Thanks for posting the link to this interview, HH. It was informative and interesting for this UPST investor. The guy answering questions about Upstart was Sanjay Datta, CFO. (Is that last name not perfect for a guy working for Upstart? Heh)
A few highlights:
A major reason they have a borrower facing front end, instead of just providing an interface to their AI model to lenders, is so they can control exactly what is asked of the borrower to get maximum data useful to the AI model, which maximizes the accuracy of their risk rating of the borrower. He mentioned an interesting tidbit that they ask borrowers if they know what their FICO score is, and for borrowers that might be a bit iffy, if they do pay attention to such things and know it, they are a better risk.
UPST is good at identifying the risk profile of a borrower. That’s what gives them their edge that the lenders pay for. They give a much more accurate and nuanced view of that than FICO. The macro environment is not directly related to that: they leave it up to the lenders to judge whether they need to charge a higher rate because a recession may be coming and defaults would rise as people lose their jobs. Upstart doesn’t see the current levels of inflation as a significant default risk by itself, but they definitely acknowledge that when people start losing jobs there’s a huge direct effect on defaulting on loans.
For the vast majority of their time in business, the environment has been one where there was more financing available than borrowers to soak it up, so a focus has always been to attract borrowers to the ecosystem. A few times that has reversed, once back in 2016 that he didn’t elaborate on, then the first couple months of the pandemic when we were in lockdown but didn’t have stimulus money yet, and now, when stimulus money has worn off: been spent, lost in the stock market decline, or whatever, so demand for loans has risen, but with the macro uncertainty of the high inflation, the war in Ukraine, the possibility of impending recession, financing has somewhat dried up.
Hence, UPST used their own balance sheet to fund some significant number of loans in Q1. They figured it was a reasonable thing to do, they had the money, and they could sell the loans on the secondary market down the line if they wanted. The market reacted very negatively, as we investors experienced, OUCH (but gave a great purchase price for one chunk of shares), so UPST now accepts that judgement: it wasn’t important enough for them to fight the market on it, so they will only be using balance sheet to fund R&D loans in the future, like they are for auto loans at present, and they will just accept volatility in loan volume as the nature of their business.
UPST makes their money by charging a transactional fee on each loan they facilitate. In the past they didn’t maximize their take rate, instead preferring higher volume to help train the AI and to build customer base who might be repeat customers. Now, in a tougher environment, they are maximizing that take rate more to shore up the profit and loss flows.
They have worked closely with the Consumer Finance Protection Bureau (CFPB) ever since the Obama administration, and one of the important things they’ve focused on is explainability of their AI’s decisions. They have roughly as many AI scientists working on this problem as they do on the actual main AI that runs the Upstart risk model. They say it’s computationally intensive and hard to do, but that they believe it’s vitally important that the AI isn’t just a black box spitting out unexplained decisions. CFPB agrees and is apparently going to be making that a requirement, which makes UPST’s moat a bit bigger: a competitor can’t just train up an AI to make the decisions: they have to do the hard work to be able to document WHY it’s making those decisions.
Those were the highlights I got out of it. Hope this was helpful to some of y’all out there. I’m still impressed with this business and their monster growth potential, though we’ll undoubtedly see a lull in the short term.