Are we good? Oh, yeah, there we go. Perfect.
Yep.
Fantastic. All right. Good job, guys. Take two. Good afternoon, everybody. My name is Pete Christiansen, here with Citi Equity Research, covering digital assets and fintech services. For this next session here, we will be covering Paul Gu, Co-founder and CEO of Upstart. Before I get started, Upstart has asked me to read this disclaimer. Today's discussion may contain forward-looking statements that may relate to future results and events, which are based on Upstart's information available as of today and 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. Without that, now we are past that.
Let us really dive in here. A week ago, Upstart reported record Contribution Profit, but lowered the full-year revenue and EBITDA guide as worsening macro conditions pressured the balance sheet income somewhat. That split screen, strong fundamentals, but modestly weaker macro-sensitive earnings, is the central question for the next half hour. To discuss it, I am joined by the person who built the underwriting models and now runs the company. Paul, welcome. Great to have you here. Let us get right into it.
Great. Happy to be here.
Let's chat a little bit about alpha and beta. On the call, you framed growth as alpha plus beta times macro. This quarter, Contribution Profit grew strongly, but the guidance cut came from balance sheet marks. Today, how much of Upstart's earnings volatility is company execution versus macro sensitivity? What's the best metrics investors should watch to see that beta is shrinking?
Yeah. The reality of the business today is it's a high alpha and it's high beta. The execution in the business has been really strong. Really couldn't have asked for anything better. The fundamentals, I think the alpha is the part that I think is most predictive of the long run of the business. It's the thing that we actually control. It's the thing that, as a sort of system, as an engine, we can expect to continue doing quarter after quarter. Whereas macro, sometimes it's good, sometimes it's bad. We don't really predict it. We don't control it. There have been quarters, just as many quarters where it's been a helpful force as quarters where it's been a headwind. In the most recent quarter, it certainly was a headwind to the business.
But, to your earlier comment, in spite of that, we still did a really spectacular Q3. I think we'll still have a really strong year. Q3 record Contribution Profits for the second quarter in a row, 48% year-on-year. That was despite a nine-point average UMI headwind. On a full-year basis, if you just look at where we are today in the year versus where I think anybody thought we'd be at the beginning of this year, just from an interest rate perspective or from a consumer perspective, I mean, it's just been a more challenging macro year. Despite that, we're still going to do really good numbers, especially on the sort of core operating business where we raised guidance for fee revenue. We raised it, the sort of implicit guide for Contribution Profit, and that happened because of that really, really strong execution.
But at the same time, the business does definitely have a very real beta component to the macro. That's true in the core operating business, but it's even more true on the balance sheet.
Following up with that, per point of UMI, which is more sensitive, fee revenue or some of the fair value marks? Do these marks, I guess, move immediately, or is there some lag, kind of that dynamic that we see on the fee revenue side?
The balance sheet side is more sensitive, especially the fair value component, because if you just think about it conceptually, fair value is, it is not just what has happened in the quarter. It is a sort of accountant's expectation of everything that will happen in the future discounted to the present, and it ends up being a fairly complex calculation. Some of the timing of things, of when loans are delinquent or charged off, or even if they are current, like the way they get impacted is a little non-linear to the way that UMI or our view of the underlying consumer moves. So it does end up being more sensitive and also a little bit of a complex calculation.
I would say it is mostly intended to try to capture everything in period, but because of the complexity of how it is accounted for, there might in practice be like a little bit of a lag to some parts of it, but maybe like most of it is quick.
Oh. We will blame the FASB then. October origination slowed relative to Q3. How much of that do you believe reflects maybe a tick up in higher funding costs, or the UMI changes in demand and credit appetite?
Yeah. Macro was definitely a challenge going from September to October, with rates being the single biggest and most obvious driver there, where for the most part, our business reprices on rates pretty quickly. We do it about once a month. Most of our funding deals with partners, we're just a true marketplace in the sense that, if market- traded interest rates go up, that passes through to our borrower within a month's time, and we tend to reprice based on once-a-month numbers. Think of this as two-year Treasuries or two-year SOFR. And those rates went up a lot, especially at the tail end of September, and so that passed right through into the October pricing of how we did loans.
So, roughly a month to pass through. Was that the case, I guess, when we had the following COVID, we had a rapid rate increase. Following COVID, that was obviously a lot harder to digest for consumers and for everybody. Is there a different dynamic now, I guess, if we were to expect rate hikes would be, I guess, slower to implement, more gradual, versus what we had in the last cycle, which was a lot more vertical, exponential?
Well, one thing I always say that I think sometimes people get confused by is, in our business, the primary interest rate that we care about are market-traded interest rates, so we're benchmarked against SOFR or Treasuries. It's not so much what the Fed is doing.
Right.
Obviously, those have a relationship, but things that are already expected by the market are generally already priced in, so it's not like there's a double whammy there. But, I mean, yeah, ultimately, you look at the inflation numbers, the direction of travel there, that's going to inform what the Fed has to do, which probably is going to inform where the market trades. They're all interconnected, and ultimately, I think our business and I think the entire American consumer, I think the American electorate, I think everybody's just going to be happier if there's a lot less inflation for us too.
Fair enough. For borrowers that, I guess, would near the 36% APR cap, should we think of rising funding costs and the UMI effectively capping volume growth? Should we think of that as part of the rationale also for expanding into some of the secured product areas that Upstart's been getting into?
I wouldn't call it a cap on volume growth. I would really call it a headwind, and the difference is we don't really ultimately care about the level of the macro, whether that's UMI or rates or anything else. What we care about is the first derivative. We care about the change. If conditions in the world stayed exactly as they are today for the rest of time, we would be thrilled about that. That would be just fine for us because ultimately, what we're doing is we're just generating alpha. Every month, every quarter, our job is, we think about our job as like, let's churn out technology and marketing wins that will generate a certain rate of growth, and let's just keep doing that quarter after quarter. If you keep doing that against a stable macro environment, inevitably you build a much bigger business.
The challenge comes when you're doing that and at the same time either interest rates are getting higher or the consumer is getting less healthy. Obviously, then those effects start canceling each other out, and you can get less results than you were hoping to get. But as long as it's the case that the sort of U.S. economy, the U.S. consumer is not in some kind of structural decline where their real wages just keep going down or inflation keeps going up or interest rates somehow just keep on going up over and over again, then the business is in a perfectly good place. The dynamics you described are the right ones.
It is true that when you have a higher UMI or you have higher interest rates, some people that you could have approved at the APR limit go over the APR limit and can no longer be approved. But then, all the work that we do every month, every quarter, those technology and marketing advances that we always talk about, those things bring people back in. If you can cut out some of the risk that comes from not understanding who's likely to default, if you can reduce some of the uncertainty that lets you access sort of lower target returns, that brings people back under the limit. That's the work we're always doing, always pulling people back down. In our business, we just need there to not be too strong of a force pulling the other way.
We'll get into secured products later. I didn't suppose that or imply that the rate cap is a reason why you got into secured products. You obviously got there from technical know-how, building out the platform and the brand. Maybe in that context, you could talk a little bit about revenue diversification when we do get into a less favorable rate environment.
Yeah. It definitely is helpful to be in more segments and in more products. I think to some extent, they're all exposed to directionally similar thing. I don't think there's any kind of lending where it's better for interest rates to be higher.
It just isn't better. But it is certainly true that relative to having just a core personal loans business, I think the entire portfolio of products tends to be more diversified. You don't have as easy of an effect where people get pushed off the approval cliff. HELOCs is an interesting business because while people don't prefer to borrow at higher HELOC rates, there's sort of this offsetting effect where higher interest rates maybe makes people have different kinds of issues with their home equity and not being able to sort of move and get a new mortgage or certainly do any kind of refi. And so, HELOCs might become, relatively speaking, more attractive than they were in another environment.
Some of the less prime products like auto-secured personal loans or Cash Line end up being really good products for people who are getting pushed off the approval cliff in personal loans. So, there's definitely some good sort of ecosystem forces. But having said that, all loan products are directionally subject to higher interest rates is worse, less healthy borrowers worse.
Fair enough. This is one of, I think, the more interesting conversation lines that you've been talking about more recently. You were saying that the model is 2.77 x more accurate than a traditional benchmark, yet 80% of the improvement opportunity still lies ahead. What exactly is the denominator behind that 87%? In your view, what does perfect underwriting model realistically look like?
Yeah. If you had 100% of the remaining error, that basically means if you were to just make completely random predictions about who's going to pay back each month and who's not going to pay back each month, that would be the amount of error you would get at 100%. 0% would mean if you got all of the cash flow predictions right, so both the defaults and prepayments in every month of a loan's life, you've got all of those predictions right, you'd be at zero, and that's the number against which we're at, like 87.
I guess as AI models become more and more sophisticated, the compute cost per loan, is that rising or is that falling, and does compute ever become a constraint on the model complexity?
Yeah. Compute is actually a pretty big issue for us. The combination of compute, cost, and latency have been sort of It's a real sort of major frontier of work for the company. I think in some ways it's actually a testament to how real the kind of Everybody likes to talk about using AI in credit, but really, I don't think there's actually nearly as much action as there is talk. One of the places you know it is whether you have to invest a lot of time and effort into managing your latency and your compute cost. For us, for better and worse, it is an area that we spend a lot of time in. The practical answer to your question is, it goes up and down because the natural forces, our models get more complex.
The amount of compute they eat up grows somewhat quickly, and that gets both very expensive and very slow for the customer. Then we have to pour more effort into the R&D side of managing that compute cost and latency. Then with some optimizations, we bring it back down, then the cycle begins anew.
I guess, are you still seeing that, I guess, as token prices are declining to some degree, you're still seeing these increases?
Well, the types of computations we do are just totally different than sort of LLM tokens. I mean, they kind of maybe have similar dynamics. But yeah, we're kind of just off in our own world there, where it's not like anybody else's advances reducing token costs are going to help fundamentally our compute cost issues. They're very proprietary algorithms that we've developed and we're using specifically for this consumer lending use case. So again, for better and worse, it just is like, it's our own battle to fight where we have to do fairly low-level optimizations to make this stuff more efficient.
Right.
We just constantly do that. As soon as we make the cost more efficient, the latency fast enough for the user experience, then we unleash the model developers to give us the next generation of models that will bring the cost and the latency right back up, then we start over.
Version 26, 27? Or what are we up to? I forgot. Maybe it was—
No, we're in the 30s now.
Oh, in the 30s?
0.
Jeez. I got to update that one. At the Investor Day, Upstart said that a 1% improvement in accuracy drove roughly 13% better conversion. This was, yep, a year and a half ago, roughly. Now that conversion rate disclosure is going away, what single metric do you believe best captures whether model improvements are creating value for borrowers and/or funding partners?
Yeah. We manage the business to Contribution Profit, and I think everything kind of rolls into Contribution Profit. I think what we found is when we tried to disaggregate that into these component metrics, they'd always get confused by mix effects or sort of refilling the funnel with more borrowers that you can now economically source. Ultimately, everything flows into Contribution Profit. When we get smarter models, we're going to cut. All else equal, we're going to convert more people. We might refill that bucket, but the way you're going to end up seeing it show up is more Contribution Profit. That's the number that every trade-off gets weighed against in operating our business and our models and everything else. That number, I think, ultimately is the one that you got to watch.
Fair enough. I think the sell side is starting to change their thinking there as we get off of a five-year, six-year metric. I really like this question right here. You just mentioned this briefly before. Every lender out there now claims to have and use AI to some degree. In your view, what's the strongest argument that Upstart's underwriting advantage compounds rather than commoditizes? If you're wrong, where would investors see that?
Yeah. Well, coming right back to Contribution Profits, I mean, there's a reason that's the metric we manage the business to, and I think the one that's worth looking at, because in some ways, it is the best single metric that encapsulates how much you're growing and also how much you're growing in a way that's actually creating value and capturing value. Obviously, you can only do that if your product is differentiated, if your underwriting is differentiated. If it's not, if you're competing in a commodity business offering the same thing as 10 other players who are able to underwrite the same way, then yeah, you're not going to be able to get the margins that you're looking for. Contribution Profit kind of captures it.
I mean, to me, the proof is ultimately in the pudding of how's our Contribution Profit growth compared to anybody else in the market? If you want to talk about underwriting specifically, ultimately you should look at segments of the market where underwriting matters, which tends to be consumers where there's some risk. If you look at the sort of consumers where there's some risk, I think you'd just be pretty hard-pressed to find anybody that has Contribution Profit growth rates anywhere in our neighborhood. I think if somebody could sustainably do the kinds of numbers that we think we're going to be able to do there, then I would admit that maybe they're doing something interesting—
Right
—on the underwriting side, but to date, I haven't seen it.
Interesting. This has been a hot topic throughout the whole conference. I think we just had a session not too long ago about it. But the use of AI agents in financial services more broadly. I guess you've argued that AI agents that may become the primary interface for shopping for financial products, in that world, how would you define Upstart's moat? Is it still some degree of distribution? Is it really the data and the underwriting, or is there something else?
Yeah. We think that agents push the world towards a place where the best product wins. I think as compared to a world where maybe in a sort of human-operated world, it's much more the case that we have very, very finite limits to our time and attention and our willingness to go and check all the options—
Right
—to find the very best one, to sort of consider new ideas, change. You end up with a lot more walled gardens, which I think frankly is what a lot of players, both incumbents and even fintechs tend to rely on, is their strategy is about like, "How do I sort of get the customer to sign up for one thing and then not check other things so that when they next need a thing, maybe my product isn't the best product, but they'll stick with me?" That's proven to be a pretty good strategy for a lot of players, and obviously many, many really valuable brands, franchises have been built on that strategy. I think that comes really under question in a world where suddenly the cost of checking all your options to make sure you get the very best one, it approaches zero.
If agents make that cost zero, then the best product just wins, and that's a world we would be really, really excited by. We'll see how long it takes to come into fruition, but it's definitely something we'd be a big supporter of.
You don't need a distribution deal, just the truth, right?
That's right.
Super interesting. I'm using Investor Day as a nice comparison timeframe again here. auto refi was a major focus at that point, but then it was later paused, and now it's back to some really exceptional growth just in the last few quarters. What did that experience teach you about where Upstart does and maybe it doesn't have an edge? I know you've rejiggered that product area quite a bit. Take us through that process that you went through.
Yeah. Today, we are really happy with the progress of our new loan bets, and I would say all of our new bets are ones we have scrutinized really carefully, made sure that they are clearing a really high hurdle rate for investments in new areas. To your point, we used to have an auto refi product that we are not doing anymore, but the remaining set of auto products, the home product, all the secured products and Cash Line are all things we are really excited about. The secured products just had a massive improvement in their Contribution Margins this year, flipped positive. Really good wins there. I think the auto refi experience specifically, I think what I took away from that is today we do not have the scale to try to educate the market on things that are not already big things in the market.
Auto refi was a product that it was not, frankly, it was not a big established market. There were not a lot of people out there looking for an auto refinance loan. In order to win, you not only needed to have a better rate, you needed to have a better rate by enough that somebody who was not looking could be convinced to care about this. That is just a high bar. I think there could be some world where we have the customer ecosystem and the sort of relationship with customers that you can do things that maybe they were not already thinking about.
But today, I think it just makes much more sense for us to focus on products where there is no question that the demand and the appetite is out there, and what is missing from the market is a technical ability to serve that demand, and that is what we want to focus on today.
It is a lot of operational things, the two that needed to be ironed out.
For sure.
Right, yeah. The secured contribution margin this quarter improved from roughly - 35% to + 17% in just one quarter, well ahead of your target. How much of that came from structural improvements like automation versus, I guess, better loan sale execution and monetization?
All of the above. Really, yeah, I mean, we've been so happy with the progress on secured products this year. We made it a major priority for us. We said, our number one goal in secured products this year was going to be getting them to contribution positive by the end of the year. We did that a quarter early, and it was exactly all those things you listed, which is in order to do that, you have to sell most of these loans to third-party buyers in order to even be able to mechanically recognize some of the revenue there. You have to make sure that those deals are actually good enough deals that you're capturing enough value on the deals that the math makes sense.
And then your cost side of the ledger has to be good enough that the amount of revenue you're recognizing is bigger, and we worked on all those areas, and I think is a big win.
So in the secured world, which is a fascinating growth aspect to the company right now, if you think about reinvestment, where does that next dollar go in the secured credit opportunity for Upstart?
Yeah. Each of the secured products is in a kind of different place. I would say the auto-secured personal loan product is the one that is probably the most similar, the most tied into the ecosystem that we already have with our core personal loan product. It's pretty adjacent customer. They similarly find us through upstart.com. So that one, I would say, sort of has the fastest path to being like a mature product in the sense it fits right into the rest of what we do.
The auto purchase product is in many ways dramatically more complicated of a product, but it's the one we've been doing the longest, and we spent a lot of time, I think longer than probably we would have expected when we first got into the business, but now we've put in the hours and we've really put in the effort and put in all the iterations, and we've kind of figured out how to work with car dealerships to distribute to the customer. So that product is, despite being quite difficult, we've put enough into it that I think we're pretty much there. I think that product I don't view as one that's going to need to take a lot more investment. The HELOC product is the least sort of mature of the group, and the HELOC product obviously has large market.
I think it's something that's really interesting for a bunch of reasons. But when you look at the HELOC product, its ultimate distribution channels, a lot of those are very likely to be distributed through home- adjacent partnerships, and those are partnerships that we don't have today, are things that we haven't historically done. That's kind of a whole new area for us, and I think building that up is probably going to take some incremental investment dollars.
We also have an existing borrower base over the years. I would imagine there's some awareness campaign that needs to happen here so that people understand Upstart is more than just a personal loan company.
Yeah.
How do you believe that awareness is tracking right now?
I think there's a funny dynamic in our business where, on the one hand, we've talked about how more than 25 million unique Americans have applied for a loan with Upstart. That's more than 10% of the relevant adult population. It's a huge number. At the same time, somehow despite there being 25 million people who've done this, probably awareness outside those 25 million is almost none at all.
Right.
People just don't know about Upstart. I think actually that is, in many ways, kind of a reflection of a remarkable amount of upside and also a remarkable amount of discipline in how we've marketed, is we have been very focused on marketing to people who need this product, need it at that moment in time, and it's been a remarkably effective strategy for reaching those people. But we haven't done any kind of broad-based marketing awareness campaigns or anything like that.
And I think our belief is that the sort of natural bridge between those two is that eventually, everybody needs credit, some kind of credit at some time, and I think there's a path where you can organically get from the first thing to the second thing by just economically reaching people at the moment they need you and not having to invest so many dollars into reaching people when they don't actually need the product.
Fair enough. That's a really good point. I want to skip around here. First of all, from marketplace to bank, it's a big topic right now. At Investor Day, there was this view, "We never want to be a bank," right. And now Upstart Bank is the largest strategic initiative. What did you come to believe about the business that maybe you didn't believe a year ago?
Whenever we've said we don't want to be a bank, really the—
It's also the semantics. I—
Yeah. The spirit of that statement was really like, when people think about a bank, they think of a business where you have to fund your own originations with your own balance sheet, and that means that your rate of growth is capped by the amount of equity capital that you've got, and we never want that to be true about our business. We don't intend that to be true even hopefully after we have our bank charter. We still very much intend to work through third-party capital partnerships, have the vast majority of our originations funded by third parties, and that's really so that the rate of growth in the business is determined by how many of these technology marketing wins we can churn out and not by the amount of equity capital that we've got.
We want to be super efficient with equity capital, and we also want the business to grow as fast as it reasonably can on the borrower side, and so that's the combination that works. Tactically, we wanted to pursue the bank charter because it helps us operate efficiently in all 50 states with direct regulatory relationships, and that's going to be a real nice tactical win for the business, but it doesn't really change the underlying strategy that we want to be able to grow untethered to our equity capital.
I would imagine it's a structural margin driver, of course, especially from the origination operations side. I know you haven't sized it yet for us, and there's a long way before we get to that point, but I guess generally, does this change the dynamics of the P&L significantly or any sense there?
I think in an accounting sense, it will certainly move some of the way that things are represented around. In a true economic sense, of course, it actually does improve the monetization of our loans because there's less third parties in the middle that are taking some of the economics, and so that's a real win. But I wouldn't call it a fundamental change to the P&L. I would say this is one of many things we could do that would drive more revenue and profitability in the business, but that's kind of what we're always doing, and I think we always get this question, like there's some confusion, "Well, you used to have this long-term outlook for growth that didn't have the bank in it. Now you have the bank.
Why isn't it additive? I think the answer to this question really is just because the bank takes work and other stuff we would do take work, and for better or worse, the thing in our business is we put in work, we get out growth, and I only have so many units of work to put in from a fixed team. If I'm putting that work towards the bank, which is a lot of work, it's a lot of work to set up a bank, it has great benefits, but also if I wasn't doing the bank, it's not like I would be doing nothing with those units of work.
True.
It would be producing other kinds of wins that would also be generating growth and profits in the business, and so that's the way in which I think there's sort of a more, in my mind, almost more predictable relationship between input and output in our business, and the bank is just one way of doing input, and there's others.
The next logical step, that kind of
Yeah, that's right.
I got you. I do want to tap on capital and credit to some degree here. So recent vintages have been generating attractive spreads, yet the beneficial interest marks have been negative for two quarters here. Why are those signals diverging, and maybe how should investors think about reconciling that?
It's pretty simple. In a better macro climate, the spreads would be even better, and I think the spreads are good now, they're attractive, and they're good spreads relative to benchmark, but I think in a better environment, the spreads would have been even bigger, and the beneficial interests are ultimately, they move relative-
Right
relative to the previous numbers, and so anytime macro gets worse, it's going to tend to put pressure on the fair value marks.
Also, I would imagine, though, the actual underlying securities themselves, which have different characteristics, I guess if we think about what the balance sheet looks like, Upstart's balance sheet right now, does that change the types of co-investments that you make after you do become a bank?
I don't think becoming a bank will necessarily change our strategy with respect to the co-investments. But separate from the bank, we definitely have an intention to continue improving the efficiency of our co-investments program and of how we do loan funding in general. We talked a lot this quarter about wanting to bring down the equity usage percentage in supporting our loan funding. That's a pretty major priority of mine. Specifically with respect to co-investments, the same thing is true. We want the amount of capital that we have to contribute to the co-investments as a percentage of all the loan funding, that number is one that we want to bring down over time. That's definitely very important, and it would be important with or without the bank.
Wrapping up here, to think about 2027, key milestones that you're looking to achieve, obviously the bank is one of them. Maybe discuss some other major priorities that you're seeing for the upcoming year.
Yeah. I think we've talked a lot about our priorities. In some sense, 2027 is just a year where I expect continuation of the core strategy, because I honestly think the core strategy is working. It's working really well. You see it in that 48% year-over-year contribution number, that Contribution Profit growth number. It would be even higher without the macro headwinds we've been seeing recently. Those are phenomenal numbers. We'd be really, really happy to keep doing things like that. That's coming from this continued focus on core personal loans. It comes from, we talked about 2026 was the year that we really wanted secured products to get profitable, and 2027 will be a year to return to growing on those products. That's sort of big focus for us. Then ultimately, I think there's this piece about operating leverage, which is it's just time.
It's time. We've been around for 14 years. I think it's time to start making a meaningful amount of profit in the business. 2027, I would like to be judged on whether we're producing records that are better than that macro-supported 2021 year on every line of the P&L. That starts with originations and Contribution Profit and goes down to adjusted EBITDA, and ultimately, GAAP net income is like, we should be doing record numbers across all of those lines of the P&L. I think if we did that in 2027, a year that probably is going to have far, far, far more challenging macro conditions than we faced back in 2021, I think that will be a pretty strong statement of the amount of progress we've made at that time.
Sounds fascinating. We are certainly behind that story and would love to see that come to fruition. So, best of luck with that. Paul, thank you again for appearing today. Great conversation. We covered a lot of ground. Thank you.
Great. Thanks, Pete.