All right. We are going to get going with our next fireside. We are very excited to have Paul Gu here, Co-founder of Upstart, longtime Chief Technology Officer, and more recently assuming the role of CEO. Paul, thanks for being here.
Yeah. Excited to be here.
Before we kick it off, I am just going to read a quick disclaimer, and then we can get into it. Today's discussion may contain forward-looking statements that relate to future results and events, which are based on Upstart's information available as of today and are subject to risks and uncertainties. Actual results may differ materially from these forward-looking statements. The discussion may also include non-GAAP financial measures, which are not a substitute for GAAP results. Please refer to the company's filings with the SEC and its IR website for additional information, including GAAP to non-GAAP reconciliations along with other disclosures. All right. The exciting legalese out of the way.
Go.
Paul, let's kick it off. Thanks for being here. I wanted to start with what feels like the defining strategic message of your first couple of quarters as CEO. You have been very direct that the core personal loan product is really the superpower of the business. It is the highest margin, most differentiated, and that your first priority was simply to do a lot more of it. So how are you doing against that goal so far, and how do you think about the remaining opportunity to continue to deliver growth in that core product?
Yeah, I think hopefully we showed in Q2 that we're well on our way. In Q2, I think we did something like 3.5 x as much growth in this core personal loan segment as we did in the prior three quarters combined. So a real trajectory change in what's growing in the business and what's not. Yeah, like I said, core personal loans is what we're really, really good at doing. I think we're world-class at being able to do risk separation in this product. As a result of that, we can identify a lot of really good borrowers that we can underwrite uniquely in the market. Therefore, we have a lot of pricing power in. So this is a high-margin product. We're very differentiated, and we can achieve the combination of high growth, high profitability, and good credit performance in the segment.
Naturally, we want to do more of it.
Yeah, that makes. So what does it actually look like to change the strategy internally and get the team focused on core personal loans?
Yeah, it's a good question. Really, it is a cross-company effort to grow any particular business. Something that we've been doing as long as core personal loans, it's never any one or two really simple, obvious things. It's like every team needs to orient its goals around doing this as opposed to something else, and there's always trade-offs in what people focus on. So for us, it's like you can think of the steps of the funnel as there's targeting and marketing, right?
There's these questions of, okay, you have a team that's working on building targeting models for different channels. Do you want them to figure out this problem? How do you get 3% more core personal loan borrowers, or do you want them to find something else? I think we were prioritizing the something else in earlier quarters, and we switched the priority back to this core segment this quarter. The same is basically true of every team as you go each step down the funnel, and you can just imagine there's a bunch of different teams at each step. There's the people who then care about, okay, once some eyeballs show up at the website, how do you convert them to applications? How do you convert those into approvals? How do you convert those into what we call rate accepts?
How do you get those people through verification? The problems of what kind of verification problems or what kinds of automation problems a core personal loan user has versus different segments or different types of loan users have are just different. We just put more emphasis for every team to focus on this as their number one thing, and you can see step function change in results.
Yeah. Just before we move on, just to be clear about the stance on the newer products, some of the secured products that you've launched. Obviously growing the core personal, I think you've been very clear that that's the main strategic objective. Have your views changed in any way about the market opportunity or the growth in some of the newer products over time?
Yeah. We definitely made some changes, and I think maybe the meta thing is I really believe in having consolidated focus and really having crisp alignment across the whole company on what are our top priorities, and you just can't have too many of them. Among other things, one example is we paused on our auto refi product. That's a product that we've been working on for a few years. It was a good product. I think it was something that, in some version of the company, we would still be working on. I think it certainly could have had a role, and there's all these different advantages that that product offered. But it wasn't as high potential, high growth, high momentum as some of our other product bets, so we consolidated that one away.
Today, every single bet that we've got left at the company is one I'm really excited about. I think all of them share the characteristics that they have extremely large addressable markets. They're adjacent to something that the company is already really good at. They have high momentum that we're capitalizing on and doubling down in.
Got it. Okay, let's maybe pivot a little bit, talk about the macro backdrop. We, of course, have the Upstart Macro Index, which is kind of your view of macro as it relates to your segment of the consumer. As of September 3rd, that was relatively stable month-over-month at around 1.5. It's drifted up a little bit year to date. So maybe if you could talk a little bit about that metric. How should investors interpret it? I know you made some changes to definitions and phraseology around it. Then maybe talk a little bit about what it's telling you today about the state of the consumer, and then how that translates to the business.
Yeah. The simplest way to think about what UMI is, it's just a linearly interpretable index of how likely a consumer is to default on unsecured consumer credit relative to pre-COVID levels. So a 1.0 reflects the years right before COVID, so you think of this as 2018, 2019, early 2020 type stuff. The fact that UMI is at 1.5 today, it's just a way of saying the same exact consumer holding constant, all the sort of borrower-level characteristics and loan-type characteristics is 50% more likely to default than they were pre-COVID. So it's actually a pretty big number. To your point about the migration this year, we've seen this number drift up 12 points since this spring, and that's quite a lot. I mean, 12 points in our business, it's a pretty significant change.
I think if we had been standing still on all of the investments we make into marketing and automation and approval and underwriting and risk separation, I mean, the business would just be a fair bit smaller. Of course, it isn't because we have made so much progress and had so much focus on those things. But yeah, I think the consumer's been under some amount of pressure this year. I think it's not surprising when you look at the fact that we kind of reversed back into the space of having more inflation than we have wage growth. That's been true probably for the last six months, mapping almost perfectly to the duration of time when UMI's been drifting up. I think you see it in credit card utilization rates. You see it in credit card delinquency rates.
I think whether you are looking at the metrics inside Upstart or outside, I think at this point, it has been pretty clear that the consumer has been, or at least the consumer is always a slightly weird word for us because sometimes we care about the opposite thing as companies that want more consumptions care about. But we would say that the American borrower is under more stress today than they were six months ago. And I think UMI has done a nice job of picking that up early and clearly.
Yeah. Then maybe if you take that conversation more forward-looking in perspective, there are a lot of cross currents out there, gas price shocks, which have been lingering. You have fading fiscal support from tax refunds, labor market that is stable, maybe cooling on the margin. I think you have said the business is fine with a stable consumer, even if it is not an improving one. How do you think about how the consumer could play out over the next 12 months, and then just range of outcomes for the business?
Yeah. Maybe I will start by saying that one thing I think investors get too focused on is the action of UMI in the short term. And we have said UMI is really, really important in the short term. It is really impactful. That is actually one reason that we somewhat stubbornly refuse to give near-term guidance on results, is that we want to be able to take credit seriously as a first priority, and that basically means we want to be fastest and most precise in responding to UMI changes. And if we are going to do that, then of course it is really hard to tell you one month from now what exactly we are going to be doing because in some sense, in the very near term, it is somewhat outside our control.
But I think investors get a little too focused on this when they think about if you are here to invest not for the next three months, but you are here to invest for the next three years, then the rate at which you are compounding durable wins in the business across marketing and automation and underwriting suddenly matter a whole lot more than whether the macro is up or down a few points in any given month. And I think that contrast becomes really clear when you look at where the business is today versus, say, where it was back in 2021. In 2021, we probably had the most supportive macroeconomic climate you could possibly imagine. You had UMIs well under 1.0. You had stimulus checks. You had near zero interest rates.
In spite of the comparison being against that world, in today's world of much higher interest rates, much higher UMI, much more consumer stress, we are doing more contribution profit now than we were back then. How is that possible? It is possible because we have had three years of compounding technology wins across the real durable parts of the business. I think in this any kind of medium to long term that just ends up mattering much, much more. When you talk about where I think the outlook is 12 months from now, I think the answer on the macro is I have no idea. I think of it as really not my job to worry about, is there going to be another war in the Middle East, or what is going to happen to the Strait of Hormuz and energy prices?
I do not know. I have my guesses, but so does everybody in this room, and I do not know that my guesses necessarily are any better than anybody else's. I think the things that are within my control are about how fast can I move the technology frontier on lending. I know that as a company, we have a tremendous ability to execute on that. Compounded over any reasonable timeframe, I have a lot of confidence that the business is going to do really well. That is how we have started to think about where we can give confident guidance in the business and where we do not really want to overstretch in what we can tell investors to expect.
Yeah. Makes sense. I want to hit on one more macro question, and then we will pivot away. I wanted to talk about the supply of credit. I think that is one thing that can impact the market, but really supply of credit from other people in the market. I think Upstart is one of the more disciplined. You just talked about very disciplined culture, wanting to be responsive to what you are seeing in the macro environment. I also think funding markets have been less discriminate in providing funding to anyone out there in the market. I think when everyone competes in the same channels for the same leads, you run the risk that underwriting could move to least common denominator.
How do you think about competitive dynamics in the market that you play in, and how to think about new entrants, less disciplined investors benefiting from a really supportive funding market?
Yeah, mostly we don't think about competitors that much. I think that's a privilege we have because of the spaces where we're really strong. We talked about this core personal loan superpower. I think we're just so differentiated there that we just have the luxury of not having to think too much about competitors or what they're doing. I think in some spaces, there's quite a bit of competition. I think we tend to, all else equal, shy away from those spaces. I think in some of our newer products, of course, being the new guy on the block, we have to think about who the existing players are and who we think we're going to take market share from over time. But by and large, I would say in the core business, it's not a first priority.
I also want to respond to this point about the capital just flowing indiscriminately. I do think, of course, that happens sometimes. But when I look at the capital markets today, I don't necessarily think that there's something fundamentally broken with how they're providing capital to the space. I actually think that over the course of the last maybe five, six years, I think a lot of things equilibrated in the sense that you look at what happened in unsecured personal loans. I think most of the players that did high volumes of internet distribution of large dollar loans to lower FICO score borrowers stopped doing it. It was just too hard to do, and I think maybe they were somewhat affected by the relatively better selection we could do in the market and kind of got pushed out.
I think there's just a lot less happening there, and so therefore, a lot less capital flowing there. In our case, we've had 100% retention of all of our capital partners in recent years. We've re-upped all of the deals that've gotten longer and larger and generally at better terms. I think that's happened because the capital recognizes where the good risk-adjusted returns are. And so I think the capital markets are working pretty efficiently, and I think even if you think it's not perfect, I think there's pretty good evidence that over a couple of years, they tend to get pretty efficient just because they can just look at how the credit performs.
Right. You get the numbers back, yeah. If that were happening, how would it show up in your business? It wouldn't show up in UMI. It's not directly a borrower statistic. Is that just lower conversion rates, people being more aggressive and people not taking the Upstart loan on the margin?
Yeah. I think it can show up in one of two places. I think one is there can be more noise around top-of-funnel marketing. If you put too much money into top-of-funnel marketing, then it just gets harder to get eyeballs and get attention, and borrowers don't really know who's who before they trek. The other would be if you get lower conversion, but in both these cases, I think it's theoretical. We don't see any evidence of that happening in our business.
Sure.
Again, despite the fact that we've seen this higher UMI in our business, I think we've still actually held up remarkably well in our ability to originate good loans.
Great. Let's talk about contribution margins and where they go from here. I think the guidance assumes a pretty meaningful step-up in the back half of the year. We started to see it in Q2. Margins improved nicely quarter-over-quarter. In your eyes, what still needs to happen to get to that full-year target, and how do you split that between continued acceleration in core personal versus the margin profile of some of the new products stepping up and kind of reaching scale?
Yeah. I always think a lot more in terms of contribution profit dollars than the margin number, which I kind of just think of as like, I don't know, it just is what it is. But of course, yes, the contribution profit dollars have to keep going up. It's as simple as that. If you look at any reasonable full-year model for how you get to the full-year guide, we have to have more contribution profits. We've been pretty clear, again, short list of priorities, everyone really focused. It's just two simple things for us. It's keep focusing on that core personal loan segment, which is already very high margin. We would love to just do more of it. The second is get the unit economics of the new products, specifically the secured products, in a good place, get them from negative to positive.
And of course, you get a bunch of leverage from flipping that around. It is as simple as that, those two things, and if we are right about that, you will see it in more contribution profit dollars.
Got it. So maybe can you talk about that second point in a little bit more detail? You saw a nice move in the contribution margin on the secured products. I think you have talked about being break-even by the end of the year, and I think there is a kind of a running list of levers that you have talked about to improve the margins. Can you talk mechanically on how that happens and your visibility into those continued improvements in the back half of the year?
Yeah. We have really high visibility, really high confidence in what needs to get done to move the secured products from unprofitable to profitable. In some sense, it is almost just like the standard playbook for how you build a new business and a new product, which is like the first thing you have to do is you have to prove people want this thing, and you do that before you worry about how dialed in the margins are. You do that before you try to optimize your operations. You do that before you try to optimize your take rates. That is what we did, and I think it has become exceedingly clear that people really want these products, both in car dealerships and in the case of the HELOC product, the combination of process and rates that we deliver there, we think are best in market.
We have got the sort of step one done. Then we were focused on, okay, we got to start onboarding capital providers because we are very committed to funding these businesses in a really capital efficient manner, having primarily third-party capital funding. That was kind of the second step that we had to prove out along with credit performance in these products. Then really the last thing we worry about is unit economics, and so that is the step we are on now, and we are rapidly improving the margins of these products. We will get them to break even before the end of this year. Then once we do that, then we will turn our attention back to really scaling these products up.
Yeah. Where do you think they can go over time? Obviously, I'm sure break even's not the destination. How do you think about continued improvements over a multi-year period?
Yeah. The long-term margin, for now I'll just refrain from speculating on just because I think when you're negative 40% or something, it's just irresponsible to speculate about how positive it can be. You got to get positive first to earn that right. But I don't really theoretically think there's any reason that they can't be as good as what our core business has today. They're really very similar kind of businesses, similar dynamics, and we think the size of the technology advantage can be just as large over time. I think the actual value capture is going to happen a little more incrementally because we're not going to sort of like- Prioritize getting that all the way up at first, I think that would be over-monetizing. I think we need to capture value as a function of how much we create.
The value we create is just a function of the level of technology differentiation, which today is not as high as in core personal loans, but is going up. So I think over the course of a few years, it'll get quite a bit higher.
Yeah. All right. I wanted to maybe talk to the other driver in the back half of the year, which is continue to model improvements on the core product. I think one of the things that's harder to understand from the outside as a follower of the company is just how you actually drive more volume growth. You talked about the technology wins that have been able to offset a lot of the macro pressure so far this year. Q2 alone, I think you mentioned three new personal loan models, a lot more variables in the model. What does that process look like from start to finish? Is there an internal roadmap of known potential enhancement opportunities that require engineering staffing? Is it more of a continuous brainstorming and testing process? How do you get confidence to compound that process over a 12 - 18-month period?
Yeah. We've been making our models better for a really long time, essentially since we started the company back in 2012. We've just, every month, every quarter, worked to make the models better. For all those 14 years, we've never run out of things to do. To the question, yes, there absolutely is a whole backlog of things that we would do. Of course, what we actually work on 12 months from now will a little bit be path dependent, what works and doesn't work today. In some sense, I think it's not that different than what any kind of research roadmap would look like at any model company.
I think if you were going to go ask OpenAI what their roadmap looks like, I think there's just a lot of different ideas for making models better, and we benefit from a lot of the same fundamental research that goes into new types of neural networks or new types of chips for training. These kinds of things help us, and we look at a lot of the papers that come out on new kinds of learning algorithms, new kinds of cost functions, and we develop a lot of our own. The search space is really pretty large. It gets even larger longitudinally when you think about what the constraints to better models are, where one of those constraints is our kind of engineering input.
That's a constraint that is a factor in the function of how much model improvement output we get. There's other ones that are kind of time-bound for us, which is you need to have more training data. You can't increase the levels of complexity of your model without having sufficiently more training data. At some point, you just kind of have exhausted all of the signal you can get from the amount of training data that you've got. But the good news is the training data is increasing, and it's increasing exponentially. We've got, I don't know, 140 some million training data points now, a lot more than we had a couple of years ago. As that number goes up, it kind of just naturally unlocks more sophisticated models.
Then there's this other piece of it, which is really about compute speed and cost. This is where I say, sometimes you get these kind of outside things going on in trips or kind of usage of our algorithmic efficiency. They become pretty important. I think if you're in the world of traditional finance, this is irrelevant. Your compute costs are irrelevant. But in our business, it's much more like a model business, and compute costs can be quite substantial, and kind of become a limiting factor. If your model gets too complex relative to the speed that they can be processed on, then you've got a customer who's waiting there for three minutes while you're trying to figure out the math. That's a pretty bad user experience. You've got to have teams that are optimizing model speed.
You've got teams that are optimizing model costs. You've got teams that are doing actual sort of fundamental research on the algorithms. Then you're just over here, just waiting for more data to show up. All of these things are kind of happening in concert, and that's what makes for such sort of a powerful data flywheel.
Yeah. No, that's super helpful. Maybe we'll just talk about the broader market opportunity. I think the process is very bottoms up, like how do we get the model to respond to kind of the opportunity that's in front of us right now? How do you think about more of a top-down framework? It's kind of difficult to get market share data in this space, but I have to imagine Upstart's the largest game in town for near-prime personal loans. Just how do you frame the addressable market, where your current penetration is, and then how do you think about market share versus kind of market share creation, expanding the TAM to borrowers who simply just aren't in the borrowing activity that we see today in outstanding consumer credit?
Yeah. I think the good news in our business is that I think unlike a lot of others, it's kind of just like the market is extremely large. I think the simplest way to think about it is always just you look at the amount of credit card debt, it's like $1.2 trillion or something, and about half the consumers in market are sort of not considered super-prime or they're south of 720. It ends up being an extremely large market. Now the vast majority of these people don't even know what personal loans are. But the good news is that it's not sort of like for us, just you've got to kind of convince them on sort of subjective experiential factors that they should prefer one thing over the other.
In our case, it's just like we're just here to save people money. If because of that better ability to separate risk, if we can get people lower rates than they would get elsewhere, it's sort of an objective good. Now, that doesn't mean you automatically win the consumer. They still have to find out about you and all the things that actually have to happen to get them to become a user of our product that they're not so familiar with. But I do think the fundamentals are very good. Very large market, objective sort of money savings for people, and who doesn't want to save money, right? So I think it's a very winnable battle.
The way that we've been going after it, even if you don't believe in any kind of step function changes in brand recognition or the sort of number of customers we have relationships with, the thing that is really mechanical is just like, all we do every month is we just say, how good are our conversion rates? This should be a kind of population-adjusted conversion rate. How good is our conversion rate? That dictates how many people we can afford to reach based on various marketing channels. The most direct is you look at a direct mail kind of marginal CAC curve, and it's just like, well, we send out every piece of mail that has a positive return on the marginal send. If your conversion rates get better, you can send a little bit more.
That is a really mechanical way to just reach more and more people, and that is probably the most direct way you can think of the business as growing. It's like, we make better models, better models allow us to reach more people, and boom, now you have both more people and higher conversion.
Yeah. Let's talk about risk separation then. You talk about the inaccuracy gap, 87% of default risk still not being captured by your models, which I think is significantly better than what we see in sort of incumbent models. Hopefully, I got that definition of the 87% right. You can either read that as enormous runway, or you can read it as something about credit risk is very structural and maybe just isn't knowable at the time of underwriting. How do you get comfortable it's the former, not the latter? How do you think about when you hit the point of diminishing returns on that statistic?
Just look at what AI has done in other fields. I think at this point, it should not be controversial that with better models, you can get dramatically higher levels of intelligence. I think that's pretty self-evident in 2026. I think in some sense, that's the founding belief of the company, is that we've always believed that if you had better models, you could get dramatically better risk separation. For 14 years we've been continuously right about that the next sort of step of model improvement unlocks more accuracy in the models. That hasn't slowed down in any way, which I think is probably contrary to what almost anyone would've predicted, that maybe we would just chew off some low-hanging fruit in the first three years of the company, and then it would level off after that. But that's not how the graph looks at all.
It's just kind of almost shockingly sort of a straight line of improvements to model accuracy. If we look at our backlog, it's a long backlog. We've been doing this a long time. It keeps going this direction. I think kind of Occam's razor, the simplest thing to project out is that you think it'll keep improving.
When you think about the pace of improvement, I think Upstart's already a kind of deeply AI-enabled company. Has been for a really long time, since before ChatGPT hit the world. You talked about some customer support type use cases, but when you think about that kind of core operational, like we're going after model improvements today, what types of improvements in efficiency are you seeing in terms of the ability to iterate on the product and ship faster? Some of the different buckets that you talked about of kind of known potential ways to improve the model.
Yeah, I think we're doing pretty well in this department. I was looking at some of that ramp data about how much sort of tokens different companies use, and I think we're probably, as a technology company, something like a top 5% adopter of this technology. Not like top 1%, but definitely up there.
I think we're seeing pretty good results in terms of the amount of code that's getting written, the sort of time to close tickets, and ultimately, I think that should show up as just more revenue and more revenue growth per employee. Of course, it's always a little hard to perfectly attribute causality to these things. I don't know the precise answer to this question, but I do think if you look either bottoms up or tops down, the numbers are pretty good on what we're getting out of this, and we'll just have to keep making it better over time.
Okay, we've got a couple of minutes left. I want to jump around a little bit, but I want to talk about the bank charter. You've got conditional SEC approval to launch in early 2027. Can you talk about just the investment that was required to do that? What sort of drag was it on just focus of the organization, investments required, and then just how do you frame the biggest benefits that you expect to realize once that occurs?
Yeah, huge investment. It's definitely the largest discrete single project that the company has done and is doing in 2026. Of course, that makes it a cost center with no benefit in 2026. Hopefully, that will reverse once the bank is open and launched in early 2027. By the way, I think that level of investment is totally justified. I mean, it's a big deal to open a national bank, and I think we want to make sure to do it right and to do it in a way that's compliant and is going to take care of the public interest and safety and soundness. I think we're putting a lot of effort into this. I think we're going to do it right, and hopefully, we're going to get great benefits next year when we're live.
Just for the avoidance of doubt, the casual observer may look at what's happening in fintech more broadly and say everyone's becoming a bank. What are the specific use cases for the bank for Upstart? I think you've been pretty clear that this is not signaling a change in the balance sheet strategy or the funding strategy of the business.
Yeah, that's right. For us, there's just a bunch of operational benefits that have to do with simplifying so we can reach more borrowers more easily. Today, we have almost 100 originating partners. They all have kind of slightly different regulators, slightly different rules, slightly different disclosures, slightly different states where they can offer products and which products. All of that simplifies away. We can reach more people without a lot of that operational complexity that we deal with every single day. On the sort of funding side of it, I would say it's not like we're totally just not going to use the sort of enormous benefit that is deposit funding, but it's more like today we have about $1 billion of loans on our balance sheet. Most of that is not levered, and so it's just a very inefficient way to use equity capital.
If you tell me, "Hey, now you have a way to lever that stuff at a really low cost of funds," of course, we are going to use some of that, and that will actually be really efficient from an equity capital perspective. So maybe contrary to, I think sometimes people worry like, "Oh, well, in order to open a bank, you are going to need a whole bunch more equity capital." It is like, no, today we have $1 billion of loan assets that are mostly unlevered, just going back and levering that actually creates net new
You have the capital.
Net new equity or net new sort of cash. So it is actually quite equity efficient to do this.
Yeah, makes sense. Apologize for the CFO-type question, but you mentioned the big investment in the year. Fixed expenses this year were up around 30% year to date. Pretty meaningful step up. How do you think about a more normalized level of expense growth, particularly once you are through this initial investment on the bank side?
Yeah, we've said that Opex growth is going to normalize a lot in the back half of this year and go down to low single digit type quarter-on-quarter growth. I think that's a much more normalized level for the business. I think coming back to this AI and productivity stuff, I think we are starting to see it show up in really nice ways. I think that maybe like the bank thing, like some of our new products, there's just a whole bunch of different things hitting all at once that made it so in the first half of this year, we're really in this place where we have one really profitable contributing thing, which is core personal loans. It's paying for everything else that we're doing, all of which are unprofitable, even at a contribution level.
This is our new products in auto and HELOCs, this bank effort. Of course, some of the ways in which you ramp the AI for internal productivity spend is very, at first, you're just very experimental about it, let everybody sort of do their own thing, and then you start to worry about making it efficient later. I think we're starting to turn the corner on a lot of these things from a margin perspective, from harvesting more benefits than cost. The bank thing, of course, as we're saying early next year. I think a lot of the timelines are such that our expectation is we're just going to be able to get a lot more operating leverage out of the business compared to what we did earlier this year.
Got it. Makes sense. Well, I think that just about takes us to time, but thanks for being here. Thank you for doing this. Really appreciate the conversation.
Fantastic. Thank you.
Thanks, as well. Great.