of becomes lower priority and it'll present better opportunities for me. But I think the reality of the way this industry is going to evolve is people are still going to lean on the judgment side. We've just now sort of embarked on a whole new set of tools to make us more efficient. That's the way I I kind of think about it.
00:34:31 · Brett Caughran, Raj Shah
[00:34:31] S2: Yeah. That that makes sense. So the hearing interesting rumblings about LLMs sort of developing judgment in various tests and competitions too, which starts to get sort of scary. About if all of a sudden LLMs could actually give you a 55% batting average on earnings prints. Yes. That's sort of a totally different conversation. Not to say that's my base case, but hearing more people talk about that as a... [00:35:00] S0: But it's funny, right Brett? If you play that out, if enough LLMs do that, then it sort of already defeats the if enough people are using an LLM that's giving you sort of that same answer, then you won't get that earnings move, right, after the print. Because everyone's sort of... I feel like it's kind of akin to, like, all data started creeping out, right, whenever that was ten years ago in the industry. And the first early adopters of it were actually generating some alpha. And then after that, you had these companies like beating comps and the stocks were going straight down. You're like, god, why is that happening? You're like, oh, well, it was happening because everybody's looking at the same set of data. It's probably a similar analog to like what could happen if people become too LLM dependent. Yeah. I think the the most amazing, sometimes humbling thing about this job is the fact that you can be right and still lose money or be wrong and make money. And I don't think that's gonna change, you know, with this.