Personal Finance and AI with Cleo founder and CEO Barney Hussey-Yeo

My impression of the limit, I think, looks radically different than where we are now. Most of the machinery of intermediation was designed for people. The same thing is true of finance. What we are trying to accomplish in financial markets at some level is shockingly simple. Once you have tens, hundreds of millions of users, you radically understand them, you can start to have people and agents bid on those users for the products they want. Like, I'm a user. I don't actually want a credit card. I just want to borrow some money. And you have just introduced market mechanisms into something completely broken that puts so many people in harm every single day. And if you can do that, you can fundamentally change what all financial products look like. Welcome to the Darby Mill series. Welcome to the Darby Mill series, pushing AI to the limit. I'm Ajay, co-founder of Intrepid Growth Partners. And today we're diving into how AI can transform personal finance decisions for everyday people. I'm joined by my collaborators, Rich, pioneer of reinforcement learning, Sendhil, who studies how AI influences human decisions, and Niamh, an applied AI scientist who optimizes foundation models. Our guest today is Barney, founder and CEO of Clio. Cleo's mission is to help people make smarter money decisions by predicting where they might slip and offering guidance to help them make better spending decisions. And with that, let's start the show. Awesome. Well, that was a good intro as well. But the quick, too-mini-abridged version of Cleo is she is an AI assistant that helps you make better decisions with your money, very simply. So I started the business in 2016. So this is pre-Transformers. I was doing this on hard mode. This was all supervised learning, traditional NLP. It was a little bit contrarian to build an AI assistant back then, but we did it anyway. Built up an amazing corpus of data and we learned a lot through the process, obviously. But essentially, when a user comes in, they connect all their bank accounts to Cleo. So they connect their financial data, their debts, we get all their spending, their income. And it's an amazing data source. You learn so much context about that individual through their financial history. It's one of the richest and most interesting forms of data that we produce as humans. And there's lots of analogies, whether you go to health or education, where you can see an agent like this. It helps you make better decisions every day. So Clio does that for financial services. And over time, we have built a big engaged user base, which allows us to explore the search space. We'll get all the reinforcement learning quotes in through the podcast, I'm sure. But it allows us to really think about those users, give them the right financial advice. And it's a proactive service. So yes, Clio can answer questions and you can think about what you want. Maybe your retirement planning can ask a question. But I think the big paradigm shift here and the big paradigm shift in consumer apps is going to be having these reasoning models constantly working 24-7 to think about the data that you're creating, your life, your context, coming up with the next action that you should take, the next policy, and giving you that advice proactively. So we think of this as a proactive service, how lots of our users engage, and that's how we drive the action and the behavior change today. And on the back of that, we've added financial products. The financial products allow us to Let people take action to move money, to get better financial products, to save all the gambits of a retail bank we're adding into the ecosystem, both our own kind of infrastructure, but also partnering with best in class players as well. So you can think of it as the intelligent layer above all of your finances. It's trying to help you make better financial decisions every day. Wonderful. Thank you, Barney. And, and just so our listeners have a sense of scale, because there's many, many, you know, There are a lot of apps out there that are targeted in this market for financial services. But yours is, if not the biggest, certainly one of the biggest. Can you just give listeners a sense of how many people use the application? Yeah, millions of active users, millions of paying subscribers. We're 350 million ARR today. We're profitable. We're growing 2x. So it's a rare combination of a large-scale ARR base, big active user base. Profitable and fast-growing. You don't normally get the ARR at this size and the fast-growing and the profitable. You normally get one of the two if you're lucky. And we've got all three. So it's kind of one of the first at-scale, domain-specific, very vertically integrated AI agents. And it serves 100% of the US audience today. But we're going international as of next week, actually. So it's... Yeah. And just for context, the audience is relatively young. So median age is 27, 18 to kind of 36 is the target demo. And we help them level up through their financial health. Excellent, Barney. Thanks very much. Let me turn it over to Niamh. Barney, great to see you. Congrats on all the progress. I think we met way back in 2017 when this was like a mere concept and an MVP on Facebook Messenger. So it's incredible to see how far it's come. Just on the clarification question, just what you touched on earlier around that shift from kind of reactive, consultative, conversational mechanism with the user to that That more proactive suggestion approach. Is there a particular onboarding period you require or you found kind of averages out whereby you have to earn the user's trust first before you can start doing things proactively? Or actually, is the app now able to kind of jump in there fairly swiftly and the users react and not feel like it's overreach? Yeah, we've done a lot of work on the onboarding through the years. And the thing that we found that was kind of the most important was we want to give them some value up front. Take that financial data and then we play it back to you in a way you've probably never seen before. Most people don't really understand are they spending less than they earn? They don't understand where their money's going by category, by merchant. They don't understand how their spending differs from someone like them in the same kind of neighborhood with the same income level. So able to play back your entire financial life, give you real good context there, which kind of is this aha moment for users. We have a fun kind of silly feature called roast mode, which will make fun of you Maybe if you'll kind of... your postmates spending and how many ubers you've taken and just building a bit of fun and engagement since product was like really critical to getting trust in the early days especially. But then we have asked people for their goals so it's generally like a Okay. Hey. then it's not that complicated to be honest for our users it's hey I want to get out of debt, hey I want to save for a deposit for a mortgage, I want to pay down my student loan. And they're all kind of savings goals if you think about them in some way. And they're all going to go over a multi-year horizon. So they come up with that goal. We have a little voice interaction with the user. We interview this out of them. And then we set you a plan. So it's a multi-year plan, sometimes even like a decade long. And we break that down into paycheck, paycheck steps. So it's giving you a goal over a long time horizon. And then every day we check in with you to try and help you hit that goal and change your behavior or help you take an action that's going to get you towards That goal in the future. So that's how we think about onboarding, at least, anyway. Thanks, Steve. Thanks, Barney. Rich, over to you. Oh, hello, Barney. I'm really excited to hear about this, your company. Some quick questions. Do you use large language models at all? Yes, a lot. We didn't, though, because we were pre-LLMs, remember? So everything was supervised learning, intent classifiers, NLP. We had a recommender system with a contextual bandit that did all the push stuff to start with. Now everything's LLM-based. You've got to think, we were pre-generative, so we had loads of writers, loads of behavioral scientists. They created amazing training data for us. They wrote loads of the insights. They wrote loads of the responses. That became great supervised fine-tuning data, but then in a reinforcing context. Yes, that makes sense that you're now using them thoroughly in your system. This isn't a question, but I wanted to mention that there's something very similar that I'm familiar with from the research literature. There's people doing clinical trials on basically behavior modification for addicts. And and they it's all about this, you know, helping the addict avoid their addiction and, you know, warning them that they might be entering a situation where they'd be tempted and all this sort of thing. It seems like a similar. Yeah, no, 100 percent. And if you think about. The interventions that we take or the advice we give, a lot of it has had to be grounded in what is the literature and what are people doing today. So even if you go talk to financial experts or financial advisors, they generally do serve a higher net worth population, but there's lots of people serving everyday consumers as well. You talk to the debt charities, StepChange, for example. we learned a huge amount about essentially the policies that they created by hand and like the literature and what they did and we're able to encode that in the system right in terms of uh when we had an intent we'd write the response to it when we had insights we'd write what could be potentially viable so there's like a human kind of bootstrapping to that process Cool. Now, I think maybe my, I have this main question, which I always ask when I see an AI system. AI system normally has a lot of knowledge, and your system has a lot of knowledge about what to say next and how to organize the understanding of each individual. But just, you know, the main knowledge is what to say next and what to recommend. And to be a good chat, to be a good motivator, you have to say the right things. And so the question is, what keeps the system going? What keeps the knowledge in the system correct? So I see two sources of knowledge in your system. One is, well, there's a large language model, which has a lot of knowledge of what people tend to typically say in various situations. And then there is, you know, you and your company and you, I presume, are spending a lot of time, you know, crafting the system, both, you know, what it should say to motivate and what it should suggest and how it should understand each person so that there's a lot of that is kind of manually handcrafted and can you can you speak to just you know how much of a burden this this this crafting is and and how much of it can come automatically from the large language model. That was the big change, right? Post-generative, like pre-generative, everything had to be crafted, but it created an incredible data source, right? For supervised fine tuning. And now, you know, and we're in the post-training world, like everything today is post-training, right? It's kind of wild that you were... It's just wild how reinforcement learning is now, 30 years later, it is the thing. I want to ask you questions now, Rich, but I'll save that for another day. But if we think about the objective function or the reward model for Clio, that's the thing we've kind of iterated over a huge amount over time. If you think about the first version of the recommender system, and it's a contextual bandit, and we're thinking, we're just going to do this for engagement, or we're just going to do this for retention. And then the system starts getting... Really fun, actually, but a bit spammy and a bit clickbaity in a way. And then we go, oh, we've got a need to make revenue and build a business model. We're going to optimize everything and the objective function for revenue there. And you go, oh, my God, it's just pushing up sales 24-7 to users. This is not good. So... The long time horizon kind of policy or reward model objective function we think about is financial health and like the multi-steps you have to get there. So we spent a huge amount of time understanding the transactional data. What we get from a bank is a string, a date, an amount. So it's kind of very raw. It's not a great understanding of financial health in that sense. So you've got to parse out of that all the important components and you've got to build higher level concepts until you get to what financial health actually is. So you go from that string, you turn it into maybe a merchant, a category, you get metadata about where they're spending that money at that merchant and category, and then you start to build up concepts like, that's your income, that's your bills, that's all these different variables, and that goes into a model of financial health for us. And then we're able to use that model, that score, it's like a zero to 100 score, 100 is AJ is the venture capitalist now, Intrepid. I'm excited. And, you know, one is the other end of the spectrum on that, but we're able to go, was that insight, did that insight lead to an outcome from the financial data that is? grounded in truth that allows us to then go and retrain the model and get data from it and yeah either to do the reinforcement learning and update the reward function or do some preference optimization on the DPO side of thing that was a very long answer Rich but like you know simply you've got reward models and objective functions and what we found is really important is like aligning that to the north star of the company which is can you improve your financial health health from this thing. And the interesting thing is, It kind of encapsulates some of those other metrics we talked about earlier about engagement and revenue, because to improve someone's financial health, hey, they've got to actually engage with the app. Hey, to improve someone's financial health, they're probably going to take the financial products that are going to make them better off. So it's kind of our North Star, the business. But it's like, if you think meta about all of these AI systems in the wild, the objective function is like the thing that changes society the most, right? Every feed that we look at, whatever that company is optimizing the objective function for, like that is... That is how we view the world. And that is increasingly how we're going to view the world, especially as AI becomes more prevalent in everything we do. So if I was a politician, I'd be really pushing for making all of this open source so you know what people are trying to optimize towards because that is the world today. Does that make sense, Rich? Yeah, yeah. So one of your major levers is the objective function and what you train towards. And you mentioned building up of these concepts that are essential to financial health. That can't come from the large language model because it's all post-training. Thank you. That comes from, it seems to me, it comes again from the construction process that you and your other experts are doing at your company. Yeah. Okay, good. Thank you very much. Thanks, Rich. Rich, before we go to Sindel, Rich, could you just for our listeners, Give a quick description of, you know, it seemed like when you asked Barney the question, you said, where does the knowledge, you know, what's the ratio of knowledge, the source of knowledge? And I thought you were asking me effectively a trick question, which is you said, well, one source might be the language models, and one source might be the human engineer knowledge that you put in at your company, Clio. And I thought you were laying a trap for him because what you are really going towards is, but we're moving to the age of experience. And that's knowledge generated by neither category one nor category two. But he anticipated that answer and went right to, well, we handcraft the objective function, but then the system learns the knowledge through the behaviors of the users. And that knowledge is neither knowledge that Clio knew nor that was embedded in the language model that was created through experience of using the system. In response to the handcrafted objective function. And then carry that one more step further. I think Barney also said that building up these important concepts of financial health, that wasn't entirely through the reward function and the interaction. That, again, was there was some participation by the engineers, which is a nice mix, really. All right. Let's thank you. Sindal, we'll turn it over to you. Great. I have two questions. My first question is, feels like, and both of them revolve around my impression of what's been hard in this financial planning space. So the first is, it's very hard to get a data frame that gives you a view of what's happening with the customer. So if you had access to their bank debit card and you saw all their transactions, you don't have access to a different debit card, a different credit card. You don't necessarily have access to their 401ks. Even if you had access to all of that, you then don't have access to shocks that are happening in their lives. Like it turns out that my cousin is sick and I'm not well. So there's a sort of snapshotting problem, tracking problem. In that finances happened in a very decentralized way across quite a few things, some of which are digital, some of which are not even digital. How do you even just think about that problem in your implementation of Clio? Yeah, in two different vectors. One is on the financial side, that's actually just getting better and better. So Plaid, Method, all of these providers are allowing us to connect to more and more financial accounts over time. So the view, just like the transactional data and the open banking view has got better and it continues to get better. But the other point that you made, I think is so critical and so interesting. The context of the user, the context of their life, the context of what drives them, what motivates them, who the hell they are as a person, is not all captured in that transactional statement. So the conversational interface is the evolution of the GUI, right? We've had 30 years of a four-inch screen and a few pixels. The conversational interface allows us to capture... Everyone calls it memory, but it allows us to capture your preferences, what you care about, how your day was. We have people that have two-hour conversations with Clio multiple days a week. It's more than just a financial assistant to a lot of people, it's a real relationship. And I think that's incredibly powerful, obviously, to... Provide a better service. Yes. But it's just incredibly powerful as a new paradigm in computing. We're going to interact with voice way more often. We're going to have more of these conversations. And it's going to allow us to have just radically better experiences with computing. And computing will be less as, we're all looking at screens, right, today. In the future, it's just going to be all around us. We're going to talk to it. You don't have to look at a screen. You pull it. Computing is going to be more proactive for you, do things for you, Versus you having to like. click around and find the box. I've got to move my Google Chrome over here to record this goddamn podcast and blah, blah, blah. It's going to become this proactive, hyper-intelligent assistant for these big domains of your life, health, education, wealth, your relationships, work. All of these domains will have a really deep, vertically integrated domain-specific AI. And it won't be this generalized model, this OpenAI. It will be specific and really deeply, vertically integrated. Get more data that way, you get more outcome. And I think they're going to be really powerful. I think they're going to change the world. They're going to be the apps that make a real difference in society and the economy and people's behavior. So if you can do that in one of the big domains, which is money, it's the thing that drives everything, right? It's the thing that has the most attachment to... Your stress levels, your anxiety, your fears, your wants, your dreams, life is unlocked because of that. Look at all the populism and politics that's happening today. It is driven by the lack of money and the lack of progress of the bottom 90% of the income earners in the Western world. So if you're going to make a big change, you want to do it in one of those big verticals. Health is another great one, but there's many. But so if I'm hearing, am I hearing you saying the imperfect capture problem is something we're getting better at, which I agree. But so is the view, let's just punt on that because in the medium term, that'll disappear or, I mean, we live in 2026. What, what view should I take about the imperfect capture problem right now? Like I log onto your app, Plaid has some of my stuff. I don't even know if my benefits cards are loaded up to it. If I have food stamp, I. You know, and the context may for the two hour user, maybe some is captured, but maybe I don't know what to tell. Like right now or is the view that let's not worry too much about that right now will provide some value. Any answer is fine. I'm just trying to understand what we do about the imperfect capture problem on the path towards full capture. Yeah. Data will get better, but you need to build a relationship with users and get that context out of them. You have to build a relationship with them. You have to have conversations with them. You have to fill in the blanks that are important for your system. And that's going to be different per system. For us, it's getting their goals. It's getting what we're missing. It's saying, hey, we see you are spending on a credit card. Tell us about that credit card. Hey, we don't have the APR on that credit card. Which type of credit card is it? And you can only do this in like incremental, you think about this as a class. Classical UI problem is incrementally onboarding a user through the cycle, and we do that very explicitly. So I think all these services will gather little bits of data in a very natural human way for you to fill in that problem. You'll never get everything because we can't download links, we don't have a neural link yet. Maybe that is the final solution for all of this, but that relationship is so critical. And it's okay. Let me ask my second question. It's related to the relationship. I feel like the other stumbling block of every financial service product has been effectively monetization. So let me just give you what I think of as, it's going to sound pessimistic, but I don't think it's kind of an unrealistic view of what has happened in the past. Doesn't mean it's going to be different in the future, which is that people don't really value advice enough to to meaningfully pay for it. That's just been a consistent finding in the literature and in the world. So then what ends up happening is we end up giving the advice bundled through some other modernization strategy. And typically that has, not always, but that has distorted its way into... putting people into products that like typically finance, consumer finance has made money by and large by putting people into products that charge them stuff they didn't realize they were being charged for and got the money back at some level. But then they think they're getting stuff out of it. So that's the equilibrium right now in the market. And I think of that as a sort of a A market exchange problem, which is that the consumer isn't demanding the thing that they probably should be demanding. So how do you think about how to monetize Clio to not fall into that trap? Yeah, I think that the glib answer is, hey, you have... Hundreds of millions, I don't know if it's billions now, of people paying for ChatGPT. As intelligence exponentially grows, you're going to be willing to pay for intelligence. If you've got 50 PhD researchers working for you, it's very different to the types of advice that you had from any software service in the 90s, 2000s, 2010s. So I think that's the glib answer. I think the more specific, nuanced answer in financial services and to our goals is We do have to build products that are better than the incumbents. You need competition. You need markets. You need to build credit products that are cheaper, that are better, that are less opaque. You need to build wealth products that are less predatory and don't charge all those active fees. So we really think about this of, hey, we're going to provide the best products on the market for the user, which 99% of our users don't have the best products on the market today. They could switch to something better. And then two, where are the opportunities and where there is a need for market competition? We need innovation. We need to build better financial products and it needs to come from technology companies. It won't come, it's not going to come from retail banks. It's not going to come from legacy players that are, you know, got the regulatory capture and are very happy with their extractive systems. It's going to come from people like me and people that are going through YC today. And it's going to come from founders building great new technology companies because the world gets better with innovation. You know, we're on the Derby Mill podcast. How has the world got better over the last 200 years? It's technology and innovation. And I think market forces and better products are so critical to that financial health journey. So everything we do is thinking about that. So then if I go back to, but how do you imagine actually, so then am I hearing you saying that the way you want to charge for Clio is sort of a subscription model with no dollar charges on? Like, we don't make any money off of any other bid. You're just going to charge users X dollars for Clio. Is that how you're conceptualizing it? Yeah, today, three tiers, depending on some different jobs to be done. You've got one just the AI tier. So just like the chat GPT example, better reasoning models, more tokens, voice, people pay for that. And then we've got two others in financial products. Thanks. And they pay just a clear, transparent transaction fee for it. No APRs, no hidden costs, blah, blah, blah. And the most important thing when we stack our products up versus the chases of the world, are they cheaper? Is it cutting into our fees for our users? Is it cutting into any alternative forms of credit they may have? And the overwhelming answer is yes, and we wouldn't be doing it if it wasn't. So I think subscription business models are here to stay in the future. But I can also see a world where microtransactions and a more transactional-based view of intelligence and a view of outcome-based, if she takes a task for you and achieves something for you and gets you a saving, there's definitely a way to monetize in that way. So I think as commerce becomes cheaper with lower payment costs, you can also do more of that, which you can obviously do in the East that we can't do in the West today. Did that make sense? That was a very long answer as well. I keep getting long answers. I think I'm trying to parse because I think the subscription model is a clean transaction. The we're going to give you better products has always been the story of financial services to date when we bundled advice with it. And that is, I'm not saying you guys have gone down that path, but I'm saying the market forces have always led that down the path of products that look cheap. After all. Let me just give you the inherent inconsistency of any advice product, whether in finance or anything. The person being advised doesn't know what is good advice. That's the inherent market failure. If you're going to build a market around giving advice, you now have to somehow wrestle Because it is just as... compelling to give there's we did an audit study a long time ago, but where if you give people the certain kinds of advice that's transparently bad, they find that advisor to be much better than an advisor who gives them but but it makes sense like you wouldn't need advice if you knew like, there's a reason that we have the FDA like the market for medical advice by itself is not going to settle into the perfectly right equilibrium. So I'm only saying this because Because in some sense, there's some... There's market economics underneath this that struggles, and it seems like there's a technical problem over here, but there's a sort of pure market problem here that stands in the way of the technology. That's why I think I'm asking these questions, but your answers were helpful for me. Thank you. No, just a slight add to it though. It's like, it's such a good question. You can see this, the build out of all the competitors that I have. I look at Robinhood and like how they monetize and how that, how you think about the business model in a company like that. It's very, very interesting. Right. And it all comes back to us for this concept of financial health. And like, if the advice actually ladders up to the objective function and is that objective function true. So you've got to be able to audit those objective functions and you've got to be Hey, is Bonnie actually trying to improve financial health there? And how much is he improving financial health? When Instagram is serving you your feed, what is the objective function there? Like, that is the real question a politician or an academic should be asking. But as an inherent... You've got to trust me unless I can be transparent and you can audit the actual objective function, right? All right. Can I build on that for a second? Rich, you're going to like this. The most innovative thing I've seen in this space that I thought was really interesting, and I haven't followed up on what happened to it, but when HIV was quite big, there was a South African insurance company that sold life insurance to people with HIV because the mortality rates were very high. And their business model was they would sell life insurance, and then they owned the behavioral problem of adherence to the HIV medication. So their model was they would sell life insurance, and then they had this whole behavioral team that would make sure that you got onto the antiretrovirals and took the antiretrovirals. So in some sense, it's like perfect alignment. It's what you were saying. If I could own the outcome. Then we're in like really good position. It's like, if you could, like, if imagine your financial product was, I am going to give you social security. I'm going to be the provider of social security when you retire. Great. Now I own that problem. So now I'm perfectly incentivized in the middle. Now in some funny way, it's almost like the problem of finances. It'd be nice if you could transact and truly own the final outcome, the long-term outcome. But we rarely transact on the long-term outcome. We often transact on these very Proximate outcomes where there's a lot of chance for messing around. We're getting us distracted. You should take over and say this is not where you wanted it to go. No, I think I think Sendel, that's a great setup for where we'll go next. And then give Barney a break. He gets to listen and and and we all weigh in on where does this go at the limit? And and I'll just. You know, begin this by the very final sentence. There was something that when we did our last session with Toby from Shopify, at the very end of that podcast, we were talking about goals and we were saying, you know, how he had this great data because, you know, so many companies are doing things on trying to optimize on advertising, but he's got data on actual sales. And, um, And at the very end of the podcast, in his sort of last sentence, he said, oh, by the way, we're not optimizing on the individual transaction, we're optimizing on lifetime revenue. And that one sentence completely changes the nature of the interaction, because it's, you know, it is thinking it's determining a different goal and, and central that goal is very aligned with what you just said. And so if we are now I'm moving to part two of this, which is what happens when, you know, as as machine intelligence advances towards the limit and we think about, you know, the Barney's opening comments about financial health as the long term objective. And so if that's the objective. And the machine intelligence gets better and better and better. And as Barney described, the data that it has in order to make its recommendations is becoming more and more high fidelity. Then we can think about either what does this look like at an end state or even the transition path from where we are today to that. So we'll go back to our original order. Niamh, anything you want to do in weighing in on where this goes? And then Rich and Zebel. Yeah. Absolutely. And again, it harks back to the holy grail that we were all trying with the first robo-advisory wave around 2014 of how do you get like a CFO, a private banker, a wealth manager in your pocket. But the challenge then was we didn't really have all the technological tools to achieve it, where now it's becoming within the realms of possibility, plus folk have become more familiar with all these mechanisms that they're willing to kind of engage and trust it. So it's becoming a a more tractable problem. I think you can look across it, not only across the entire journey of the individual, but holistically at a financial level as well, almost optimizing the P&L, the balance sheet and the cash flow of the end of one user based on your immediate needs, short, medium, long term goals. So you'll have like the compliance side, which will be, OK, we know where you're based. We know the tax deadlines in that country. Identify whether they need to file or not and also draft it for approval. Then you have the saving or the cost side, which I think we've already touched upon, as to how can you expand the product suite to incorporate live events and broader goals. Because as Barney alluded to, half the battle is awareness. It's kind of crazy that at school you're not taught Legal and financial health as like a 101 from the start. So, you know, they discover that you've got your first child. Are you aware that you're entitled to these benefits? And there's these tax efficient products and like college funds, etc. Or you're hoping to buy a home. Let's see if we can improve your credit score with renting in the interim and then help find a mortgage. So, there's all these passive mechanisms you can do, too, alongside these short-term Simple optimizations that you can do with capturing discounts and optimizing subscriptions. And even growing with the ecosystem as a whole, on the revenue side, if folk are cash strapped, you can look at additional revenue streams such as... Just take photos of items you no longer want and Clio kind of uploads it to Vinted or eBay to sell it or you're going on vacation, you can auto-rent it on Airbnb whilst you're away. So you've really got an ability to play with every single lever in accordance with that holistic objective function. And I think that's now possible in a way that it wasn't before because everything has become online, the open banking system's kicked in, There are all these connectors and tools and mechanisms where you can adjust various accounts and auto-spin them up in an agentic manner. So, yeah, it's super interesting. Excellent. Thanks, Niamh. Rich, and then Barney, at the end of this, we'll offer you a chance to react to any of the points that are raised. I think the first order, it doesn't look that different than what... Clio does now. I mean, it will be interacting with users, figuring out what their goals are, and then helping keep track of things and motivating them to pursue their goals. Obviously, you know, you can extend this, but that's not probably that different from what Barney's thinking as well nowadays. The extensions would be things like helping the user figure out what their goals should be. You know, they thought they were going to just try to save for a house but maybe they should also put some aside for savings or for investments or maybe buy some bitcoin or because you can't really trust the current financial banks and stuff Good stuff. so helping to set the goals and also extensions into the lines of What they might do about it. Like they might say, oh, you need to go back to school and learn another job that will pay more. Or you might need to quit your job and you really enjoy some other life. Thank you very much. So those directions are natural extensions. But in a sense, it's the same. It's interacting with the user, figuring out what the goals are, and helping to achieve them and motivate towards them. So I think it's all a very human activity and as such, it may not look that different at the limit. And Rich, can you just describe for listeners when they're thinking about this business in the context of the age of experience, what does experience mean in this context? And how could someone like Barney? Generate more experience data in order to move on that trajectory. Okay. I am sort of looking past that. Assuming, I mean, the major source of experience is the conversation. What works, what doesn't work, what engages the user, what loses customers. There is a There's a lot more. Or, uh... I'm one of the guys that's kind of skeptical about large language models. I think they... are not as strong as they should be, but sometimes I thought Barney was also looking past that. We could, we'd have better interactions and better development. Maybe there'd be less manual tweaking of the, less engineering that the engineers at Clio have to do to get a good interaction. Yeah, the interaction, the experience would be, I said these things and And they got the guy to save some more. Or maybe the whole interaction is full of data that he is harvesting. Barney's harvesting towards the reinforcement objectives. But they're all kind of short-term objectives. They don't have really long-term objectives. Those are not intertwined the way they should be. That was a lot of words. Maybe we'll get a better understanding as we go along. Okay. Rich, let me come back to you one more time on this, which is, you know, in your initial response, you suggested Barney's system starts going, you know, going from the current type of interventions, which are like, for example, maybe you can order in this Saturday night rather than going out, you know, because it looks like you're otherwise going to exceed your monthly budget. To maybe you can quit your job and go back to school. And you know what I mean? Now, that that is very much widening the aperture and and it's making all of these recommendations moving you up a gradient with regards to your financial health objective. And it feels like once you are starting to make those types of recommendations, that in other words, now you're a big span of decision-making is being driven by optimizing financial health. And then you start worrying about the competition of other goals that are not, you know, if they're not using another application that's worrying about health, long-term health objectives and long-term happiness objectives, that, you know, you're kind of guiding the person financially driven lifestyle. So we can come back to that. I would think that the financial application would have to enlarge its scope to include some of those broader objectives, like they do already, you know. Okay, Sendhil. So let me try an analogy. I think my impression of the limit, I think, looks radically different than where we are now. So let me try it by way of analogy. So let's think of, you know, finance Through the lens of, say, Getting from place A to place B, but instead of a financial place A to place B, you were trying to get physical place A to place B. So, and let's suppose that in the limit, we start having self-driving cars. Well, at that point, I don't need driver's education anymore. I don't need to know the rules of the road. But then, once we've done that, why would we continue to use the same rules of the road for a bunch of self-driving cars? That were entirely designed for people. So the rules of the road should change too. I mean, it would be absurd to keep the rules of the road the same. It wouldn't make sense. Like you have a yellow light because it takes time for people to react to the thing. Why would we have lights? Why are we spending it? The whole thing, all of traffic would begin to become very weird. We'd start to appreciate like, well, there's all this evidence on what causes traffic jams. And we can't do anything about it with human drivers, but we can certainly affect the rate at which cars speed up or slow down on the highway so that we... There's just an endless supply of stuff that now starts coming into action because the very basics... So what does that tell us? It says that most of the machinery of intermediation was designed for people. The same thing is true of finance. What we are trying to accomplish in financial markets at some level is shockingly simple. I am either trying to move money between today and tomorrow or move money from tomorrow to today. And in doing that activity, I'm going to be willing to take on some risk and pay some premium. That's it. There's not much else happening in finance, but because, but then, but right. I mean, that is all finance is. There's nothing else. Like that's it. But because as people... We don't like all those things are kind of complicated. Most institutions are built around the limited cognitive machinery of people and most financial product proliferation is kind of a response to that fairly limited, whatever complexity of it. So I actually think you will start to see it just to make a very simple example. Right now, imagine we just had my agent able to look at credit card offers and suppose it was great at it, perfect at it. Do we think there's going to be mileage credit cards? Do we think there's going to be bonus points for credit cards? Like, oh, spend here. You get these points that you can spend on this. Where's the efficiency in any of that? There isn't efficiency. It's just a, as a human, I'm like, oh, I got a free trip to this, blah, blah, blah. It's like, these are things that are appealing. The whole credit card market to some degree exists because there are things that are appealing to the human purchaser. Maybe there's some deep efficiency and it'll persist, but just if you play with the fiction, once the agents were doing the choosing, that's not what the credit card market's going to look like. The credit card market's going to look like a commodity because all it is is you have a credit score and a risk associated with you. There's a cost of capital of giving you that cost of capital. It is the most commoditized thing in the world. The only thing that makes it not commoditized is customer acquisition and the oddities of customer acquisition. But that cost It's just gone to zero because my agent's doing it for me. So a bunch of these markets are just going to utterly collapse, which I think is a good thing, because I think then we're going to leave the hard problem, which is, I think, the problem you ended on, Rich, which are the kind of hard problems of like, hmm, do I really want to be eating out with DoorDash all the time? Is that worth it to me? It costs so much, but it saves me this hassle cost. Help me understand how much the hassle cost is costing me in the long term. There you go. That's a hard problem. And so I kind of feel like we're going to see this market collapse from what looks like finance, which I think of as just the detritus, if that's the right way of saying the word, of the stuff in the middle, to the hard parts of finance, which is the intertemporal trade off that we as people must make and understanding the cost of the intertemporal trade off. And I think that's what the limit is going to be. It's going to be helping us struggle with those basic questions. Awesome. Settle. Thank you. Barney, the last couple of minutes are for you. Anything you want to react to? I think that was pretty good, Sandal. My big thing has always been, once you have tens, hundreds of millions of users, you radically understand them, you can start to have people and agents bid on those users for the products they want. I'm a user. I don't actually want a credit card. I just want to borrow some money. And I want to borrow $1,000 and I want to borrow it for this thing. Hey, go out to the market agent, all the other agents, interact with the agent, tell them about me, my data, who I am, and then give me the best market pricing for that credit card. And you have just introduced market mechanisms into something completely broken that puts so many people in harm every single day. And if you can do that, you can fundamentally change what all financial products look like. Because you're 100% right. You've got, we're going to spend some money. We're going to try and invest and save some money. We're going to borrow some money. It's really not that complicated. And we can simplify what is the most opaque, painful, you know, burden on society, which I think at retail banks, don't say that to, don't tell Jamie Diamond I said that, and we can turn them into kind of commodities and dumb pipes, which I think they should be. They should be the, they should be the commodity products and like the intelligence and the layer is where the value accrues to. So everything gets dramatically cheaper for the consumer. They make radically better decisions and people improve their financial health in a really meaningful direction, which in An economy and a time and a political environment like we are today, I think, is the big, big thing we need to solve. So I thought some really good thoughts, guys. This is super interesting. Awesome. Well, thank you very much, Barney. Real pleasure to have you here. And congratulations on the success of Clio. You know, it's. It's a remarkable business. And you've got the kind of data and it seems like you've built the kind of trust with your users. To actually, you know, move us towards a world like the kind that Neve and Rich and Sendel were describing. Thank you so much, guys. This was really fun. I appreciate it. All right. Bye, guys. Thank you, Barney. Thank you, Rich, Barney, and everybody. All right. That's a wrap for today. Thanks to Rich, Sendel, and Neve. And a special thanks to Barney from Clio. We've also posted links to our hosts and guests' social media feeds and webpages in the show notes, as well as our podcast webpage located online at insights.intrepidgp.com slash podcast. Thanks, everyone, for listening. The views, opinions and information expressed in this podcast are those of the hosts and guests and do not necessarily reflect the official policy or position of Intrepid Growth Partners. This content is for informational purposes only and should not be considered as financial, investment or legal advice.