00:32:44 · Joe O'Donnell
[00:32:44] S0: Well, it's terrific. AI sounds really, really, AI honestly is neat. But the actual process of getting an AI model to operate in your image is just horrifically bad. It requires a ton of manual effort. A ton. Know, I mean, to take... You know, just to go back to the disclosure statement example. In order to get a model that can parse through a disclosure statement at the level of a very good analyst, it requires very good analysts going through, you know, whatever, annual reports manually and saying, here's what I would say. And then they look at what the AI says and, you know, through fine tuning, reward functions, etcetera, all that kind of stuff. Over time, you hopefully get, an equivalence. But the only way to do that is by a huge amount. I cannot understate how much manual effort to actually make that work. And one of the reasons for that is somewhat obvious, but another reason which is obvious is the thing I was talking about, the long tail of results. I think there's relatively large Venn diagram overlap between very interesting insights and things that only happen very rarely. And so you... That needs to all be codified. You need to have a sufficiently large sample set that you've actually fine tuned to actually allow the AI models to understand when they see something like that and what it means. And it just requires enormous amount of manual. And that's just disclosures. There's hundreds of those, literally hundreds of those examples that are required to get performant fine tune models, performant defined as what a good investor would think about this particular thing. It needs to be equivalent. My view is that, let's say, I don't wanna pick on anyone in particular, but let's say you could get a result that was 95% as good as a good analyst for task x. Some some tasks, by the way, that's fine. But tasks that are centered around making the right decision, that is not fine. And if you... If it's not possible to get to 100% equivalence, then the default, what I would at least recommend for professional... For public markets investors, the default setting for that would be do it yourself. Do it the old do it the old fashioned way. Because, as I was saying before, AI is a... The the day... One of the dangers of of of overusing AI is higher conviction when you didn't deserve it.
00:35:29 · Joe O'Donnell, Khe Hy
And you never know what the missing percent is or how big the percent is or what what even category it's in, or whether it's a false positive or false negative without doing the AB testing. The only way you can avoid doing the AB testing is if you have AI models that, you rely on either that you buy yourself or that you buy from someone hopefully like us, that where you're convinced it is doing the job at the equivalent of of you or your investment team. [00:36:02] S2: And is it fair to say so someone like an Anthropic is hiring a lot of former bankers, former PE folks, investors, and they're doing a kind of a mass, like a production line version of what you described. And then Canary's alpha in this situation is like the people doing those evaluations and the fine tuning are just more in line with a hedge fund analyst. Like if they're more specific, is that a fair analogy?
00:36:30 · Joe O'Donnell
[00:36:30] S0: I think that's part of it. It's this multiple layered thing. So I think you could throw infinity investment bankers at a problem that applies to public markets investors and you would not get to the correct level of of accuracy. And I also think there just tends to be adverse selection. You know, that's what I was just saying about what it actually empirically means to go train a model. This isn't a job. You know? No one's leaving to take investment bank. I I suspect at least maybe there's a counterexample, but no one's... The head of TMT banking at Goldman Sachs isn't leaving to go train models for Anthropic. There might be a former Goldman Sachs investment banker that is training models for Anthropic. My suspicion is it's not their best bankers. That may be wrong, but that'd be my guess. And I think for certain industries, that's fine. You know, if they're more formulaic industries or lower leverage industries, totally fine. You don't need the absolute top people doing it to get to results that can be impactful at at... Through all stacks of... Or all parts of the organization. I do not think that's the case for public markets investing. And, you know, hiring... This is an industry where if one is a very talented investor, it can be very lucrative. And, you know, the... I guess I've seen some some very large salaries thrown to to... Or compensation packages thrown to some of the AI engineers. And so perhaps one day, those companies will do the same for... To train for public markets investing, but I haven't seen it yet, so we'll see. And... But I I think until that that happens, it's unlikely that for this particular use case, that will be satisfying. I I say more generally speaking, whenever I see a firm that says we do AI for our financial services, to me, that's that's... They've already lost. I I don't... That that doesn't mean it. There's so many different things that that could possibly mean that... And they're all very specific with lots of nuance. There's no such thing as AI for financial services. Yeah. We build specifically for public markets investors and and, you know, obviously, more more broadly the ecosystem around public markets investors. But I think we...
00:38:44 · Joe O'Donnell, Brett Caughran
You've actually started to see this with, you without naming names, but you've seen certain other companies specialize in investment banking, for example, which makes sense. It's it's... Building AI for investment banking seems to me to be a really interesting business model. But building AI for financial services means you probably don't understand the problem statement. [00:39:08] S1: I did I did a test on this exact point, Joe, with AI and Excel. And AI, Excel will build a great investment banking model and your balance sheet will always balance. But almost none of my models who actually forecast out the balance sheet to balance balance the the balance sheet in the forecast period. So it's funny, you actually have to like, I totally agree with you. You have to untrain the AI to actually make it approximate what a public market investor does. Which brings you to a question on Excel. I mean, to your point, there hasn't been a great software stack for public market investors because complexity of what we do. We've sort of been stuck with a Bloomberg Launchpad and Microsoft Excel to be the workbench, the analytical workbench. How do you see Excel shifting, both in terms of ingress, egress with AI tools? But how do you think the operating system of the public market investor will shift in the coming months and years?
00:40:18 · Joe O'Donnell
[00:40:18] S0: It's an interesting question. I suspect it will be different for different investors based on their strategy, based on their own preferences. But what I'd say, one thing that seems obvious to me is that a lot of those things will just be abstracted away. Mean, what do people use Excel for? Well, one is building a financial model and forecasting key KPIs and stuff like that. Some of it is more bespoke analysis. I would say in the category of bespoke analysis, a lot of that stuff can be abstracted away or already is to a certain extent. I mean, know, our super analyst product, for example, does a lot of work that would have traditionally been done in Excel without opening Excel. And the answers to those questions are just emailed to our users and available in our web app, in our MCP. And so there's just no need to get in... You know, open anything new to to get to the the same results. I think, you know, build build me a model, that kind of stuff. I I think if... It is very likely that some version of people interacting with Excel for that explicit purpose of, you know, forecasting EPS or whatever. I don't see... It doesn't seem like there's a real problem with doing that in Excel. And I don't think... Again, depends on the model. If if someone has a coverage universe of 500 companies, for example, it's different than a concentrated portfolio where they own six stocks. So perhaps there's there's more of a... An automation case in the former case than than the latter case. But those... As I was saying before, if the foundation models themselves haven't already solved most of those problems, it's only a matter of time. And that time frame, who knows? We'll see. But I would personally be surprised within twenty four months if, that's not a a solved problem. But I would argue, you know, somewhat facetiously, that already is a solved problem. You know, the old the old version of build me a model was email a sell side analyst and say, send me your model. And that's the same thing. And there are products like Canalyst that exist or still exist that did that sort of, you know, the old fashioned, more deterministic way. And everyone's got a different thing. You know, for me personally, I had a different kind of heuristic, which was if I was really relying on some nuanced understanding of a financial model, probably wasn't that good of an investment.
00:42:44 · Joe O'Donnell, Brett Caughran
You needed to build a model somewhat on a backwards looking basis to understand risk reward and stuff like that. But most of my investments, I already knew whether they were good investments or not before I built a model around them or built a Nuance model around it. So I think it'll just be different for different people. And Excel by itself, I don't hear a lot of people complaining. I don't talk to a lot of people. I don't hear a lot of people complaining about using Excel for those kinds of purposes. You know, Bloomer perhaps is another story. [00:43:18] S1: What, you talked earlier in the sort of agentic analysts about, obviously the data piece, but curious about where you're at on integrations to other data, even simple things like sell side research, pulling in the right news stack, and obviously pulling in expert network calls from players like AlphaSense have continued to be a challenge to just even get all of this external data into one environment. You've talked about creating your own data sets, but can you talk about this integration challenge of pulling in these other publicly available semi proprietary data sets where the barrier has been more commercial than technical?