What if Upstart is a Good Business?
Good business model, great management, cyclical market.
Everyone knows 2022 was a brutal year for Upstart. Consensus says it’s a sinking ship — a “shitco” in Twitter speak. Revenue is in free fall, and the loan book has grown to over $1 billion despite claims it’s a technology company.
It’s easy to write Upstart off as a product of zero interest rates, a money-losing machine, or simply a lender with a website. That’s the safe option and, judging by the recent stock price trajectory, the correct option.
But let’s step back. Is anyone surprised by the disappointing results of 2022? Did anyone think the 2021 boost would continue in perpetuity? (To be fair, the market certainly did when asking for nearly $400 per share in 2021.)
But let’s be realistic: Upstart wants to disrupt credit. That’s no easy feat, and it certainly wasn’t going to happen in linear fashion. Credit markets will be credit markets, and we just underwent the fastest pace of interest rate hikes in 40 years. Two of the nation’s largest banks went bust in the last week! Lenders adjust to a new environment slowly, and as they do, lending activity is curtailed. Would-be disruptors must accommodate the market — not the other way around.
In the long run, however, the current difficulties won’t matter to the Upstart story if the premise of the company is sound. The premise is simple: (1) Americans are more creditworthy than the current system implies, and (2) AI models improve the accuracy with which lenders can measure default risk.
The most important thing in the long run is model accuracy: the ability to separate high risk borrowers from low risk borrowers. Dave Girouard confirmed this on the Q3 2022 call:
Performance of credit is and always will be our highest priority.
Results in this department have been good (as shown in the chart), but the outcomes are still representative of the risk undertaken by lenders. More accurate underwriting does not eliminate risk — it improves the odds of success.
For loans sold to institutional investors (who underwrite riskier borrowers in exchange for higher potential yields), results were good for a time, but they have suffered recently as higher risk borrowers are disproportionately impacted by stimulus wearing off and higher inflation. From the 10-K:
For loans purchased by institutional investors, all vintages from 2018 through 2020 are forecasted to deliver returns at or in excess of the targets such institutional investors were expecting to receive, while our 2021 through mid-2022 vintages have underperformed relative to target returns. If a loan buyer invested equally in all vintages of loans transacted across our platform since 2018, they would expect a positive return on all vintages thus far.
For loans retained by bank partners (who tend to have more conservative underwriting standards), the results have been excellent. Also per the 10-K:
Since 2018 through mid-2022, all quarterly vintages of loans retained by our lending partners to date are currently forecasted to meet or exceed the target returns set at the time of loan origination.
While the market interprets the underperformance of institutional loans as a signal Upstart’s risk model does not work, I disagree — as does Girouard (Q2 call):
It may be natural for you to question whether Upstart's AI-powered risk models aren't working as designed, but we're confident this isn't the case. That, in fact, our models continue to improve with respect to accuracy and risk separation…
In some ways, the bifurcation of performance between bank partners and institutional investors affirms Upstart’s value proposition. Lenders can customize risk controls and monthly loan volumes just like you or I can adjust the volume on a speaker. In 2021, institutional investors (hedge funds and other leveraged buyers) reached for yield by underwriting riskier borrowers, a fact evident from the dramatic increase in loans funded on the platform.
In doing so, they became less insulated from the impact of a broad deterioration in the macro environment — which is exactly what happened. When the macro picture worsened, default rates jumped across all loans — not just at Upstart — and institutions have been the first to take the hit.
On the other hand, bank lenders (who trade yield for safety) continue to earn consistent (albeit lower) returns while growing loan volumes.
Critically, Upstart is an enabler, not a partaker. Credit markets are cyclical because lenders are apt to loosen standards in good times and tighten them in bad times. Upstart does not encourage nor create this behavior, but it enables those who will do so anyway, to at least do so with better tools / information at their disposal (note all loan vintages still have positive returns overall).
The counter-argument invokes Upstart’s use of its own balance sheet when capital markets dried up over the last year. This is a fair argument (the company is undoubtedly a risky proposition in such early stages), but management has clearly telegraphed their intentions with the balance sheet, and the move does not represent a long-term shift in strategy. Rather, it’s a stopgap intended to keep the marketplace fluid while they lock up longer-term funding agreements. Per the Q2 2022 call:
While we continue to believe that it doesn't make sense for Upstart to become a bank, we've decided it may make sense to, at times, leverage our own balance sheet as a transitional bridge to this committed funding. I acknowledge that this is a shift relative to what we planned and communicated earlier this year, but a changing and volatile environment suggests we need to be flexible and responsive in our approach.
We're taking this step for a few reasons. First, there's an obvious information asymmetry where we understand better than anybody how our model is performing today and how well it's calibrated for the current economic environment. Secondly, we believe the opportunity to generate outsized profits on our platform is unusually high right now. And third, we can bring a level of stability to our business that's important to our longer-term goals while we work to put these committed capital structures in place.
More recently on the Q4 call, Girouard noted the balance sheet had reached the likely upper limit and shared a positive update on the committed capital initiative (though nothing concrete yet):
In our earnings call in August, I told you that we would begin to investigate partnerships that could provide more reliable and persistent funding to the Upstart platform. I'm happy to report that we're in late-stage discussions with multiple potential partners in support of this goal.
If Upstart can bring on a form of through-the-cycle committed capital, it will not only represent a positive for the marketplace but demonstrate the long-term value lenders see despite prevailing uncertainty in markets.
Business Model-Market Fit
Despite Upstart’s trials in 2022, we learned a major point: it has pricing power. Demand for credit is inelastic — people always want to borrow money — but supply, not so much: when the future looks uncertain, banks (and capital markets more broadly) shore up lending activity to protect their downside. That’s exactly what happened over the past year (Q2 call):
Despite the fact that our bank partners have seen consistently strong credit performance, meaning portfolios performing at or above plan across quarterly cohort, several of them have paused or reduced originations due to fear about the future of the economy. To be clear, these lenders and institutional investors have not left Upstart platform, but have temporarily paused or reduced their originations.
The important thing for Upstart is that because borrower demand is inelastic, it can (a) price loans higher and (b) cut performance marketing when the supply of capital is limited. Simply put, it can increase take rates to protect profitability when loan volumes fall. We’ve seen this exact phenomenon in the last year.
Of course, Upstart remains sub-scale, so the improved contribution margin has not been enough to offset the negative operating leverage. But overall, it demonstrates that Upstart’s business model is well-suited to address the volatility of credit markets.
People First
A business with a disruptive mandate is only as good as the people behind it. Fortunately, all three co-founders remain at Upstart more than a decade after its founding, and they retain meaningful “skin in the game” (~19% ownership among all executives and directors as a group). From the Q1 2022 call:
When the economy gets turbulent and nimbleness is at a premium, the advantages of a founder-led company with a closely knit and tenured leadership team become apparent. And that's what you have in Upstart: three founders involved in the business day in and day out and a proven leadership team, half of which have been with Upstart almost since inception.
The team founded the company with full awareness of what market they were entering. A temporary market dislocation will not deter them from pursuing the longer-term opportunity. Also from the Q1 call:
…although we serve a cyclical industry, we represent a secular change that the financial services industry desperately needs. Artificial intelligence will reshape the economics of lending in ways that will reverberate for decades. We're today pursuing opportunities that represent more than $6 trillion in annual origination, so there's little question about the scale of the addressable market. We see a clear path to building a company with more than $10 billion in revenue in the coming years and are maniacally focused on achieving that goal.
While pursuing this goal, the team has allocated capital with skill, managing to take full advantage of the market’s optimism in 2021 and pessimism in 2022. Below is a brief timeline:
December 2020: IPO at $20/share
Raised $167.4 million net of fees
April 2021: follow-on offering of ~2.3 million shares at $120/share
Raised $263.9 million net of fees
August 2021: issued 0.25% convertible notes due 2026
Raised $645.5 million net of fees
Conversion price of ~$285; maximum dilution of ~2.3 million shares
February 2022: authorized buyback program up to $400 million
Bought back 5.9 million shares at average cost of $30.24/share
Total cost of $177.9 million
Essentially, the company raised ~$731 million and retired 1.3 million shares for free (though it will have to pay back the convertibles unless the stock is a 20x prior to 2026). In theory, the market’s irrationality funded Upstart’s loan purchases last year at very low cost to the company.
Prior to going public, the team’s track record was equally impressive. They repeatedly encountered difficulty fundraising and only raised ~$160 million total. Of this amount, they only used ~$70 million prior to going public (and for reference, generated $168 million of operating cash flow in 2021).
Dave Girouard recalls (I recommend the full interview if interested):
In many ways, it ended up being a real strength of the company that we repeatedly raised less money than we intended or expected to. Our thin bank account meant we got good at building the company without throwing money at problems.
This doesn’t sound like a cash-incinerating fintech lending startup to me. And this point is reinforced by another clip from the conversation:
If you're looking for a crazy, visionary founder, I don't think anyone describes me that way. I look like the CEO you bring in when the founder goes off the rails.
If we double-click on Girouard for a moment, his resume is quite impressive. After joining Google in 2004, he built the Google Apps (Gmail, Drive, Calendar, etc.) business from zero to over $1 billion before leaving to found Upstart in 2012. When he left, his role was taken over by Sundar Pichai (current Google CEO). The man is no joke, and he has a record of execution beyond Upstart to back it up.
Conclusion
One question dominates my thinking on Upstart: why would AI not create a significant improvement in the accuracy of credit underwriting models?1
I can’t arrive at a decent answer. Over time, underwriting has become more quantitative and incorporated more data. AI simply takes this to the next playing field, aggregating metric tons of data and using more sophisticated modeling techniques (instead of simply linear regression).
But why is Upstart the answer? Because AI models are ill-suited to in-house development by banks or other lenders. First, individual lenders (other than the JP Morgans of the world) lack the volume of repayment data needed to train an AI model. Second, these firms can’t afford to employ expensive data scientists and machine learning engineers. The solution is a third-party technology plug-in (Upstart) that can aggregate repayment data from many lenders and spread the R&D cost across those same customers.
In closing, I’ll return to the titular question: what if Upstart is a good business?
If Upstart is a good business (if because it has yet to be conclusively proven), I have little doubt investors at today’s prices will earn great returns over time. If we place Upstart in Howard Marks’ risk/reward framework, it is now on the far right: the distribution of outcomes has likely reached its widest point.
Today’s investors could make multiples of their money, but they could also lose their shirt. In the spirit of full disclosure, my money is optimistic, but who knows if I’m any good at investing? (I sure don’t.) Either way, I’ll be around to see what happens on the other side of this credit cycle.
A great quote from Girouard in a similar vein:
When it comes to lending, I think the whole world sort of came to the conclusion that it's kind of a commodity — anybody can lend if you have money and some basic analytics. And if you do it well, you'll make some profits, and if you don't, you probably won't. But for whatever reason, people didn't seem to believe that you could apply modern cloud computing and data science to create a dramatically better product.
People would say, 'Yeah, is your lending model really going to be any better?' It's almost like everything they've learned about computer science and the potential of the software, they threw out the window when they looked at a lending-related business and said, 'Nah, that's just a human instinct business. All the software and data in the world won't make a difference.' I feel like that's the way some in the industry to this day still think about it.