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

Welcome back, everybody. We are very pleased today to welcome Sanjay Datta, CFO of Upstart, to our conference. Sanjay, thank you so much for being here today. Really appreciate it.

Thanks for having me, Ramsey.

Why don't we start with just a broader question about Upstart? For those in the audience who are not familiar, what do you bring to the table in terms of the value proposition? What was the legacy system of underwriting sort of not capturing that you guys are able to sort of address and capture?

Sure. I guess I mean -- I think of ourselves as sort of principally doing 2 things. The first one is -- I mean there's a lot of people who would repay loans who don't really have access to bank credit. And almost everyone who does have access to bank credit is subsidizing losses for others who are given loans that shouldn't have been given loans. And so our observation is if you have a more accurate risk model, if you can better pinpoint propensity to pay back, A, you'll be able to approve a lot more of the population sort of into the lendable set, if you will. And then you theoretically would be able to keep more of the losses out of the system, which means you can lower prices for everyone. So I mean the sort of the central tenet of our company is smarter risk models equals higher approvals and better offers. And I guess the second thing that we think a lot about is a lot of the traditional loan processes require a lot of manual review and documentation. And it's really because of the risk of fraud, right? So you're giving them information, they're underwriting on the base of it, and they have to make sure you're not lying. If you can be really good at predicting fraud, then you can automate a lot of the verification process and you can make the access of credit much more effortless. And you can selectively apply the friction of manual documentation where it's most needed. So those are the sort of the 2 pillars, more approvals/better offers and more sort of frictionless access.

And I want to talk about the broader, sort of, market opportunity for your products, maybe even outside of the core personal loan business that you all started with. Let's talk about your move into auto, sort of how would you characterize in terms of how far down the path are you with auto? Give us an update on how that rollout is going.

Sure, yes. I mean I guess there's sort of really 2 approach vectors we're taking as a company into auto lending. The first one is auto refinance. So for people who have auto loans existing, our observation again is that there's a significant fraction of them that we would consider them as [ price ]. And so there's an opportunity to refinance. I would say that, that's a product that's very analogous to our personal loan product in that existing people who have credit in the personal loan instance, that's tends to be credit card debt, in this case, it will be auto loans. It's us reaching out through our marketing programs directly or through agent partners and essentially giving a refi offer. So that one, because it's, I would say, somewhat adjacent to what we do already, it's the one that's furthest in development. We've been, I would say, incubating that product for the better part of the year, tuning our models, getting the loss curves right and seasoned. And we're now at the phase where we're beginning to scale that and deliver it to the third-party funders, whether they be banks or credit unions or capital market institutions, so that they can earn the yield and scale the business. So that's, I would say, in the early phase of scale-up. And then the second approach is really about being at the point of sale in the dealership. And our approach there is to, first, provide the retail e-commerce software to dealers. So we acquired a company called Prodigy last year. And they essentially are working on modernizing the checkout experience for buying a car. And then into that platform, we are slowly but surely inserting our loan offers. So the underlying retail software platform is now, I think -- or has been adopted by somewhere north of 500 dealers and growing quite rapidly. And we're at the very, very early stages, maybe 2 or 3 dealers, where we've got the loan product embedded in the checkout flow. And I would say we're at the stage where we're, sort of, looking at the early data. We -- I think between now and the end of the year, would roll that out to a larger percentage of our overall footprint. And I think by the end of this year, it will start to take on meaningful scale.

And so when you look a few years out, I mean -- and I know this is a crystal ball question, but what's the split in terms of refi versus purchase on the auto side? Do you think refi will continue to be sort of the bigger opportunity or the opposite? Or maybe it's evenly split?

It's a good question. I would say it's sort of an open internal debate right now. My own personal view is that I think the bigger opportunity is in retail simply because, obviously, most of the transaction volume or most of the current volume in lending for autos happens there. That's the point of sale. It's really only if you don't get the pricing correct at the point of sale that there's an opportunity for refi. So I think to some extent, it depends on our success in penetrating the dealership base with our retail software because our retail software allows us to ask questions we care about and, therefore, price the loan accurately. And so I think if we're very successful in retail, it will probably imply that the refi opportunity is less. But the harder it is to penetrate that base, then the more opportunity I think we'll have in refi because I think we will view there to be more mispricing opportunity that we can go after. So I think we'll be -- I think what we're seeing early days is that we're getting a lot of momentum in the dealership, and I'm hopeful that we'll see a sort of a bigger business and a bigger impact at the point of sale.

That's interesting. It's kind of an interplay between those 2 and depends...

In theory, if we did every -- if we believe in our models, and we did every one at the point of sale, there wouldn't really be a need for a refi product we're [ launching ] ourselves.

Yes. And you recently signed partnerships with VW and Subaru, I think. How should we think about that? How do those work? And how do those partnerships work? And should we expect more of those down the road? Or any color there?

So those are -- those OEM-style deals tend to be sort of retail transformation-style deals. So ultimately, at the end of the day, you're selling to the dealership, to what we call the rooftop. But at the OEM level, they strike deals, which sort of take the form of endorsement, corporate-level endorsement, some customization, which will make the platform sort of more interesting and more relevant to the individual dealerships. And then in some cases, some financial incentives or some programs to try and push adoption. So it's more of a corporate-level program that will grease the wheels, if you will. We have an entire team out there trying to cut OEM -- OEM-level deals, and we would hope to have more to announce after it.

Okay. I want to get into more new products in just a second. But first, I wanted to touch on the macro environment. On your earnings call, I think you guys called out some intensification of some of the macro headwinds over the last handful of months. I guess, how should we think more broadly about the impacts of a more challenging macro backdrop on the business? And sort of what surprised you most in terms of how the macro changed between the prior quarter and the most recent quarter?

Sure. Yes, yes. Macro is a hot topic these days. I think the general dynamic is one where I kind of view ourselves, in some sense, to be a marketplace for bringing together borrowers and funding sources. I think at its simplest level, macro strife tends to be accompanied by either some combination of rising consumer defaults, rising interest rates and higher risk premiums. And any of those 3 things will drive up costs to borrowers, and that will have the impact of reducing our conversion funnel, because higher prices will tend to result in lower demand. So that's the general mechanic. With respect to the current macro, I mean, if you even just rewind 90 days ago, which doesn't seem that long ago, I mean, like the war hadn't started yet. I think there's still some debate as to whether we would actually see something like that. The treasury hasn't moved. The sort of the swap curves, I mean, the 2 year, we were still at maybe half the rate it is today. There was talk of inflation, but it wasn't yet clear that the Fed was going to move aggressively as it now has signaled it will. So I think very quickly, you've seen on the rate side, I would say, a pretty steep increase in risk premium is the sort of the -- the war broke out and the Fed started making a much more aggressive messaging around inflation. And the other aspect of the macro that has sort of started to become very clear in February is that -- in retrospect, the defaults, the consumer defaults in the economy, which to us had been unnaturally low for about 18 months due to all the stimulus, I think, pretty quickly unwound. And that sort of played out between November and February. But I think by February, it sort of became clearer as the dust was settling. So that to us was a pretty quick unwinding of stimulus and increasing consumer defaults and a pretty quick increase in risk premium and interest rates. And both of those things combined to create a bit of a constrained market on our platform.

And going forward here in terms of guidance, what do you have sort of baked in? Are you presuming some further deterioration in guidance? Or are you more -- taking a more conservative approach? Or how should we think about that?

I think we're taking a somewhat conservative approach with the caveat that there's a wide variance of scenarios right now. I mean things can get really unravel further. Tightening in inflation could push us into a recession. It's hard to imagine now that with that kind of a scenario may involve unemployment, which has a direct impact on losses as well. On the other hand, you could do a quick unwinding of some of this stuff and then return to confidence. So it's a bit hard to, sort of, thread the needle in a world where macro, I think, is this volatile. And admittedly, we are a macro-sensitive business in the short term, right? Like I think we've always said, we're a technology provider to a lending business, which is very cyclical. So I think we're being sort of conservative to mean zero, but I think there's wide error bars around that.

Yes. I don't envy you having to think through this right now because you just don't know inflation could continue or could not. And there's so many variables. The labor market is tight and that continues. So I don't envy that. So that's a tough job. And Sanjay, in your most recent quarter, you carried more loans on the balance sheet than you have historically. Can you talk a little bit about the drivers of the increase and some of the dynamics that play there?

Yes, absolutely. Let's see. I guess I would say that the majority of our time, our platform is borrower constrained, and we're working hard to sort of find more borrowers at good unit economics, and that's sort of the growth model. There are these periods of time that are related to what we talked about, these sort of macro shocks, where the funding changes rapidly and it becomes a constraint on the platform. And in those periods of time, we essentially have a decision where we can either choose to sort of step in with our balance sheet and support the volume until the new market clearing price sort of is down and we get back in equilibrium or we can just not originate a bunch of loans. And neither is ideal. In Q1, we did take the decision to step in with our balance sheet and support volume when funding became volatile. And it was in a relatively modest amount. We did sort of $4.5 billion in originations in Q1. And I think this sort of amounted to maybe $150 million. But I think, look, in retrospect, we were a little bit caught off guard with the visceral reaction by the market to this. And I think in retrospect and certainly in the future, I think the decision is just going to be to accept volume volatility and not weight into the platform with our balance sheet as a stabilizing mechanism. And so I don't know. I guess, if anything, maybe that's the lesson learned, if you will. We're sort of ramping in real time to a pretty fast-moving situation. I think that we would probably react differently had we known all the questions it would raise about the underlying business model, which to our view, has not changed at all.

Okay. And then as sort of a follow-on question, this is really interesting and good to know. But the sort of follow-up question is just in terms of having the funding capacity to fund growth, I think one view is that there's a price for everything and you can find the funding you need when you need it, depending on what the clearing price is. But how should we think more longer term? And you sort of kind of alluded to this and almost answered the question already, but just how do we think longer term about your -- Upstart's ability to make sure that those loans have -- that there's demand on the institutional side for -- or on the originator side to put the loans in the market?

Yes. I think there's things we want to do to create more resiliency and to create a faster market-growing mechanism, if you will. And in particular, the funding through the banks and the credit unions that we have was very resilient. Hasn't really changed much. It's really -- there's a spectrum of resiliency. You sort of have hold-to-maturity buyers that are a bit more resilient than the folks who are really dependent on the securities markets that tend to be quite volatile. And so it's just having more resiliency of capital. And then again, when price or risk premiums go up from those funding sources, the market should react immediately. And it doesn't today, it's a lot of manual sort of effort for us to sort of understand how our people are reacting to the changing macro conditions. So I would say there's a bunch of things we want to do to create more resiliency and faster price discovery. But at the end of the day, I do think it's important to be transparent about the fact that we are a marketplace. And I do think that in the broader sense, if risk premiums and interest rates and consumer defaults go up sharply, we're going to be exposed on a volume dimension to that. And those shocks tend to be sort of short and abrupt, and they tend to normalize somewhat quickly. But they are things that we are exposed to as a business model, and we prepare for in terms of how we will improve our P&L and our balance sheet.

Got it. Back to the products. I wanted to ask about mortgage, which is something that you guys had alluded to that you were going into. Talk about that product. What do you bring to the table there? What's the value proposition look like? And how do you -- what problem do you solve in that product?

Yes. I would say, look, most of the credit segments or products have this sort of same characteristic where there tends to be a segment of the population that's super well served and highly competed for. They tend to be sort of, from a traditional lens, super prime. And then the -- your sort of your options, if you don't qualify, rapidly fall off a cliff and sort of rates get really high and spreads get really high. And that's -- within that construct, that's the segment where we would, sort of, believe there is opportunity to come to the table with more precise underwriting and better offers and higher approvals. It's really with the well served that we -- our main value proposition is eliminating friction. And so if you think about the mortgage market, it's no different. Interestingly, if you rewind way back, like maybe 2001, you had a very different market than today, and there was what we call qualifying loans, which tends to be people who qualify for a certain backing by government institutions or government-backed institutions, I guess. And then outside of that, you've got what's called non-qualifying loans. And non-qualifying could mean you're too big, your loan is too big or you just don't have the credit metrics to qualify in. For those who didn't have the credit metrics to qualify, there's like 1 million loans a year getting done in 2001. And then, of course, that probably got too euphoric by 2008 and the whole thing caused a bit of a debacle obviously. And so sub-prime mortgages became almost a dirty word, like I almost feel dirty saying it. And now there's like 0, there's like next to no -- if you don't qualify for the qualifying stuff, you're almost out of luck, you can't get a house. And yes, we know there's a lot of good credit there. And so I think there's an area of underserved borrowers where we can bring better models to bear and do the same thing we've done in personal lending and what we hope to do in auto. And then, of course, look, I don't think anyone would disagree with the fact that the process to get a mortgage is like going to the dentist. It's not fun. And there's a lot that can be automated and improved there. And some of that manual friction is there for a reason. It's because they have to verify things. So again, it's not just automating processes, it's getting better at modeling or automating fraud and fraud detection and income lying, verifying all the other things that we care about in an automated way so that we can make the process more frictionless for a borrower. So it's the same playbook. It's just obviously a much bigger scale and different structure and different operations required. But at the end of the day, it's the same value prop.

And how do you guys internally build those models? Or what goes into basically rolling out something like this? I mean, generally speaking. And I guess, embedded in that question is, how much of what you have today in terms of models and technology can just be sort of ported over versus that kind of a fresh build?

Yes. There's sort of 3 dimensions to predictability. There's the model forms, there is the data you feed it, and then there's what we call the training history that it uses to calibrate itself, which is essentially repayment history. So the model forms are very extensible to -- across all credit types. And the way I would describe it is we have taken off-the-shelf machine learning sort of algorithms, and these are sort of the technologies that have been developed in tech companies for very different applications. And we've adapted them to the lending problem, which there's a bunch of unique characteristics about lending, it's a time series problem, cash flow matters. Your training data at any given time is incomplete, right? It's not just if the loan was defaulted or repaid, there's a bunch of loans in repayment that may default in the future. So there's a bunch of unique characteristics, and we've done a lot of work to adapt machine learning, open source technologies to the lending problem. And so those sort of framework technologies, we can apply to auto and to mortgage and to any other kind of time series credit problem. And so those are a great advantage. The second side is the data. And we think of the data as like in any given sort of loan product, you're typically worrying about the borrower and some version of a security, right? So we've done the hard work of starting with unsecured lending, which basically means all you really are doing is trying to figure out if the borrower has a propensity to repay or not based on the characteristics you know about them. And I would say we've developed a very deep understanding of the borrower. So you think about auto lending, for example. And it's really -- you care about the borrower's propensity to repay and the car, which is essentially your security in case he defaults. So like I think our data scientist would say they know 80% of what they need to know. There's a slightly different place on the waterfall that an auto loan would fall versus an unsecured loan because you want to keep your car if you have to choose. But that's why we chose auto as the next segment because it's very adjacent. It's basically a borrower and then you have to worry about a car. Mortgage is further downfield, right? It's a borrower, but now the asset is very big and a little more complex and it's more meaningful to the economics. And then when you start thinking about business lending, well, now you're -- there's a business owner, and there's an asset which is the cash flows of a business. And the extent to which one matters versus the other really depends on how big the business is and how big the loan is. So that's why we're starting at the very small end of small business lending. But I guess, this is a long way of answering your question, which is I think our models are very extensible. And there's new work to be done with data. But in any case, if you're giving credit to someone, the propensity of the borrower to repay is generally important in addition to the other things you care about. And so we have that, I think, as a strong head start. So I guess that's how someone [ looking from outside ] goes along with it.

No, no. It was really interesting and helpful. What about the international opportunity? When I first sort of learned about Upstart, I thought, wow, there's a lot of markets out there where there's very poor underwriting environments and no credit, for example. How do you guys see the global opportunity?

So it's a great -- and I think it is a huge opportunity, of course, to sort of globalize this framework. It's -- we're very interested in it. It's a harder lift than going to a new product category for the reason that, look, our -- the same paradigm sort of applies. Our models will be extensible, the sort of generalized time series prediction engines, so we have that head start going into a new market. But where -- in the U.S., we go from personal lending to auto lending, we understand the U.S. borrower. If we were to go to the U.K., we have no understanding of that borrower because the data sources are completely different over there. They don't have a FICO score. They have some different score and their credit files are different and they're impacted of whatever education they have and their propensity to repay is probably very different. So we almost have to start from scratch on the data side. And of course, there's different regulatory regimes. So I think it's a big opportunity. It's a little bit -- in the road map, it's probably behind expanding our footprint across credit segments in the U.S. But it's one that we think a lot about. It's one that maybe a good candidate for acquisition or partnership with some local entity that can bring assets to the table.

That makes sense. Sanjay, one question that I get from investors is sort of like how can I be sure -- how can I validate that Upstart's results are effectively better than the legacy, kind of, underwriting? What can I point investors to? How would you respond to that question? Like where can we see that this is a better mousetrap basically?

That's the $1 million dollar question. It's a question not only asked by equity investors but certainly back in the day by credit investors when we were new on that scene. And I don't know, I think the intuitive way to approach it is so we'll look at all the -- look at all the ways I can measure this, and we've put some of these out there. You could look at how our collateral performs against rating agency assumptions and base cases where I think it performs very uniquely. You could look at how it performs -- so obviously, we work with a lot of banks, about 50 of them now, and we always back test our data against their models or our models against their models. And you could see the delta that it creates, and we published some of that. We can generally double or triple people's approval rates without any change in loss rate. You can look at some of the tests that a regulatory such as CFPB has done a lot of testing with us against what they consider to be state-of-the-art traditional models, where traditional just means they're not using machine learning algorithms, and there's significant lift in approval and significant decrease in APR. So there's a bunch of ways or studies or points that sort of point to this. And then, of course, the reaction is, well, I don't know. There's a lot of noise out there, anyone can tell a good story, data, how do I know that it's apples-to-apples? And ultimately, what we sort of ended up with the credit markets is you just kind of have to look at outputs on some level. And I think people have been looking at our credit since 2016, 2017, and people who are specialized in that, I think they would tell you, okay, that something is different about the way this credit performs at the end of the day. And then with equity investors, it's like, okay, well, the outputs are, well, you can do your channel checks with credit buyers, and they're the ones who are specialized in this or you could look at my business results. And so the trends, stock market notwithstanding, I do think we've demonstrated an ability to grow that has kind of astounded people. And not just grow because again, the fast-growing lending platforms create immediate suspicion, right? There's been a lot of implosions. But how many have been able to grow with improving credit performance? And you could sort of understand the credit markets and profitability. And I think that the business outcomes, we hope, will be unique and hard to explain as well. And if those are being temporarily -- those have been temporarily distorted by a macro shock, and again, we've been transparent in saying there will be macro ups and downs along this journey. I think those will play out. And I think if you look at this over a longer term or if you do look back to 2017, you will see, I said at the business results that I think are hard to explain other than by saying, well, there must be some sort of differentiation in how the credit modeling is working. So I know it's not a very satisfying answer to someone who's trying to make an investment decision today, but that's the best I can come up with for you, Ramsey.

Yes. No. It's, again, super hopeful. There's a lot of data points in there that I think are really critical. One thing that I've also seen is the potential for some kind of cross-sell opportunity. And obviously, you've been -- there's repeat users. But how do you -- how can you port the productivity you're seeing in one product over in another? Are you expecting like a lot of auto customers to be existing personal loan customers? Is there anything you're seeing today that would show that there's some overlap? And also then, how do you execute on that?

Yes. Look, I think it's a huge opportunity, and it's all really in front of us. And when we think about it, the majority of the Americans we're trying to serve do require credit for their auto, for their home and for their expenses. They're moving cash around in time, like -- the average American that's in this segment that we're trying to unearth is a consumer of credit. And when you think about it, when we start to be at scale in these different products, we will have a database and we will have better credit fidelity because we'll know them across more relationships or products. We will have lower unit economics because the marketing will be much more direct. And so if we have an auto loan customer -- or I guess what's happening today is if you're a personal loan customer, we know what auto loans they have on their credit files, and we can see them and we can make a judgment as to whether they're overpriced or mispriced or not, and we can immediately reach out and offer to save them money. And the same will be true between autos and houses. So I think the prerequisite is to get big and meaningful and good at pricing in all these segments, and that's what we're in the process of doing. But once those are at scale, I think the opportunity is to cross-sell. Because at the end of the day, it's the consumer and it's an overall debt burden that we want to manage across. And I think the opportunities are very big.

I would agree with that. Sanjay, we're out of time, unfortunately. Thank you so much for being here today. Some pretty insightful answers. It was a great pleasure.

Thank you for having me. It's always a great pleasure.