Editor’s note: Upstart Co-Founder and CEO Dave Girouard shared thoughts on second-quarter 2024 results during the company’s quarterly earnings call. For more, please see Upstart’s Q2 2024 earnings press release and the earnings presentation.
Good afternoon, everyone. I’m Dave Girouard, co-founder and CEO of Upstart. Thanks for joining us on our earnings call, covering our second-quarter 2024 results.
I’ve said many times over the last couple of years that I’ve never lost an ounce of faith or optimism in the future of Upstart. And today, you can begin to see why. I’m proud and thankful for the incredible work done by Upstarters in the last two years to build a stronger and better company on so many dimensions.
The numbers and guidance we released today demonstrate that we’re turning a corner. We’ve made real progress toward returning to sequential growth and EBITDA profitability. And, I believe, toward resuming our role—once again—as the Fintech known for high growth and healthy margins. We’ve also rebuilt our funding supply by locking in important long-term funding partnerships and significantly reducing the use of our balance sheet to fund loans. We expect this trend of reduced loan funding from our balance sheet will continue through the remainder of 2024.
But this progress is not due to any dramatic improvements in macroeconomic factors or risk; any such macro wins remain in our future. Rather, our progress is the result of the dedicated efforts of more than 1,200 Upstarters. The improvements that are evident in our business today are coming from inside the house: first, significant and even dramatic AI model wins; second, a revamped and revitalized funding supply; and third, increased operating efficiency. These wins and more are providing the foundation for the Upstart comeback story that I expect we’ll share with you in the quarters and years to come.
Today I’ll provide some insight to these major initiatives and how they’re building on the progress we’ve made in recent months.
Product Progress
We continue to focus the majority of our efforts on our core personal loan product, where the opportunity for leadership in a fast-growing category is clear. Our product today is far superior to what we offered two years ago in all the dimensions that matter. Model accuracy, fraud detection, automation, funding resiliency, acquisition costs, and revenue optimization are leaps and bounds better than they were in 2022.
Most importantly, I’m thrilled to share that we very recently launched one of the largest and most impactful improvements to our core credit pricing model in our history. In fact, with this launch, 18% of all accuracy gains in this model since our inception have been delivered by our ML team in the last 12 months.
To dive a bit further, Model 18, or M18 as we call it internally, is the first to incorporate APR as a feature, or as an input to the model. It’s, of course, common to think of APR as an output of a risk model at least indirectly, but we know empirically that the APR also affects the repayment risk of a loan. All else being equal, a higher APR will select for a riskier borrower—a notion known as adverse selection. Conversely, a lower APR will select for a less risky borrower.
If you have a background in computer science or math, you quickly realize that having APR as both an input and output to the same model presents some challenges. Solving this problem requires running our risk models many times in parallel to arrive at the appropriate answer. In fact, M18 generates approximately 1 million predictions for each applicant in order to converge to the correct APR, which is six times the number of predictions of the prior model. And we believe the improvement in accuracy is well worth it. Additionally, I’m very happy to report that we expect M18 to substantially improve our funnel conversion rate.
From a competitive standpoint, I believe that significant technical obstacles such as the one I’ve described here are themselves a clear sign of progress. We’re pushing the boundaries of computing and AI to build more accurate models. And we have seen few signs that peers in the lending space are far enough along the path of AI-based modeling to even encounter these technical challenges.
We also reached another all-time high on automation of our core unsecured loan product, with 91% of loans in Q2 fully automated. As a reminder, this means no documents, no phone calls, no waiting, and no human involvement whatsoever. Two years ago, this number was 73%—and we weren’t sure reaching 90% was even possible. Driving automated approvals up while keeping fraud to minimal levels is an obvious fit for AI, so we would expect Upstart to continue to lead on this front. And automation isn’t just a win for cost and efficiency—it also provides the foundation of a fundamentally better product for the consumer.
Ultimately our strategy is to offer the best rates and best process to all—for every credit product that matters. This means continuing to expand our platform to auto loans, small-dollar relief loans, and home equity lines of credit. And we’re making great strides in each of these products.
Auto
In Q2, our auto team released new underwriting models for both our auto retail and refinance products as well as a new fraud model for auto retail. We’ve now seen multiple months of calibrated loan performance and are growing confident that our loans are performant and increasingly competitive in the market.
In the interest of continuing to move our auto business to profitability, we increased the monthly fee we charge each dealership for the use of our software. Despite this, we believe we’re still quite inexpensive relative to competitive offerings.
We’re also investing heavily in servicing and recovery for auto and saw a 33% improvement in roll rates and a 44% increase in recovery rates in the second quarter alone.
SDL
Our small dollar “relief” product continues to grow rapidly, with 57% sequential growth in the number of loans in the second quarter. Our intention with this product is to expand access to bank quality credit rather than to generate enormous profits. Nonetheless, I’m thrilled to say that in Q2, SDL became our second product to reach break-even economics. We also signed our first warehouse for SDL this past quarter. For the current quarter, we’ve identified opportunities to reduce the variable cost of these loans by more than 40%—which would represent another incredible win and opportunity to increase approval rates further. Overall, this team continues to execute like pros and is helping Upstart expand its impact on the American consumer rapidly and responsibly.
HELOC
As of today, our Home Equity Line of Credit is available in 30 states, covering 51% of the U.S. population. We exited Q2 with an instant approval rate for HELOC applicants of 42%, up from 36% in Q1. This means we’re able to instantly verify applicants’ income and identity without the need for tedious document uploads. Consistent with our experience in personal loans, instantly approved applicants convert almost twice as often as other applicants. With respect to credit performance of our HELOCs, things couldn’t be better. With more than 300 HELOCs originated, we have zero defaults to date.
Finally, we’ve seen significant interest from Upstart’s bank and credit union partners in our HELOC product and hope to launch our first lending partnership before the end of the year.
Servicing
We continue to invest enormously in servicing and collections. To give you a sense of this, in the last two years we’ve tripled the number of Upstarters on our servicing product and engineering teams. And this investment is paying off. We’ve made it radically easier for borrowers to make payments in whatever way works for them. We’ve implemented new channels for reaching borrowers who are delinquent. These efforts and more have helped drive delinquency rates down by 16% year over year and have helped reduce support costs per current loan by 30%. We’ve also now increased the number of borrowers enrolled in autopay for 36 consecutive weeks. Much of our team’s efforts to date have prepared our servicing infrastructure for the deployment of AI models that we believe will enable us to build a significantly differentiated loan-servicing capability.
Funding Supply
Two years ago, we told you that we would upgrade the funding supply on the Upstart platform. We aimed to move a significant portion of our funding from at-will monthly agreements to longer term committed partnerships. Given the importance and complexity of these relationships, we cautioned that this would take some time. I’m pleased to share that we’ve now accomplished this goal.
We ended Q2 with well over half of the institutional funding on our platform coming from committed capital and other co-investment partnerships. We began with the announcement of our first partnership with Castlelake 15 months ago. This partnership has since been renewed. We’ve since added significant partnerships with Ares and Centerbridge. Other institutional investors that have been with us for much longer have also returned to the platform. We continue to pursue additional opportunities to broaden and deepen our funding supply as Upstart returns to growth mode.
We also said back then that we’d use our own balance sheet as a transitional bridge to this better state. You can see from the numbers we released today that we’ve begun to reduce the use of our balance sheet to fund loans. We’re hopeful this will continue through the rest of the year, though I’d like to always reserve the option to use our balance sheet to do the right thing for our business.
I’m also pleased to report that banks and credit unions continue to return to the Upstart platform. We’ve signed eight new lenders since Q1. Performance and lender demand on the platform are creating a competitive environment which is beginning to reduce prices for Upstart borrowers. In fact, lenders representing about half of the monthly available funding on Upstart from lenders have reduced their target returns recently as their liquidity has improved and their demand for loans has increased. This is the first time in two years that we’ve seen loan prices drop on Upstart.
For many reasons, transforming credit with AI is complex and challenging. Tackling the world’s most entrenched problems with AI is difficult—and it doesn’t happen overnight. But to those who ultimately solve these problems, there comes a tremendous reward.
Today we’re tackling problems that we weren’t even aware of a couple years ago. My perspective is that, top to bottom, we’ve gone through a significant reinvention of the company, both from a technology and business model perspective. We’re confident we’re on the right track and making rapid progress. And this is just beginning to show in our financials.
Despite the fact that many trillions of dollars in credit are originated each year, our competition in AI is scarce. In Generative AI, you have a significant number of well-funded and talented competitors, such as OpenAI, Google, Anthropic, and Meta, at the cutting edge of model building. In AI for lending, you have Upstart.
Thank you for joining us. As we’ve discussed today, the actions we’ve taken over the past two years are beginning to pay off, and we’re well set up for the remainder of 2024 and beyond. I hope you all enjoy the rest of your summer and we look forward to speaking with you again in the fall.