Upstart has been a pioneer in leveraging artificial intelligence in underwriting. They were talking about AI before any other online lender and it has been a core part of their underwriting from day one. Today, they announced a new offering called Credit Decision API. We had the chance to speak with Paul Gu, co-founder and head of product and Alex Rouse who works in product management at Upstart to discuss their new offering.
It’s important to understand that all of Upstart’s offerings for financial institutions run on the same core technology and model that powers Upstart’s own loans. Upstart has also offered a white label solution for banks for some time now which is a true end-to-end solution. The addition of the Credit Decision API will now allow banks to access just one piece of the end to end solution which is the credit decisioning. The benefit is that banks are able to plug into the API and integrate Upstart’s model into their current workflows.
This is targeted towards mid-sized and large banks who either just want the underwriting expertise or are simply not ready for a full end to end lending solution. Upstart’s model has been proven to provide great outcomes for their lenders and now more financial institutions will be able to take advantage of their expertise. The press release has further details on the features.
New Verticals and COVID-19 Performance
Beyond personal loans, Upstart has an auto loan program in development which leverages the same technology of their personal loan models. It also adds in pieces of information that are more relevant for this specific asset class. In addition to personal loans and auto, Upstart also has an internal model for student loans.
We couldn’t speak with the team at Upstart without asking them about the current lending environment amid COVID-19. While Paul Gu was quick to say that they can’t comment on how their models stack up directly to their competitors since they don’t have that data, he was able to comment on performance more broadly. dv01, a company that acts as a hub between lenders and the capital markets has recently shared multiple reports, rolling up the data from the leading online lenders.
Gu stated that based on the dv01 numbers they are doing far better than what the rest of the industry is seeing. During a normal lending environment Gu shared that they see loss rates at about half of industry benchmarks. During the crisis their model seems to be holding up even better. The lenders collectively, as part of dv01’s reported data, have hardships at around 10% of the loan book at the end of April. Upstart came in at 4% or less than half, meaning that the normal 50% delta is even more pronounced in this current crisis. The team feels like their model, which includes 1,500 traditional and non traditional data points is responding well to the new lending environment. Their model is maintaining the same risk separation between low and high risk populations and for new originations they believe the model is even more responsive.
While Upstart is not in the business of predicting the future they have been making upgrades to their infrastructure to take into account macroeconomic trends like unemployment claims in real-time. Their model is designed to be agnostic to economic cycles.
We also asked the team why they decided to release this new offering now. Firstly the company has been working on Credit Decision API for some time with some clients already in beta. Lenders are now more conservative and are likely to be in this state for some time. Thus there is a lot of more interest in risk management and their models have demonstrated the ability to separate risk. The alternative for some of these banks would be to just lend to no one.
They expect more banks to be on-boarded throughout the summer with this new offering which is priced on a per applicant basis. Going forward we may see other parts of Upstart’s end to end solution broken out into various APIs so it will be interesting to see how Upstart’s offering evolves over time.