Upstart Holdings, Inc. (UPST) Earnings Call Transcript & Summary

Great. Thank you, everybody. We're just about to get started. So welcome to the Upstart presentation at the Goldman Sachs Communacopia and Technology Conference. I have the privilege of introducing Sanjay Datta, Upstart's CFO. Prior to joining Upstart in 2016, Sanjay was the VP of Advertising Finance at Google. My name is Mike Ng. I cover Upstart and fintech here at Goldman. And we have about 35 minutes for today's presentation inclusive of Q&A. So if you have a question, please feel free to raise your hand, and we'll get a mic runner over to you. So first, Sanjay, thank you so much for coming to participate in our conference and being with us here today.

Yes, thanks for having me.

So maybe to kick things off with a higher-level strategic question, Upstart has made a tremendous amount of progress in unsecured personal loans. It's expanding into auto with your in-dealership offerings. Could you just talk about your 5-year view? What does success at Upstart look like? What are some of the key milestones and initiatives that you're watching and pursuing as you kind of undergo continued growth at Upstart?

Sure. 5- to 10-year as well. Let's see. Well, I would say the [ pure ] fundamental vision of AI and lending is unchanged in the sense that I think that all flavors of credit, all segments of credit will benefit from more predictive risk models. And we believe that in a 10-year time frame, almost all lenders will be using some versions of these technologies in all segments of credit. I think the technology thesis is pretty incontrovertible, and the economic value you create by getting better risk assessment in credit is pretty massive. So it's pretty core to our belief that that's a change that will happen over a 10-year time frame. In terms of hurdles, well, the technology case is sound, we believe. I think the economic model of better risk within lending is also compelling. I think to get there, there's going to be a lot of regulatory conversations, obviously. I think just the general sort of discussion around AI and machine learning and how it gets applied to sort of more and more use cases in the world is a hot topic right now. I think it's particularly acute in an area like lending where there's a lot of regulation. I think there's some kind of adoption curve of this technology on the funding side of the ecosystem. So banks and other lending institutions, other types of institutional investors have to get comfortable with the performance of that technology. And I think there's some, around that, some resiliency in supply chain of capital that supports all of this. Obviously, we've seen it in the past a couple of quarters. The supply chains that are currently underpinning this current model are not as resilient as we would have wanted them to be. So I think those are some things that come to mind, but they're a little bit -- I don't want to call them on the margin. I think the core things around the technology and the economic model to us have been proven already. So it's just a question of some of those more larger conversations around how regulatory would think about this stuff.

Great. That's a great overview. And as you mentioned, artificial intelligence and AI has never been more top of mind than it is now. And much of the Upstart competitive advantage has been related to the AI-based algorithm that the company has built and trained data on. Could you talk a little bit about Upstart's underwriting model, how it may compare to utilizing FICO and other more traditional underwriting models? What ultimately makes Upstart's platform unique and more difficult to replicate?

Sure. Yes, AI is definitely bandied about in the press a lot of these days and in the investment circles, of course. The core concept in credit is pretty simple. The technology is used to predict risk in credit up until very recently, and the core of it has been this idea of a credit score. I guess like as an input, they're not particularly sophisticated compared to what we have in other sectors of the world, like consumer Internet, in that they are tending to be linear models. They have a lot of restrictive assumptions that make it such that their ability to predict outcomes are less sophisticated. And of course, if you look at the outputs, if you try to correlate credit outcomes to FICO scores, you won't find a very strong correlation. And so we had the simple idea of taking a lot of the technology that was being adapted for initially other uses. I mean you think about things like email spam filters and search engines and consumer recommendations. Those are all using nonlinear predictive models that relax the assumptions of interdependence between input variables that have been enabled by compute power. That's all machine learning really is. A linear regression is a machine model. The machine model is just taking amounts of data and training a model to predict outputs. But they have a lot of constraining assumptions. And when you relax those, you need a lot of compute power. We now have the compute power to be able to explore nonlinear surfaces. And so it seemed obvious to us, and many of us at the company come from a technology background, it seemed obvious that they would be a better prediction engine than a, essentially, a linear regression. What makes it unique and difficult to replicate, well, the unique part is very simply, I think for a long time, we were the only ones applying this to credit. Even today, I think there is very few, if any, parties out there that are applying modern machine learning at scale to the problem of credit prediction. Why is that? I mean some of that is regulatory. I mean, historically, there is a lot of regulatory risk in lending. And so if you want to apply new models to lending, you have to invest a significant amount of resource in engagement with the regulatory bodies. And some of it is just, frankly, because of what I said. This stuff is pretty difficult, and it's not native to financial services. I mean most of the stuff was developed in the consumer and digital Internet sphere, and there needs to be a cross-pollination of technology over time. A lot of the people who traditionally build these types of models, and it's not a massive labor pool, are coming from the big tech companies. Now why is it difficult to replicate? It's not impossible to replicate, of course. But credit has this unique thing, which is that in order to build these models, machine models require what's called training data, which in the case of credit is just historical examples of whether you got paid back or not. And in order to accumulate that training data, you need to lend a lot of money to a lot of people and then sit around and wait to see if they pay you back or not. And that happens month-over-month, quarter-over-quarter. There's no real great way to accelerate that process. So it's not like other applications where you can just ingest a bunch of historical training data. It's not that people haven't been lending for the last 10 years, as we have. But no one has collected against those repayments where the applicants went to school, what they studied, where they work. So ultimately, you want to price new and alternative variables that are not considered by credit score. And so we now have 10 years of history and of data around how to think about how to price where someone went to school or where they're working. What does it mean to work at Goldman Sachs versus one of your competitors or versus a different industry? What does it mean to be a research analyst versus a banker versus a wealth manager? All of those roles and functions and areas of study have different implications on risk, and we've learned a lot about them. So for someone to try to replicate that and, in particular, to do it while now competing against us who's developed, I would say, a lot of insight about that stuff can be very expensive and unpalatable. And it just takes time. There's no shortcutting at it. So that's, I think, the difficulty of others that will come down the path and what they will have to replicate over time.

Great. That makes a lot of sense. You talked a little bit at the beginning of our discussion around the resiliency of capital. Over the course of this year, Upstart announced several new long-term funding partnerships to build more resiliency into that funding model and ensure that the platform can continue to operate well even in more challenging cycles. First, could you just talk a little bit about how Upstart's funding model has evolved over time? And what does the optimal funding channel mix look like for Upstart, if there is one?

Let's see. Well, historically, we've had 2 notional channels for funding. One has involved, essentially, getting our technology to banks and credit unions. They would use that to originate for their own balance sheet. As the originating bank, they tend to predominantly be interested in primer borrowers or lower-risk borrowers, just by the regulatory construct of what a bank is. And then for, what I would call, the torso of the borrowing base, we've historically had the origination done on our behalf with a partner and then sold that asset to the institutional world. You could think of like private credit funds [ canonically ]. The third sort of funding construct that you're describing, which are more of these long-term committed structures, I would say, are a subvariant of the second category. They're sort of a type of an institutional relationship but with a capital partner that has the ability to think over a more fulsome investment cycle. They're not as reliant on leverage and trading as some of our existing partners. And they are structures in which we will contribute some modest risk capital. So those are, I would say, how they are new and a bit different than some of our existing relationships. Ideal mix, I think that over time, as we rescale the business, I think these more permanent, durable partnerships could form something like half of our funding base. I think that would be probably a nice base of durable capital that will create a bit more -- buffer the volatility of the external markets a little bit more. And of the remaining, I think there's always -- banks will always be the dominant competitors for very prime borrowers just because the cost of their capital is quite inexpensive, although it's changing. And I think there will always be a remaining maybe, call it, 30% banks. And there'll also be a place for what I would think of as the spot market. So funds coming in and out and trading at will. They create price discovery. They create liquidity. So that's maybe, I would say, in unsecured lending an ideal mix. As we expand to more and more secured products, I do think that banks will remain the predominant funding sources for those products, things like auto lending and residential lending. I mean banks have -- those are lower-risk products for which banks have very efficient funding chain.

Great. I'd love to explore the new committed capital partnerships a little bit more in detail. You mentioned a little bit about the coinvest structure. So I was wondering if you could just talk a little bit about the economic impacts on Upstart's business model from some of these new funding structures. What trade-offs are you making, if any, to bring more stable funding into the mix? And what does a typical like coinvest structure look like?

Sure. Well, a typical structure is pretty straightforward. You could think of it as a partnership or a JV where we have a counterparty that is the majority partner. We will bring some very modest, single-digit percentage of the overall partnership as our own equity. But it's important because it demonstrates alignment of interest with someone who's committing forward over a term of 1, 2, 3 years. And so they want to know that our motivations a year from now will be aligned with theirs. So that's important. It also demonstrates us standing behind or eating our cooking, I guess, you might say. And that's an important aspect of what essentially is a very deep strategic partnership. It's not a transactional relationship as we've been maybe more used to in the past. That structure, of course, may include some financing as well. That structure will be a sort of a loan buyer on our platform. The economic impact is interesting. It sort of depends on a couple of things. One is the extent to which this structure is targeting a return that may be higher than what you might think of as being delivered on the spot market. If there is a premium being delivered to that structure, it needs to come from 1 of 2 places. It may be passed on to the borrower and/or it may be absorbed by our take rates. And the extent to which it's one or the other will depend on the elasticity of the borrower demand. So for example, in an environment like today where demand is very inelastic, there's not a lot of alternatives out there. There's a high demand -- a high fundamental demand from borrowers for credit right now. You might believe that a lot of that premium, to the extent it exists, is getting absorbed by borrowers. You might imagine a world where -- that goes back to the very competitive world from a lender standpoint, and you might imagine that some of that will be absorbed into our take rate. So that would show up as a contraction of contribution margin. But it's probably worth saying that what gave us the insight and the inspiration to do this to begin with is that if you've been following our financials through what's been a very tricky, I would say, difficult period for a lending platform, we've flexed our contribution margins and our take rates up quite a bit in ways that are, I think, a bit nonstandard in the lending world. And the reason is because in difficult periods like now, a lot of the platforms and the issuers tend to hyper-compete over very prime borrowers. And if you have the type of risk models where you can still decision-risk in maybe the torso of the borrower base, there's still a lot of margin to play with. And the fact that we've been able to flex our margins and our take rates up a lot has made us ask ourselves the question, would we have traded some of that in exchange for more durable, more resilient funding base? And the answer is, of course, yes. And so that's what led us to sort of put some of our economics, I would say, at stake in a relationship where the underlying funding -- the trade-off is a much more resilient supply chain of capital.

Great. And I appreciate the comments that you made around how the target yield for some of these committed capital partners can be funded either from the borrowers or yourselves. I was wondering if you could talk about how the funding mix may be different in different economic environments as well. So for instance, if the ABS market or demand for whole loans pick up a lot, would you see like that part of the funding mix grow dramatically? Said differently, are some of these more durable capital solutions more of an interim solution? It doesn't sound like it is, but would just love to hear you talk about funding and...

Yes. This is meant to be a long-term strategy that will set us up for the next macro shock, whatever it is. And the current macro shock was some combination of the lockdown and the stimulus from the perspective of a lending business. The next one, god knows what it will be. When we're back at scale, like I said, I mean, I think 50-plus percent of our funding base at scale we would like to be in this sort of committed form. And then the banks, call it, 30; and maybe the spot market, 20. And the next time a macro shock hits, the spot market will contract, as we've seen in this sort of most recent version. The bank money is more resilient, but they're going through their own industry-specific things. And what is at scale, 50% of our funding base, should remain, in absolute dollar terms, the same. And that's kind of the point. There's other funding channels that have other interesting characteristics but are not durable, and they will flex with the economy but the committed funding base should not. And that's why it needs to be a long-term strategy. It wouldn't make sense to have a long-term committed sort of vehicle that is not a long-term strategy, obviously. It doesn't have the kind of velocity. If we were just relying on that to sort of recover from the current lending environment, it wouldn't be the ideal way to do that, in my opinion.

Okay. That's very clear. So can you talk about whether or not funding is still a constraint on origination growth? It was at one point. And obviously, you guys have done a lot in improving your funding position. Are you actively seeking more committed capital partners? Or is the funding and the demand from the borrower side more aligned right now and you don't need that? Could you just mark-to-market on that?

Yes. I think if you were to talk to the team like today, what they would say is the funding we've put in place is not the constraint on our business right now. The constraint is that it's very, very hard to approve borrowers right now for a bunch of macro reasons we can get into. And so finding the next borrower and being able to approve them is the constraint on the business. That said, that can turn very quickly. There's a bunch of macro behaviors happening right now on the consumer side that can change quickly. And mathematically, they're almost going to have to change. And when that happens, we're going to want to have the funding available. So there's still teams going out and continuing to pursue discussions with the capital sources. We probably can't put their dollars to work tomorrow. But a month or a quarter from now, we probably will be able to, and that's why we need to have that sort of ready.

Great. That's really helpful.

So following on from your last point on talking about constraints, have you thought about linking up with Pagaya Technologies in order to increase loan origination? Because when I analyze sort of your volume growth, I think it would be a good tie-up in the sense that you're not competing because they're not B2C. And it feels like they could -- you could increase your volumes by, say, probably like 20% with an attractive FICO score and attractive returns. Is that something you've thought about?

Yes. So we know the Pagaya guys well. They are another example of a company that does a lot of the same things we do. As you said, they are in a bit of a different business model than ours. I'll say this. We talk to them regularly. I think that the uplift they're able to find with other platforms in their kinds of programs is largely because their models are very differentiated. And many of the issuers have very sort of credit score-centric models so they can create huge uplift. With us -- the type of model they pursue is that like if the platform doesn't want the loan, they'll send it to Pagaya and Pagaya can price it. I don't think they can create a 20% uplift on our approval on our models. We do experiments with them, and we sort of are in a good-natured partnership with them. And we try to learn how our models are different. And there may or may not be a point in the future in which we sort of find a way to work together. But it's certainly, I don't think, for us or for them, would create the kind of economic benefit that they have when they work with someone who has a very traditional model, which is sort of [ they're deploying ].

Great. You mentioned some of the impacts from the macro. Maybe you can just talk about how you see the current macro environment impacting your business, what your outlook for the third quarter and the rest of the year assumes. And I think it would be helpful if you could touch on what you're seeing from the Upstart Macro Index, UMI, and how the company uses that to guide credit decisions to?

Sure. Yes. So there's a bunch of topics in there. The macro is a very strange thing right now. As you know, there's a lot of -- there's a very wide distribution of opinions on it even more than usual. There's anxiety, at least recently, around inflation and rates. It seems to be subsiding a little bit. On the other hand, the labor market is as strong as it's ever been. Consumption is as strong as it's ever been in real terms. And so it's a bit hard to know what to make of it. From the perspective of how it's impacted lending, our view is very clear. So it's been a turbulent time, and almost all of it is attributed to the stimulus. And in particular, the stimulus did 2 things. I mean it did a lot of things. But from the perspective of someone who wants their loan paid back, it did 2 very particular things. One, I think it was a very significant, if not the most significant, factor in the inflation we have. There's the supply side stuff as well. But like if you pump $5 trillion into the economy, unsurprisingly, it's going to contribute to price inflation. And the other thing it did was it caused a relatively sizable -- after the volatility of the lockdown, it caused a very sizable increase in real consumption relative to income. And of course, that was [indiscernible] by the stimulus. And when the stimulus ended, the consumption didn't go back down. Instead, what happened was consumers have run their personal balance sheets down to a level that's razor-thin. And I don't know why they did that. I have theories. But those are the facts. If you look at the savings rates in the economy, if you look at the deposit base and how it's contracted, and that's really at the root of a lot of the problems we see in the banking sector, it's created this different sort of pattern and behavior in how consumption and income play together such that everyone's balance sheets, and particularly those who are less affluent, are much thinner than they were pre-COVID. And the combination of the inflation and the fact that even in real terms, people are, essentially, living on a relative basis beyond their means compared to what they were pre-COVID, has meant that [ defaultiness ] in the credit world has gone up. It's gone up a lot. You mentioned what we call the UMI, the Upstart Macro Index, is essentially our best expression of that. It's nothing more than -- it's an index which tracks losses in our loan book after controlling for all changes in borrower characteristics over time and all changes in our underwriting models over time. So it's a mix-adjusted, apples-to-apples view of how the same borrower is defaulting, at least in our book but we have a very broad book, over time due to the economy or, I guess, well, due to things that are extraneous to us. And it would show you that if that index read as a 1.0 pre-COVID, with all the stimulus, it went down to 0.5 or 0.6, which meant losses are half or 60% of what they were pre-COVID. So they were very good. And that index is now sitting at 1.7 almost. So it meant that after the stimulus ended, savings rates plunged to the lowest level since the World War. If you look at the history of the personal savings rate, which is printed by the Fed, it's somewhere between 3% and 15% since they started printing the number, and it's currently at 3%. So that coincided with this index now sitting at a level which is 70% higher than it was pre-COVID for these borrowers in this particular product. So that's, I think, how we would best express the economic lens from the perspective of a lender certainly in the unsecured world and a little bit about what we think is at the root of it.

Great. Why don't I sneak one more question in before I see if there are any other questions from the audience. But I do want to ask about your outlook on U.S. unsecured personal loans, obviously, your most established loan product. The market was about $170 billion in 2022. What's your view on the potential growth in that unsecured personal lending market? What's driving the demand? Is Upstart helping to grow the market by helping to lend to these underserved borrowers? Or are you taking share from established players or both?

Well, neither in 2022. To the extent the pie grew, we were not at the root of that. We've reacted to this severe increase in default rates by reducing approval rates and contracting. But if you want to talk about it in more of a secular level, yes, unsecured lending is a very young product. It was sort of -- it was birthed by the Internet. It was very hard to do an unsecured loan for $10,000 if you -- in a bank branch. So as a result -- banks had the product, but they didn't push it. It wasn't economical for them. And you sort of had to like know about it and go and ask the manager. So there was no real unsecured term loan category before the digital players started making it economically feasible, and it's obviously been on quite a growth clip. Pre-COVID, it was the only flavor of credit that, I think, had significant growth. Now the numbers have all changed a little bit, but it was growing because it's a useful product. I mean I think it's easy to obtain. It's digital-native. And in many cases, the pricing was getting good enough such that the benefits you were to get on pricing from getting a secured loan, like a HELOC or even an auto loan in some cases, weren't worth the extra sort of trouble you had to go through in the liens and notarizations and all this stuff. And so it started, I think, on the one end, eating into the credit card market; and on the other end, eating into the secured market like places like HELOCs and such. And yes, I think a big part of how we participated in that was the growth of the market. The way we think about market share, by the way, wasn't as us taking from a fixed pie. If you think about us at our peak volume when credit was really good post-stimulus, our conversion rates, by that, I mean, of all the people who applied for a loan, how many got one, was about 20%. Right now, it's about 10%. So our approval rates have pretty much halved. But of the 80% that were looking for a loan and didn't get one from us, you can see in their credit reports, they didn't typically get it somewhere else. If we couldn't approve them, they probably weren't getting the loan, or if they didn't like our offer, they weren't getting a better one elsewhere, by and large. So it meant that 80% of the market was unconverted even at the peak of our volume. And our growth was really a story about improving that conversion rate. When we were improving our conversion rate, we're essentially pulling people into the market. So I think you had the $170 billion...

Yes, last year, in 2022.

Okay. That's the 20% of the market that's converting. There is a group of people 4x that size that are looking for the product, have applied for it and either not getting approved or the rate was just too high and they're like, let's not -- I'm not -- it's not good enough yet. And the risk models getting better are what's going to improve both the approval rates and the acceptance rates because it's going to bring pricing down [indiscernible]. So I view it as a giant unconverted market that we will convert through better risk pricing.

Great. Any questions from the audience? We got one up here, please. Just the mic is just going to go...

Sure. What's the progress on auto loans and also on getting more dealerships there on board with your software?

Great question. So yes, auto loans was, I think, the next big thrust for us when, I think, the market became a stiff headwind. And I would say that as a result, the business itself and the scaling of the lending in auto is a little bit subject to the recovery or to the moderation of those headwinds. I think under the hood, our strategy involves a couple of things. One is getting more and more dealers to use our software to sell the car, not to finance it. [ So it's not new ] with lending, but we want more and more dealers to use our software to sell the car. So we've got some software [ packages ] trying to modernize the car buying experience for the user and the dealer at the point of sale. For anyone who's bought a car in a dealership recently, it's a pretty archaic process from a systems perspective. And so there's a couple of companies that, including us, that are trying to modernize that. And I think we're probably one of the companies that's in the lead in terms of getting OEMs and dealership groups to modernize their stack. And then, of course, within that, we introduced our lending product. So it's a bit of a Trojan horse strategy. I think the number of dealers using our software now is somewhere north of 800. It continues to grow nicely. The penetration of the lending product into that 800 sits at about, I don't know, Jason, I want to say like 60? [indiscernible] public numbers.

I think 40 is the last we [indiscernible].

40. So it's a very small penetration of the overall base. What's gating that is the availability of the capital. So auto lending segment right now is an area where a lot of the funding sources are very conservative. A lot's happening with used car prices. A lot's happening with subprime auto default rates. And so I think we have a lot of interested funding parties, but they're all waiting to become comfortable with the macro. When they do and they're ready to start lending at volume, we'll start rolling out the lending products to more and more of the 800 installed base, and we'll continue to try to grow the 800. So all that to say from the perspective of our financials, not much looks different than last quarter. But I think under the hood, there's more and more spring-loaded potential energy waiting to be unleashed once there's sort of a cooperative funding market.

Any other questions from the audience? Sanjay, maybe in the last minute that we have, I was wondering if you could talk a little bit about what's most top of mind for you as we kind of navigate through the current climate, right? We've talked a lot about funding capacity for auto, funding for personal loans, improving the top of funnel, R&D. There's a lot of things that are certainly top of mind for investors. So what are you most focused on?

Yes. At a notional level, it's like -- I think there's a lot of amazing progress in our business that is not seeing the light of day because of the macro environment but will. And so the notional philosophy right now is just to continue making the business as good as possible, knowing that at some point, it will manifest itself. In particular, in the spirit of never wasting a good crisis, I think, we have and will continue to get much better at understanding and reading and reacting to macro shocks. Historically, that was not our focus as a company. We used to try and focus on borrower-level evaluation and relative ranking of risk across a borrower pool. We will never be in the business of predicting the next macro shock. But I think the ability to react to it more quickly than anyone and more precisely than anyone, I think, is within our capabilities, creating a much more resilient supply chain of money. So that the next time some macro shock happens, and it will, we will have a more resilient supply chain. I think that's an area we made progress and continue to. And then the other area, frankly, which has always been a huge opportunity for us as a machine learning company that we've never put much focus on is in applying those predictive models to servicing and collections. And because that's become a very important topic right now, we are putting -- to the extent we're doing any net hiring and resource application, it's in that area. You can imagine how predictive models might be very useful in that sort of activity. So those are areas that I would say are areas that have not been traditional focuses of the business that we're taking this opportunity to really shore up so that when we come out of this, we're going to be a much stronger company.

Great. That's really exciting to hear. That's a good way to cap off the session. So Sanjay, thank you so much. It's been such a privilege to be able to share the stage with you.

My pleasure. Thanks for having me.

Thanks, Sanjay.