First, a big thank you to GauchoRico for providing the link https://www.youtube.com/watch?v=C2cKcXOwibA, a very informative interview with Paul Gu of Upstart. It is impossible to watch this interview without being extremely impressed with his technical knowledge and innovative thinking. While not being able to compete with the financial knowledge and stock savviness of those here, I’d like to comment on what struck me about the company from a technology perspective along with some personal notes about what I see ahead.
Data
There’s a path from data to information to knowledge with each being an abstraction of the prior. Today’s loan application analysis begins with credit purchase history (data), which is distilled into information (like a FICO score), and then supplemented by loan application info to produce knowledge about the applicant’s loan worthiness. Lenders analyze this knowledge to produce a decision on whether or not the loan should be approved and under what terms. The problem is that the lenders don’t know if the data they’re collecting is actually the most useful predictors of risk; they use the traditional parameters, set clip levels and conditions on various things, and then make a subjective judgement on how/if to proceed with the loan.
In the interview Paul said something to the effect that they have 1,000 data points on each customer record. You’d think that’d delay the decision process, but it doesn’t – it’s lightning fast. My suspicion is that the application contains only the parameters necessary for UPST to apply the scoring algorithm and that the remaining data points are added later and cleansed so that the model can be improved over time. This allows for minimal upfront data collection and speed in scoring. (At least that’s how I’d do it.)
As an aside, I really liked Paul mentioning an “efficient frontier” in terms of data – that you don’t want to gather ALL data but focus on getting the RIGHT data.
Datamining
I really, really liked Paul’s discussion of the company’s approach and he’s spot on in saying the field of loan processing was over ripe for just the kind of innovations UPST is bringing. I’m not current in the field as it applies to lending, but it sure sounds to me it’s really a disruptive technology.
Growth Potential
Because of its approach, UPST is not only going to get a sizeable piece of the pie, but also grow the pie of potential customers. Because of the company’s predictive model, it will qualify people who previously may have been rejected and lenders will LOVE the ability to decrease risk while also growing the business – what’s not to like? As word of success spreads, the need for a competitive advantage (or just to keep up!) may lead to dramatic growth. [Look out, FICO, somebody’s coming after you!]
Technical Moat
Nothing UPST does can’t be replicated, but the same could be said of AMZN. Certainly, other companies could purchase the same data, run some off-the-shelf datamining product, and produce their own less accurate but “good enough” models. What’s to prevent that from happening?
Setting up and executing a datamining operation of this scale is a pretty daunting task so among the only lenders which I think would dare tackle this would be major banks like BOA. Even so, I think that’s a small risk as they haven’t even tried to eliminate FICO. Speaking of FICO, they’d face the same challenges along with the very real fear of being rendered obsolete. That’s some powerful motivation.
Looking elsewhere, I can see a tech company trying to mirror what UPST is doing. IBM, Oracle, and others have pretty advanced data mining tools they could apply to create a scoring algorithm and then mimic UPST and/or license the scoring algorithm itself on a subscription basis.
In either case, UPST should be well established with documented (and publicized!) case studies which show their worth. That track record and good pricing should be able to stave off competition.
My Loan
Just for kicks, I used UPST to apply for a fictional $25K auto loan. The application was very simple and the results were lightning fast, but UPST’s focus on assessing individuals means that it doesn’t really address family loans. Example: Alan who’s retired and has a pension income might well be denied a car loan while his wife of 23 years and makes $200K a year would speed through. They need to take care of that. When they do, it’ll open the market even wider.
I bought some UPST this morning and I think I’ll be adding more in the future.