Why Upstart Is Building a Bank From Scratch Instead of Buying One With CEO Paul Gu

Paul Gu dropped out of Yale in 2010 to join the first class of Thiel Fellows, spent time at the quant fund D.E. Shaw, and then co-founded Upstart on a simple premise: the techniques Wall Street uses to price corporate risk should work at least as well on consumer credit. Fourteen years later he took over as CEO from co-founder Dave Girouard, and six weeks after that the OCC granted conditional approval for Upstart Bank. This conversation covers what has actually changed at the top of the company, why Upstart went for a full de novo national charter rather than buying an existing bank, and where AI is reshaping the parts of lending that nobody talks about.

What We Covered

• Dropping out of Yale for the first Thiel Fellowship class
• What D.E. Shaw taught him about applying quant techniques to personal finance
• The income share agreement idea that brought the co-founders together
• What changed when he took over as CEO on May 1
• Losing the balance of a three-founder culture, for better and worse
• The core personal loan business and the future prime borrower
• The trifecta of growth, profitability and credit performance
• Auto and home, and the race to contribution margin positive
• What conditional approval from the OCC actually means
• Why a de novo charter rather than acquiring a bank
• Where the existing bank and credit union partners land after Upstart Bank opens
• AI in loan verification and servicing, beyond the underwriting model
• What happens when AI agents start applying for loans on people’s behalf
• Ninety-one percent automation and whether 100% is reachable
• The case that fixing the cost of credit makes most Americans 10% wealthier

Key Takeaways

• Upstart went de novo rather than buying a bank because acquiring one means inheriting someone else’s loan book, underwriting practices and operational history, and the whole pitch depends on being able to stand behind every decision inside the bank when a regulator asks.
• The bank changes who originates, not who funds. Upstart Bank will become the principal originator, but Gu is explicit that the company is not becoming a large, equity intensive balance sheet business, and the bank and credit union partners keep buying the assets.
• The next AI wins are in verification rather than underwriting. A HELOC can carry several thousand dollars of human verification cost because county property records are non-standard and non-deterministic, which is exactly the kind of work a generalized reasoning agent is suited to.
• Gu welcomes a world where AI agents apply for loans on borrowers’ behalf, because agents have unlimited time to search and no brand loyalty to defend, which favors the lender with the best rate rather than the biggest marketing budget.

About Paul Gu

Paul Gu is co-founder and CEO of Upstart, the AI lending platform he started in 2012 after dropping out of Yale as one of the first Thiel Fellows and spending time at the quantitative hedge fund D.E. Shaw. He spent most of his career as the technical half of the founding partnership, running product, engineering and machine learning before taking general management of the auto and home businesses. He succeeded co-founder Dave Girouard as CEO on May 1, 2026.

Cleaned Transcript

Paul (00:10): I always say that there’s kind of this trifecta in credit. It’s hard to have high growth, high profitability, and good credit performance at the same time. And I think history mostly shows that when lending adjacent businesses are growing fast, something else is going wrong in that triangle. And I think if you look at our history in this core personal loan segment, we’ve been able to achieve all three of these things. And you know, today, if you look at that core business, it’s growing extremely fast, you know, about as fast as the whole business, which we’ve guided to this secular thirty five percent annualized growth rate at pretty large scale. So really fast rate of growth. This is a business that has over sixty percent contribution margins. And of course we’ve been doing it a long time with a lot of investors who are really happy with the credit performance.

Peter (00:54): This is the Fintech One-on-One Podcast, the show for fintech enthusiasts looking to better understand the leaders shaping fintech and banking today. My name is Peter Renton, and since 2013, I’ve been conducting in-depth interviews with fintech founders and banking executives. My guest today is Paul Gu, co-founder and CEO of Upstart. Paul famously dropped out of Yale in 2010 and became part of the very first Thiel Fellowship class, spending time at a quantitative trading firm, D.E. Shaw. He then teamed up with Dave Girouard to found Upstart, bringing the kind of technology used in quantitative finance to consumer credit instead. Dave stepped down as CEO earlier this year and now serves as Executive Chairman, with Paul running the company day to day. In our conversation, we talk about what has actually changed since he took over the top job, how Upstart describes itself today across personal loans, auto and home lending, and the recent conditional approval of Upstart Bank from the OCC and what’s left before it can open its doors. We also dig into how Upstart is using AI beyond just coding, particularly in loan verification and servicing, and what happens as AI agents start applying for loans on people’s behalf. And Paul also gives his theory that fixing the cost of credit could make Americans 10% wealthier. Now let’s get on with the show.

Peter (02:29): Welcome to the podcast, Paul.

Paul (02:30): Thanks Peter, excited to be here.

Peter (02:32): Great to have you. So let’s kick it off at the beginning, shall we? We’ve known each other for a long time and met you soon after you co-founded Upstart. But go back to, you know, you were at Yale, you dropped out, you took a Thiel Fellowship. Tell us a little bit about that time period and also like how you met Dave and decided to dive into consumer credit.

Paul (02:55): It was 2010 when the Thiel Fellowship first came out. This was a new program at the time. It was buzzy. It was in the news. I actually read about it on TechCrunch. And you know, I had had this idea that I wanted to start a company at some point. I wanted to be an entrepreneur, but really I was pretty far away from that world. I think Yale at that time in 2010 didn’t have a ton of tech. It had a lot of people going into consulting and finance, but it didn’t really have all that much tech. It had relatively limited ties to Silicon Valley. And for me, the Thiel Fellowship was like, wow, this is just such a great forcing function to get serious about this thing that otherwise I don’t know how I would get from point A to point B. That was a real fortunate opportunity that that came up in my life. And so I decided to go for it. I dropped out of college, and then I spent a little bit of time at this quant fund called D.E. Shaw, just kind of like messing around, learning. And, you know, the thing I noticed there was that there were some of the smartest people I’d ever met in my life. Like, you know, you’d have all these people who were like world champion bridge players or chess players, people who were just really, really good at thinking. And they were using really modern novel computation techniques, algorithms, but they were doing it all in service of what seemed to me to be a fairly narrow problem, which is this problem of how you arbitrage securities prices to just be a little bit more efficient. And while intellectually, I mean, you just could not have more fun at a place, it just seemed like such a niche problem that maybe didn’t have a lot of relevance to most real people. And what I got really excited by was thinking about, well, could you essentially do the same kinds of things being done in the world of quantitative Wall Street, but do it in a domain that would matter to a lot more people. And so pretty quickly, you know, you take just a half step from what I think of as corporate finance to personal finance, you say, well, of course, like if you can do something in personal finance, it’s going to be relevant to a whole lot more people. And once I started looking over there, I was like, wow, it doesn’t really seem like there are a lot of people that are doing similar things, that are taking the same kinds of technologies, the same kinds of algorithms, and doing what they’re doing at D.E. Shaw, but on the personal side of finance.

Paul (05:15): And as a result of that, there are a lot of people who are structurally underserved in credit, mispriced, overpriced, et cetera. And that combination to me, like, here’s a technology. We know the technology works really well. We know it mints money when you apply it on the corporate side of finance. And this is a huge opportunity, lots of people being underserved and not a lot of people going after this application of technology here. And so that’s kind of how I ended up in the world of consumer credit.

Peter (05:46): And then how did you meet your fellow co-founders?

Paul (05:49): Yeah. So as I started thinking about this, I had another, at the time I guess it was unfortunate, but of course later turned out to be quite fortunate, little twist in the story, which was that I thought, well, okay, if we can do this, the very best product packaging to put around our kind of sort of machine learning powered personal finance product is going to be something we call an income share agreement. So I thought, well, if we can sort of figure out these structurally mispriced consumers, what we should do is we should give them some money and say, you owe us a certain percent of your income over the next five or ten years. And so we dubbed that thing an income share agreement. And I was really excited by this. It was kind of like the closest thing to an equity structure, but for people. And this idea was very niche, of course. And so not a lot of people knew about this. And I was trying to meet people who knew something about doing this. And there were just a few people in the United States who had ever done anything resembling an income share agreement. And my co-founder Dave, he was at Google and he had just decided to leave Google and start an income share agreement company. And so he was much in the same mode of trying to find these few people who knew anything about doing this. And we both got connected to the same woman and she introduced the two of us. She said, you know, I do know something about this. I think it’s a really bad idea. No one should do this idea. But if you insist, then here’s this other crazy person I just met who also seems to want to do this idea. And so we ended up getting introduced. We spent some time messing around with the idea together. Turned out we had two sort of complementary versions of the idea that we ended up mushing together. And that’s how Upstart got started. So yeah.

Peter (07:31): I remember you did that for a couple of years, and then you pivoted into personal loans, which obviously has been the mainstay of your business for more than a decade now. But so then, you know, Dave obviously stepped down as CEO. I think it was May 1st. And you know, I’ve had him on the show a couple of times, and he’s obviously very well respected in the fintech space as a CEO. You obviously have been watching him and working with him for, yeah, since the beginning. So tell us a little bit about what has actually changed about how you spend your time now that you have taken over the CEO role, and what’s been some of the hardest challenges for you?

Paul (08:13): Yeah, I mean Dave and I have been talking about and planning for this transition for a pretty long time. And obviously, you know, we worked in really close partnership for a large number of years. I mean, really, essentially, my entire career, my entire adult life was spent watching and learning from Dave. So in one sense, it doesn’t feel like a big change. I think it’s sort of been like we’ve been on this journey together for a long time, and now that journey just continues. I think we always knew that it was going to be a very long journey to do what we want to do. I think at least after the first couple of years of the company, when we kind of outgrew our sort of initial just naive enthusiasm about how quick this could go, we realized like, wow, there’s a lot that you have to both actually build and then prove to the world before you can really do this at scale. And it’s going to take a very long time. But fortunately between the two of us, we probably can span a fairly large number of years to bring this vision into reality. So in essence, I think the transition has long been planned, has felt really natural and like a continuation of the vision. I think probably the place where it feels more different is, you know, we started the company with three co-founders, over time became two, and now day-to-day operating is just me. And I think that that is different in that probably culturally as a company, I think in the earlier years, if you ask me to describe what’s the Upstart culture like, I would have focused on a lot of the ways in which it was a really balanced culture, where Dave, Anna and I, we worked really well together, but we were pretty different people with sort of really different kinds of starting perspectives, really different life backgrounds. And as a result of that, I think the culture revolved around a certain kind of balance. And now, for better and worse, I think it’s a little less balanced. Probably we’re a little bit able to move faster on certain kinds of decisions as a result. And I think we’re more sharp and more decisive. But, you know, on the other hand, probably less balanced. And I think that is something that I’m working to make sure that, you know, we can get the best of all worlds.

Peter (10:17): Right. Okay. So Dave is obviously still involved, he’s Executive Chairman. I mean, how often do you chat with him?

Paul (10:22): We probably talk once a week, like fairly often.

Peter (10:26): Okay, but you’re doing all the day to day now, it sounds.

Paul (10:30): That’s right.

Peter (10:32): Okay. So then let’s take a step back, because Upstart’s changed a lot since the early days and you’ve got a much broader product suite today than ever before. So maybe just start with, how do you describe Upstart today?

Paul (10:44): Yeah. So there’s probably just a few important things to know about Upstart. The first and most kind of basic building block is that our core business, that is a sort of mature business that is sort of the foundation on which we are building the rest of the company, is what we call core personal loans. Core personal loans means two things to us. One is that it’s personal loans, of course, self-explanatory. But the other is that we really have a core customer. And for Upstart, that core customer tends to be what we call a sort of future prime customer, or a customer that typically is not regarded by industry as super prime. Now that’s not to say we don’t serve super prime people. We actually serve borrowers probably across the entire economic spectrum, broader than any other player out there. It’s a really great point of pride for us. But our core customer, the one that we spend the most time thinking about, the one that I think we are most differentiated in our ability to serve, is a customer that most of the industry would regard as more difficult to underwrite, not someone who has the sort of like a top quartile FICO score, things like that. And that borrower for us is one where you know, you have to be really good at underwriting to actually understand the risk and properly separate the risk. So you don’t get adversely selected. So you don’t end up with a lot of volatility in your credit performance. And that’s kind of the place where we’ve invested the most time over the years to get really good. And as a result of that, we think we have a very large amount of technology differentiation in serving this customer that translates through, and you know, you can see that in our margins in the segment. I always say that there’s kind of this trifecta in credit. It’s hard to have high growth, high profitability, and good credit performance at the same time. And I think history mostly shows that when lending adjacent businesses are growing fast, something else is going wrong in that triangle. And I think if you look at our history in this core personal loan segment, we’ve been able to achieve all three of these things. And, you know, today, if you look at that core business, it’s growing extremely fast, about as fast as the whole business, which we’ve guided to this secular 35% annualized growth rate at pretty large scale. So really fast rate of growth. This is a business that has over 60% contribution margins. And of course, we’ve been doing it a long time with a lot of investors who are really happy with the credit performance.

Paul (13:06): And we’re able to do that because of that technology differentiation. And so that’s this big, really profitable, fast growing core. And so that’s the first thing to know about the company. Then in more recent years, we said, well, of course, the even bigger opportunity in consumer credit is not limited to unsecured loans. You really need to be in the secured products if you want to be in the parts of credit that are relevant. Not, you know, a big market, but really like the whole market of consumer credit, something that’s almost unimaginably large. And so we have built products in auto and home, starting with auto purchase and HELOC. We also have an auto-secured personal loan. And these products today are in a place where they have proven borrower demand, they have proven product market fit, they’re just starting to get capital acceptance. And so they’re really at the sort of cusp of being able to scale and turn into real businesses. The last thing that they really have to prove before they can do that is their own unit economics. And so the big focus for us is get the unit economics to a good place for these new products. And we’ve, you know, communicated that this year all of these products are going to be contribution margin positive. And so we’re racing towards that right now. And the implication of doing that is we will have a business where in the near term, you know, you have this fast growing, highly profitable core personal loan business. And I think we’re about to prove that we are then going to have home and auto businesses that will also be quite profitable. And of course, given their relatively much larger TAMs, will give us the space to continue compounding and growing at those very high rates for a lot more years to come.

Peter (14:49): Okay, well, I want to dig into the recent news, which the news that prompted this conversation, and that is the conditional approval of Upstart Bank, July 23rd, by the OCC. Maybe just start off with, for those who don’t know the ins and outs of the chartering space well, what does conditional approval actually mean? And what still has to happen before you can open your doors?

Paul (15:14): Well, there’s a lot of hard work involved in opening a bank, as there should be. And as you know, we’ve gone through the process, we’ve done just countless numbers of diligence processes with the OCC and also the FDIC and Federal Reserve. So it’s really a sort of a multi-agency process. The OCC is the sort of primary regulator whose bank charter it is that we are seeking. And so as you noted, they issued us a conditional approval. Conditional approval basically says that we are approved to open this bank, provided that we meet certain conditions. So they’re just making explicit the list of conditions that Upstart needs to meet in order to open this bank. And of course, these are conditions that we are prepared to meet and working to meet. So there’s a bunch of work ahead, but I would say the plan and the requirements are very clear for what needs to happen from the OCC side between now and when we can open the bank. There is still some ongoing work with the other regulators, which we are doing. And I’d say, you know, if you zoom out a little bit, it’s just all in the interest of like, we want to make sure that we take the responsibility of being a bank extremely seriously. We want to make sure that depositors are well taken care of, that there’s the right level of rigor and safety and soundness in the system. And so the process is designed to do that. And I think it’s doing exactly that.

Peter (16:27): Right. And I noticed you have Annie Delgado, who I know quite well, who is going to be the new CEO, chief risk officer at Upstart. I mean, that seems to me to be sending a signal, right, that you’re really leading with risk. Is that, was that intentional?

Paul (16:45): Absolutely. I mean, Annie is really a unique talent, a uniquely talented leader. And she brings two things. You know, I think she started at Upstart in the risk and compliance function. So she of course knows those more deeply than anybody. But of course, for us, our aspiration isn’t to be just like a normal, another bank that exists. We really want to be the bank that brings sort of the union of AI and credit to the banking system. And so we want to be the face of applying AI to credit to the regulatory system. And I think there’s nobody better in the world than Annie to do that because of all the work we’ve done over the years with every flavor of regulator, national and state. I mean, there’s probably not a single regulator we haven’t spent many, many hours with over the past ten years or so that we’ve been, you know, at scale and engaged with them. And Annie has led the charge on all of those fronts. So she both knows about what it means to apply AI to credit in a way that’s responsible, thoughtful, safe, fair, et cetera. But also she knows Upstart and she knows the regulatory partners. And so we absolutely thought that she would just be the very best person to run the bank side of our business.

Peter (17:58): And then why do de novo versus acquiring a bank? Did you go and do some research down that route or did you just say de novo all the way?

Paul (18:08): You know, we considered the different possibilities, but I think pretty early we had a preference for going de novo. And mostly just because, again, like for us, this isn’t about just like having a bank arm or it’s there being another bank. It’s not principally for us a sort of financial strategy. It is like we know that the nature of what we’re doing in lending credit is really different, that we are going to be much more successful if we can have a direct relationship with the regulator and representing exactly what we’re doing. And we want to be able to stand proudly and represent all of the things that are happening inside our bank and our bank holding company. And I think the idea that, you know, if we were to start with something that already had a long legacy, there were a lot of loans done a certain way, underwriting practices, operational practices that we couldn’t necessarily stand behind, it would really muddy the picture of like, you know, what is this new thing that we’re doing? What are these underwriting models that we’ve developed? What are these kind of like AI enhanced lending practices? And I think that story would get confused. And we really want to be able to just unambiguously stand behind every single thing that we’re doing at this bank and say like this is the future of credit, and be able to represent that to the regulators.

Peter (19:23): I want to talk about funding loans for a minute, because you’ve been working with banks for a long, long time on the funding side. And your CFO said on your last earnings call that you expect to move the bulk of originations over to Upstart Bank relatively quickly. So where does that leave your bank partners and how’s this going to work?

Paul (19:46): Yeah, so the vast majority of our bank partners are partners that we will continue to work with closely. All of them, of course, read in on this change and have been expecting it for some time. And with most of our bank partners, what they’re interested in is they want to have great risk adjusted assets, and some of them want to have an ability to build and deepen relationships with these customers. And neither of those things will change. It is the case that Upstart Bank will become the principal originator of the loans. But then after the loans are originated, we aren’t changing our strategy with respect to how loans are ultimately funded. And so we aren’t going to become a large, big balance sheet, equity intensive business. And that means that we are still going to be working really, really closely with all of our capital partners, including a large number of banks and credit unions that are looking for great returning assets. And so we’ll provide those, and then certain of those partners are also interested in ways they can deepen their relationship with these customers, and we’re very happy to facilitate that as well.

Peter (20:50): Do you have any sense on timing? I know you can’t give me a date, because with the government it’s, you know, it’s up to them, but internally are you preparing like mid 2027 or what are you thinking about?

Paul (21:01): Early part of next year.

Peter (21:03): Early part of next year. Okay. That’s great. Okay. So I want to dive into AI, because I know that’s really a thing that you’re very well versed in. I mean, Upstart made their name, I feel like, in the fintech lending space as a pioneer in AI, bringing, I think you were the first one to really be using machine learning at scale. And so you’ve been doing this for a long time. But what I’m curious about is obviously the last several years we’ve had the rise of generative AI, now more recently, agentic AI. And I just want to know, with given your skill set and what you’ve developed inside Upstart, how else are you using these technologies at Upstart?

Paul (21:47): There’s like a whole bunch of things, but then I would say maybe the most interesting ones I think are all about the process of verifying loans. I think that’s where it gets really interesting. I think there’s some interesting stuff in the servicing of loans. But if we talk about verification for a moment, you know, you think about something like HELOC, any kind of secured product, I mean it has a lot of steps. And a lot of those steps are steps that historically people have had to spend a lot of hours on. I mean, if you look at a typical bank or credit union process for originating a HELOC, I mean, there’s several thousand dollars of people verification costs that are involved because of the number of hours dealing with property records and, you know, local government records and things like this. And that work is not very suitable for maybe traditional forms of automation because some of it is non-standard data elements. Like if you go to the county and you get some records about the property lines, like there might be like a hand-drawn PDF or something. And that’s just going to look different from one county to another county. And so it was always like, okay, well, we just have a lot of people spend their time kind of manually trying to verify these things and go back and forth with the borrower and so on and so forth. And this task, because it’s a little bit non-standard, a little bit non-deterministic, but still, of course, a task that is highly trainable, that you know, if you see enough examples, you kind of know what to do, is a perfect thing for a generalized reasoning agent to work on. And so I think that’s a really interesting place. I think it’s an opportunity to move the needle in an economically significant way on the price of the credit, in a way that of course you can ultimately pass down to the customer, and you can make the process much faster, real time, you know, just like all the benefits of traditional automation, I think you can now do to non-standard, non-deterministic problems. So I think that’s really interesting in verification. I mentioned, you know, I think servicing is interesting because in servicing, you always want to be able to help your borrower make the right sort of choices as much as possible. But you know, it’s like rich people have financial advisors, but most people can’t afford to have a financial advisor. Like the math doesn’t really math for doing that until, of course, like the advisory work can be provided by AI.

Paul (24:01): And then suddenly, you know, you could have really, really good advisory and also something that’s accessible to everyone. I think this is a little bit still further out than where we are today, but I think it’s something that is coming. And I think it’s going to make people much better able to make the right decisions for themselves. And you just look at what happens in loan repayment. And you know, it’s just, people are faced with a lot of complex choices of like which loan should I pay first? When should I pay what in what order? You know, maybe like finances are getting a little tight this month. What should I do? And I think those sorts of questions are things that we’re going to see a lot of leverage on AI from.

Peter (24:41): So then let’s extend that a little bit. And even now, just any of the frontier models, you can be your personal financial advisor if you give them enough information. But what I want to know is how you’re thinking about this, because soon, and it may have already happened, you will have an AI agent apply for a loan at Upstart. And soon that may happen at scale. How are you preparing for that? How are you thinking about that future?

Paul (25:08): Yeah, I mean I think you’re right. It may well have happened. I mean, I think agents, that people can empower their agents to take almost any kind of action on their behalf now. And I think that’s great. I think we’d welcome that world. I think it may be here and we, you know, wouldn’t even necessarily know. And I think that would also be fine. I mean, as long as, to me, as long as an agent is obviously legally acting on behalf of a person, and they’re just helping the person kind of smooth process and make things easier, then I think that’s a great win. And I do think that it will usher in a world where there’s sort of a greater ability for people to shop across different providers. And so then you get more objectivity in who wins. And that’s a world we’d be very excited about, because we think we do objectively have a combination of best rates and best process that is very attractive. But, you know, maybe what we don’t have is as much like sort of built up brand recognition from many, many years of doing sort of large scale marketing. And I think historically, because people were in short supply of time, they just couldn’t afford to always find the very best option for them out there. And I think in a world with agents, agents have no such limitation on their time. And I think agents are going to be able to really like search wide and deep on behalf of their users to find the very lowest prices out there in the market. And even if, you know, sometimes there’s differences in process, whatever, if they can help them navigate that. And so I think it’s going to make the market more efficient. And I think that is better for the consumer. It will lead to better financial outcomes, lower prices. And I also think that’s a world that Upstart is very well suited to live in.

Peter (26:48): Are you thinking about how, like, eventually there will be an AI agent communicating with an Upstart agent? Is that on your product roadmap? I mean, what are you actually doing internally?

Paul (27:01): Yeah. So I don’t think that you need the agent to communicate with an agent because, again, this is like, agents are really good for doing generalized reasoning work when the sort of process is non-deterministic. For something like, you know, filling out an application for credit, that is deterministic work. So you actually just want the agent to be able to interact with an API. Really frankly, I mean the agents are going to get very good at web control. So that’s why I say like you don’t even have to know that it’s an agent, you know, it’s just like Upstart, you just build for the consumer. Obviously, at some point there’s enough agents that you want to make an agent optimized workflow that’s just like sort of faster, will not burn as many tokens on the user side. And of course, you may want to be in the agent game yourself and offer that as an option. I think that, to your point about connecting up data sources, I think there is, you know, consumers don’t want to have too many agents that have access to too much of their sensitive data. And so it’s like, well, if I’m already going to be sharing my sensitive data with a particular company that I’m doing sensitive business with, like Upstart, then, you know, maybe that’s a good place for an agent to be hosted. And so I think those are some of the interesting questions. I definitely think though that we are going to support a world where lots of agents are shopping and taking action on people’s behalf.

Peter (28:10): So let’s go back to today, and I think I read where you said now 91% of your loans are now fully automated. That number I’ve been following over the years, it keeps going up. Are we ever going to get to a hundred, do you think? And what would that take?

Paul (28:25): There’s sort of a push and pull on this. I think when we say ever, ever, I’d say probably yes, but it’s going to be a really long time. It’s going to be a really long time because there’s still a lot of new stuff for us to do, new frontiers. And each new frontier, you kind of like reset the bar and you start out a fair bit lower. And so if you were just to look at our kind of like mature core personal loan business or things adjacent to that, we’ve got really high rates of automation. But then if you look at some of our newer products, like the secured products, something like the HELOC, I mean, it’s dramatically lower. And that’s because both it’s new to us and so we aren’t as familiar with some of the things, and it’s because those frontiers have harder problems. So, you know, the fact that you also, in addition to verifying things about the consumer, you have things to verify about the property or the asset, the car, the home, whatever it is, like those things are both new problems and harder problems. And so you have to work on those things. But I actually think, I mean, I think fundamentally, especially, you know, back to this point about like general reasoning agents. I mean, you have the ability to do anything that a human can do with AI. And as long as that’s true, then all of it is automatable. You just have to sort of do the work and make sure that your fidelity is high enough. And that’s just, you know, working down a backlog of work. And that’s our favorite kind of problem. It’s just like give us a long technical roadmap and let us execute.

Peter (29:44): So one of the things I’ve heard you say before is that you think that, I think you might have even said it already, but most Americans are either mispriced or over verified in credit. And fixing the cost of credit can make every person in society ten percent wealthier, which is a massive number if you extrapolate that out over the GDP of a country. What do you mean there? How do you back up that number?

Paul (30:10): Yeah, sure. So the ten percent number, so I think it is a number that is approximately the right number when you look at maybe most Americans sort of by count. I think, you know, like many things in our world, there’s some amount of kind of these inequality effects. So, obviously, I don’t really have a plan to make people who are already very wealthy ten percent wealthier. That’s kind of not the business that I’m in. So I have to cut off that part of the distribution. So, you know, if you’re already a person who is in the top one percent, Upstart’s probably not going to make you ten percent wealthier. Maybe someone else will. Maybe there’s sort of, you know, a wealth management business. That’s obviously not the business that I’m in. But for most Americans, like if I think if you were to look at maybe like eighty percent of the population are fairly regular users of credit. And for this part of the population, the price of credit really matters. The ability to access credit when you need it really matters. If you look at the percent of people’s discretionary income that goes to paying interest, and this is probably your closest proxy, you know, in the sort of lower parts of the income distribution, I mean it can reach like 15% of the income going to just cover the sort of monthly interest service costs. You know, that goes down a little bit as you go up, but you get the sense. Like we’re talking in the range of 10% for the sort of bulk of Americans. And so then it’s like, okay, well, what could you lend to people at if you were a lot more confident about who would pay you back? Of course, the basic empirical fact is that the vast majority of loans are successfully repaid. Of course, otherwise, you know, none of us would be in business. But it’s like maybe 80% of consumer credit is successfully repaid. And that means that 80% of the time, if you actually just were to know that ex ante, then you would have been able to lend to them at a dramatically lower rate, something like the risk-free rate. And if you just quantify the difference between the risk-free rate and the rate that’s being paid now, I mean, that’s basically where you get the gap.

Paul (32:33): It’s like, okay, I think we could actually, then that’s the theoretical limit, is like we could make, you know, for the majority of Americans, we could make them have 10% more money than they have today. And the way you do that is you get much higher levels of separation accuracy. You drive down the cost of verification. And those are kind of the two things, right? Which is like it costs money to make a loan, and sometimes you lose money because you don’t know who will pay you back. And if you can solve those two problems, then all of that excess interest cost above and beyond risk free goes away. And then boom, you know, suddenly you’ve made everyone ten percent better off financially.

Peter (32:43): So last question. If I get you back on the show, say in five years’ time, what does Upstart look like as a whole?

Paul (32:51): I would want two things to be true. One is I want to unambiguously have the best credit product for every American in every situation. So that means like full coverage of use cases, full coverage of the whole economic spectrum. And then the second thing I want to be true is I want people to know that. So I want Upstart to be the most trusted brand in consumer credit. And that means like not only is it actually true that we’ve got objectively the best product for people across the whole spectrum, but also that that’s something that’s known. People know who we are, they’re familiar with the brand. And I think if you put those two things together, then you’re going to be able to do a lot of good for a lot of people at scale.

Peter (33:30): Okay, it’s a great place to leave it. Well, Paul, it’s always great to chat with you. Thanks so much for coming on the show and best of luck to you.

Paul (33:37): Great, thank you.

Peter (33:43): I really liked Paul’s explanation for why Upstart chose to build a bank from scratch rather than buy one. He said they wanted to be able to stand behind every single decision inside the bank and represent it honestly to regulators without inheriting someone else’s legacy underwriting practices. That’s probably the harder path. De novo charters take years and a mountain of regulatory diligence, but it tells you how seriously Upstart is treating this next chapter. When your entire pitch is that AI can underwrite credit more fairly, you don’t want to inherit somebody else’s loan book or legacy practices. Anyway, that’s it for today’s show. If you enjoy these episodes, please go ahead and subscribe, tell a friend, or leave a review. And thanks so much for listening.