Ann Berry (00:00):
Anyone who's tried to get an auto loan or a mortgage or any kind of personal loan has seen, they have a FICO score. This measures individual credit worthiness. A basic FICO score ranges from 300 to 850 with a higher score tending to mean the borrower can access more credit at cheaper interest rates. FICO scores have been around for 35 years and are so widespread that 90% of top lenders use them. Landlords often also use FCO to assess renters and insurers use it in underwriting decisions. The problem is, is that it's far from perfect. The FCO score calculation mainly looks at how much of your available credit you've used, how timely you've been in repaying loans and what kinds of credit you've had. It doesn't account for recent changes in financial health, such as switching out jobs as is often a lag before FICO scores. Update. The limited variables used means that data points like earnings potential aren't factored in making FICO scores especially restrictive for young adults without much credit history yet, for example.
(00:50)
So a couple of FinTech companies are trying more sophisticated ways to assess credit worthiness. One company is upstart, founded in 2012 by a team outer Google. I went on upstart's website and was asked the usual questions like my income, but also additional ones like where was I educated, what did I study, and what's my position at my current employer? Upstart went public on the NASDAQ in 2020 at $20 per share and a market cap of about $1.5 billion. And the stock price has been on a roller coaster ever since reaching a high of $401 in October, 2021 before dropping to trade over the past 52 weeks in the 20 to $90 range, upstart most recently reported third quarter revenue of $161 million up 20% versus last year along with new funding agreements. We sat down with CEO, Dave Girouard to dig into how upstart's products work, why some on Wall Street are skeptical of AI in FinTech and where growth comes from next. So Dave, let's start with the basics. What is wrong with the personal loan market in terms of measuring people's credit worthiness and what is it that you guys set out to do to fix it when you started this business coming out of Google in 2012?
Dave Girouard (01:56):
Well, the personal loan market is really just a starting point for us. The issues that we saw on credit access actually span virtually all flavors of credit, and that's basically that the systems used to evaluate who should get a loan and at what price, I mean, are quite antiquated. The credit score was invented more than 30 years ago now, and it's still kind of been the state of the art on can I get a loan at what price, what size loan, et cetera. And coming out of Google, we really thought, you know what? There's got to be a better way using a lot more sophisticated math, a lot more data, what's become known as AI to make much more sophisticated credit decisions to better understand risk. And the impact of that would be dramatically improving credit access for almost everyone.
Ann Berry (02:42):
So let's talk about what that means in terms of credit access. I took a look at your website, Dave and I saw that upstart, you guys talk about having 38% lower rates typically. So that's the interest that I'd be charged as the borrower. You talk about 91% of loans fully automated. This blew me away, 101% higher approval rate than traditional models. What are the different kinds of inputs that you are using in your data analysis to assess someone's credit worthiness that's different from just the good old FIO score?
Dave Girouard (03:11):
Sure. Well, most lenders, they don't just use a FIO score. They use A-F-I-C-O score and maybe 10 or 11 or 12 other variables they might get from a credit report like how many trade lines you have outstanding and how much outstanding debt you have, et cetera. Our model uses more than a thousand variables, and you can think of all sorts of information about your credit and your use of credit in the past also about where you work and how long you've worked there. Just a ton of information actually where you went to school and what you studied. So we're really trying to build more of that 360 degree understanding of the person really all in the interest of finding more ways to prove that you're credit worthy. And that's essentially what our model does is it keeps mining and finding new ways to prove credit worthiness because in the real world, about 50% of the country has access to prime credit based on their credit score, the type of credit you could get from a bank about half the country, but more than 80% of the country has never actually defaulted on anything. So you have this major discrepancy between who has access to credit and who really should based on the real data.
Ann Berry (04:21):
I remember when AI started to make moves into credit analysis, Dave, and there was a lot of chatter. It was speculative chatter about whether completely different variables like an individual social media profile and the activities we seem to be doing there or what our shopping habits were would start to make their way into these kinds of models. Does upstart use any of these really non-traditional variables in its analysis?
Dave Girouard (04:45):
No. I mean there's plenty of information that's much more obviously useful if people make their rent payments, things of that nature. Then going into esoteric things like social media also really just there's no basis for using it. How would you know whether or not it actually is predictive in any sense? So we don't really use anything that you would think of in that sense, social media or data of that. Like
Ann Berry (05:12):
You talk a lot in the upstart materials about using your models as a way to fight bias and that there's been endemic biases and a lot of credit issuance. Talk to us about what the problem has been and what you guys are doing so differently.
Dave Girouard (05:28):
Well, credit access has always been skewed for sure, and some of it is just the inputs, meaning some demographics have lower income historically than others, but credit score is one of the more biased tools out there. Historically, about 30% of black Americans have credit scores in the lowest decile of the US and that really limits access to credit. So using such a simple tool as that tends to have a lot of bias it when you start to look to other ways to prove credit worthiness, as I mentioned, whether it be someone could be a nurse or they could be in the military, both of which are high indications of you're going to be employed, you're very unlikely to become unemployed. Things like that that are just indicate credit worthiness are just different ways to prove someone's credit worthy. And so we can actually reduce bias by the use of ai.
Ann Berry (06:22):
You've also got this overlay, which looks at the macroeconomic environment, Dave, the UMI. Talk to us about what that is.
Dave Girouard (06:31):
Sure. Well, one of the things we learned in the last couple of years is when the economy's changing quickly, your lending models need to respond very, very quickly. And one of the things that's very hard for a lender is to separate the idea of the risk of an applicant and of a loan from risk that is endemic in the overall economy and how that's changing. So what we've been able to do over time is to actually unbundle these and separate them and really better understand how is the economy today impacting the performance of credit? And this is what we came to call the upstart macro index and we publish it publicly so it's available and it's just really been a good way to understand what's gone on in the last few years, particularly a lot of really unusual sort of outcomes from the pandemic, the stimulus, the de stimulus that happened when the pandemic was ending, et cetera. So a lot of important for lenders to make smart decisions, not just gut-based decisions, but data-based decisions on their lending programs. And that led to us to the development of Upstart macro index.
Ann Berry (07:37):
And what's the index telling you today about the health of the American consumer?
Dave Girouard (07:42):
Well, the American consumer actually, and this is a little counterintuitive. Unemployment rates are still near historical lows in the range of four, a bit over 4%, but at the same time, Americans are not as financially well off as they have been in the past. The way we understand this is essentially they're spending more than their income compared to historical norms. The personal savings rate is information that Federal Reserve publishes regularly every month. And historically the personal savings rate in the us, which is really just measuring all the income minus all the expenditures, and the net is the savings rate. And that savings rate is typically in historically in the 8% range. Today it's it's more like 4%. And that is really Americans who are still kind of, maybe it's a little bit of post covid revenge spending or something of that nature, but their spending habits have gone up. The government stimulus has long gone and there's been inflation of course, and inflation has made that a bit worse. So that nets out that today default rates are still significantly higher than they were in a normal sort of long period of time. We're probably in the range of 50% higher from our point of view. And that's something that our system adjusts for. So it's part of why we built UI so we can properly calibrate to that type of information.
Ann Berry (09:06):
And Dave, is your analysis able to capture buy now, pay later balances that consumers have got outstanding? Cause I know that's been a bit of a gray area. Not all folks trying to assess credit worthiness have been able to get their arms around how much is out there and who's got it. Exactly.
Dave Girouard (09:21):
Yeah, that's a good point. Buy now, pay later is a new form of credit, and for the most part, they don't always report to credit reporting agencies, which is the only way that we would have visibility to that. Some of it is really just what they kind of described as pay in for you just pay it off pretty quickly, which isn't much different than a 30 day grace period on a credit card, so that's no big deal. But if they end up having longer term significant credit balances that they're carrying, ultimately it's important for the whole system that become visible to all because you don't want people getting overindebted and in a place where they can't afford the loans that they have.
Ann Berry (10:01):
So Dave, to take a big step back at Upstart, you've talked about the ways in which you assess credit worthiness better than traditional models. Let's talk about now how the loans actually get issued. You guys don't carry a ton of loans on your own balance sheet. You made the decision, for example, not to acquire a bank. You're using other people's money, you're providing the underwriting tools, but it's other people's money that's actually being distributed to consumers. Talk about that decision. Why did you decide not to become a loan issuer yourself?
Dave Girouard (10:30):
Well, I think for a lot of reasons, banks are regulated in a very specific way. They have the unique ability to take deposits and make loans from those deposits, and with that goes a lot of responsibility and really limited ability to take risk, which is why, as I kind of said earlier, only a fraction of the country can actually get a loan from a bank because loan's ability to take on risk is heavily managed by their regulators, and that's important to know because if you want to push the bounds of credit, invent something completely new. We all know AI is new and different and it's scary to some people. So doing that as a bank would be, I think very challenging. What we've really built as a marketplace with a lot of lenders on one side as well as credit investors, because when the credit may not make sense for a bank to hold, it can be sold on to a credit investor who's more adapted to take on that type of risk and expect the type of returns from it. So in our view, a marketplace structure where there's consumers, borrowers on one side, there's a lot of lenders in various sources of capital on the other, is actually the most efficient way to push the technology forward and to have the biggest impact that we can.
Ann Berry (11:46):
I saw, Dave, that you'd signed a $2 billion capital supply agreement with Blue Owl, and there's been a lot of discussion about the private credit industry, so non-banks, some cases called Shadow Banks moving into capital supply in new and creative ways. Talk to us about that partnership with the private credit industry. Do you think that's going to grow?
Dave Girouard (12:08):
Sure. Well, I think it's funny. The term shadow bank almost seems to me a little comical these days because it just means sources of funding that are not from the banks themselves, but private credit has really taken off independent of upstart of course, and banks have receded in lots of aspects of lending, whether that be in commercial lending, whether that be in mortgages where banks have really given up market share for a long time. So that's a trend that certainly comes before us, but we began working with Alaya who Blue Owl acquired and the sort of deal came together about the time that Alaya became part of Blue Owl, but basically I think a lot of the private credit market has looked first towards direct lending towards companies. That's sort of the bread and butter of what private credit is, but I think that's also become quite heavily competed. And so they also looking for other classes and asset backed products like ours, which is consumer loans is just another category. So Blue s become an important partner. We have skin in the game, we co-invest with them, which is a structure that we decided to move toward a couple of years ago really to ensure alignment with partners and to make sure those partners would be with us through whatever economies we face in the future.
Ann Berry (13:29):
Let's shift gears a little bit and talk about your recent earnings came out last week. It was a fantastic quarter for you guys, your revenue 8% beat on the top line earnings per share, 57 beat versus 7% beat versus expectations. What were some of the highlights as we break down those numbers, Dave, that drove the acceleration in your top line and drove the improvement in margins?
Dave Girouard (13:52):
Do I think people leap to the conclusion that the Fed dropped rates 50 basis points and other 25 basis points recently? Maybe it's just an improving environment, and I would say that that was a little bit helpful, but honestly, most of the wind came from new versions of our AI models that are just much more productive, much more constructive, and we launched a model we called Model 18 just a few months ago, which really has a much better to separate risk, much greater ability. What that generally means is when our models get more accurate, we end up approving more people at lower rates, and that's generally how we grow as a company. So lots of things contribute to it. We are a model driven company, so there's just always new versions of our technology coming out. They lead to higher levels of automation, they lead to more people getting approved hopefully to lower rates.
(14:41)
And as the technology continues to get better, that's always been the basis of our growth and we're finally getting back to it. I think it's been not a very conducive lending market for a couple of years, 2023, there were bank failures, there's just a loss of liquidity across the banking system, but all that really has started to improve and at the same time we've been working on the fundamentals of our technology and our business and I think finally this quarter, the earnings we reported just last week, it really started to just all come together. Finally,
Ann Berry (15:17):
Let's give you a bit more of a victory lap on that. Dave revenue was up 20% year over year to $162 million. Transaction volume grew 30% year over year. Now, you did touch on how the models have improved your conversion rate. I saw that went from about 9.5% to 16 and change percent. So your conversion rate's up, but how much of the volume growth is because A, you've got new sources of capital so you're able to actually issue more loans in the event that you want to B, how much of it's because you're seeing demand coming more from that stress consumer as you said, relative to historical levels where you said the saving rate's down to 4%. If you are issuing more to consumers who are more stressed, what is the risk that the quality of their credit worthiness is not as high as perhaps you would want it to be at these bigger volume levels?
Dave Girouard (16:08):
I think at the consumer level, we haven't really seen, I mean it feels to us there's almost a constant consistent level of demand for credit. It doesn't sort of wave that much well, although there is a bit of seasonality to it. So I don't think it was sort of a swell in demand from consumers. I think the important thing about what we do is really identifying the appropriate amount of risk per applicant in any scenario. So if somebody is heavily indebted and really trying to refinance, et cetera, our models understand that if they've only been they're not employed or haven't been employed for very long, the model understands that. So understanding the macro environment is important, but even more important is understanding the specifics of an individual who applies and the models are particularly good at that. So it's really not a function of more demand causing a swollen growth. Really, again, just the models when they get much better. The conversion rate, the ability to get people through the funnel and to get a loan improve a lot.
Ann Berry (17:14):
Let's talk about the reaction of the market to the strong earnings report. You had your share price go up nearly 50% in one day. To your point a little bit, the Fed news helped that right this moment you're north of $60 a share, you iPod at $20 a share, but I want to touch on what some of the Wall Street analysts are saying about upstart. Dave, I have in front of me a Morgan Stanley report, a Goldman Sachs report both issued after your earnings came out, despite the strength of your results, both have upstart as sell rated, one's got your price target at $12, the other at $13 50, which is a massive decline that's being implied relative to where you're trading today. Their concern is that in scaling there's potentially lower quality credit over time. What do you say to Goldman Sachs and Morgan Stanley? What are they missing? Why are they so disconnected from where your share prices is today?
Dave Girouard (18:09):
I think they lack a fundamental understanding of our business and who we are and how we do what we do. Some of them are new analysts or they're just analysts that maybe they have too much to do. I don't know. But we actually have a lot of analysts who have moved us to a buy rating a lot. The majority of have actually moved prices up pretty significantly closer to the range where we are and some above where we are. So we look across all of 'em and say, we're the type of business that is hard to understand. Performance and credit is something that proves itself over a long period of time. And so there's just plenty of reasons historically, particularly if you come from financial services to be skeptical, there can be really a new and different way to originate credit. And so some of these analysts, they just come from that world.
(18:55)
They come from evaluating banks and looking at things sort of looking backwards. But AI is a very new and different tool. It really wasn't available in any kind of commercial sense just a few years ago. So I think it's perfectly understandable that there are those who don't yet understand the potential for it. And maybe again, they'll just take more time to see the results and appreciate it. But there's certainly a lot of other analysts have just call out analysts from C analysts, from Barclays that have much more favorable looks on us. Mizuho is another one, so you can't get them all. Like I said, I think there is inherent skeptics in the domain of lending that it's always just the same, right? Everything looks good until it doesn't and everything under the sun's been invented in lending and we're a very different company. We're out to say, look, no, you can do this in a much better way, in a much more efficient way, in a way that's better for the lender and far better for the consumer. And it's using a tool that's suddenly we're all hearing a lot more about ai. And given what you see in AI these days from chat GBT and all the noise about that, it's not really a huge leap of faith to believe that it could make lending far more accurate and far more efficient.
Ann Berry (20:13):
The first indications that information, other than some of the traditional metrics, Dave could be used to assess credit worthiness. I remember in 2007 there was subprime lending in the mortgage space and then also student loans. You may remember there was this move into looking at student loans beyond parents FICO scores or the students obviously themselves didn't have one yet, and looking at their potential earnings trajectory was it didn't end so well in 2008. In some of those cases, a different kind of asset-backed securitization market. Yes, it wasn't called ai. Yes, there wasn't level of computational sophistication that you have today. Do you think that we understand those risks better? Do you think for some of the skepticism that we're seeing from Wall Street, it's because there's still a little bit of an overhang of new ways of trying to assess credit worthiness. There's a nervousness that if we're not there yet still we could be looking at another potentially catastrophic outcome if we get too ahead of our skis.
Dave Girouard (21:12):
Well, I think understanding risk is always a challenging problem. And as you said 20 years ago, 15 years ago, there was not the computation that was not the technologies to make the leaps forward that I think are possible today. Our company really couldn't have existed in my mind 20 years ago. There just wasn't the basis. Today you have cloud computing, which is a hugely important underpinning that enables ai. Suddenly, AI itself has taken off. That's all relatively new and credit. You can't prove it in a moment no matter what. You have to originate loans and see how loans perform. So there's a certain amount of time that has to happen, but we generate all the training data on our platform. We have the best proprietary training dataset we have. And at a simple level, I mean, if you could know five things about somebody before making them alone, or if you could know 20 things about somebody before making a loan, which would you prefer?
(22:12)
Right? You'd probably rather know the 20. So even though it will always be imperfect, the question is with much more data and much more sophisticated algorithms to understand that data, are you in a better position than just using far less data in simplistic algorithms? I think it's almost inevitable that you're in a much better position. The other point I will make, as you noted earlier, over 90% of our loans are entirely automated. That is a very powerful function because when you respond to the consumer quickly, it's not only less costly for the lender and for us to originate that loan. It's also, of course, much more valuable to the consumer to get a response in a moment, be able to get that money quickly. It also selects for better borrowers. So you have these sort of secondary effects that I think the AI technology brings to the market that makes for a fundamentally better loan product from the consumer's perspective, both in terms of the price and the process they go through. And that value accrues not just to the consumer, but also to the lenders.
Ann Berry (23:14):
You've changed your models on something like 77 million payment events. Your data set is large, Dave. There are potential competitors and actual competitors out there with bigger data sets, whether it's the Visas and the MasterCards who collect similar information, plus have the benefit of seeing what their consumer's shopping and spending habits are to folks like PayPal that have got billions of transactions they're processing every year. Do you worry at night about how those kinds of incumbents with massive data sets could come in? Do you worry about other folks like SoFi who are actually doing some of the issuing as well? Where do you feel about your position in the competitive landscape?
Dave Girouard (23:52):
I feel quite good because I think when it really comes to fundamental application of AI to credit origination, there's nobody in our class. There's nobody. And if you looked at whether it was Visa, whether it was one of the money center banks, they don't have either the risk tolerance or the skillset or the desire to really pursue this aggressively. I think they view it if you're a money center bank, you just want to sort of cherry pick the best of the best and have them as your customers not really interested in exploring how to better understand and better serve the torso of America, if you will. So there'll always be lots of competition, but most of them I think, are not focused fundamentally on AI as a way to create a better product, and eventually they will be because I think this technology will be so dominant that if you could look a decade in the future or more, what you'll find is that it will permeate all forms of credit. And at some point, everybody, every large bank, every small bank, every credit union will be like, where do we get this technology? Do we try to build it ourselves or do we try to partner with somebody? Well, we aim to be that partner, realizing that some of them may try to build it themselves, but most of 'em will recognize quite quickly it's beyond their means to do so, and we aim to be the partner to those lenders at that time.
Ann Berry (25:13):
Let's close Dave on building and growing because clearly something upstart's very focused on right now. You're out in the market raising capital. Tell us what's going on by way of your capital raise, and what is your plan for the use of proceeds?
Dave Girouard (25:26):
Well, look, I mean, we're a fast growing business. If you looked at us past our IPO in 2020, you can see the kind of growth we're capable of delivering as well as profitable growth. And as we just sort of look forward to where we want to be, I think it's important that we have just more ballast, more cash on hand and more a stronger balance sheet to be able to make sure that no matter what happens in the future, we're there also just being able to invest in the AI that we have. So for us, again, it's not a sort of particular use of funds. I think are a very efficient company, have always been a very efficient company. If you sort of looked at UPSTART compared to its peers, raised probably an order of magnitude less money as a private company, still had most of that money when we came public.
(26:12)
So we've always designed our business to be quite capital efficient and I think that's who we are and who we'll be in the future. But at the same time, I think there is a unique opportunity to be the leader in AI enabled lending. I think it's an enormously important future for the industry. And even though there's enormous number of skeptics, I think it's pretty clear to more and more people that AI has enormous potential in this category, and we want to make sure we have the capacity to pursue things as they become available, whether that be acquisitions or whether that be new product areas that we want to pursue.
Ann Berry (26:49):
And if you were to go after acquisitions, Dave, what are the kinds of targets that would pique your interest?
Dave Girouard (26:57):
Well, I mean, I always think of acquisitions as when we decide we want to get from here to there, we can build something, or maybe we find a faster way through an acquisition where there's some talent and capability and product available in the market. So acquisitions are a bit of a means to an end that we've decided is an end we want to pursue. And so they can be adjacent product areas. We don't have anything. We are not a credit card product or we're not involved in that part of the credit industry at all. We're only a US based company today. Everything, an entire business is within the us. So a lot of potential beyond the borders of the us. These are just things we think about, but we're quite selective. We've literally made one acquisition in our 12 and a half years of history. So we're not one to run out and do these things. But I do think acquisitions can be a very useful tool to accelerate your business when you really have clarity of where you want to be.
Ann Berry (27:56):
Dave, you're all excited to keep following you as you continue to build upstart. Thank you very much for joining that set. Folks, join us next time for the next episode of After Earnings. Wow. That was CEO and co-founder of Upstart, Dave Gerard. If you enjoy that episode, don't forget to like and subscribe to our social media and podcast distribution platform locations. And join us next time here on After Earnings.
