Dave Girouard, CEO and founder of Upstart, testified before Congress’ Task Force on Financial Technology for its hearing on the use of alternative data in loan underwriting and credit scoring.
Girouard spent eight years as the president of Google Enterprise, where he presided over Gmail and Google Docs. In 2012, he founded Upstart, a company that uses technology to improve loan underwriting. The company has originated $3.8 billion in loans since its founding and has partnered with BankMobile, First National Bank of Omaha and Accion Chicago.
Bank Innovation spoke with Girouard about his testimony to Congress and his thoughts on bias in automated loan underwriting. An edited version of that conversation follows:
Why do you think Congress wanted to hear from you about alternative data in lending and credit scores, and what did they ask you?
When you are using alternative data, there are natural questions around the potential for bias. As Congress is evaluating these emerging technologies, I think they just wanted to hear from somebody who is actually out there doing it.
You have five minutes to make your initial statement, and the remainder was a couple hours worth of questions, not just for me but for other witnesses who spoke. They were asking what the process was like with the Consumer Financial Protection Bureau. Is that a scalable approach for other types of institutions? I referred to a sandbox as something that could be helpful to make this technology available.
Many people worry that integrating technology into loan underwriting creates bias against certain ethnicities and demographics. How does Upstart avoid this problem?
I think the fear is well founded, as it has the potential to cause bias. Ultimately, it comes down to measuring the outcome. What we’ve been able to demonstrate is that our use of alternative data and our use of artificial intelligence is not causing bias. We’ve gone to the right regulators to verify that. It’s a fair concern, but ultimately the answer is not to debate it ad infinitum. The answer is to measure it and determine the truth.
But there’s research out there demonstrating bias in artificial intelligence. Where is that bias coming from? The last thing I want you to get from me is, “Dave thinks AI is not biased.” It certainly has the potential to be, but the bias is inherited from humans. The bias inherits from the training data that it learns from, so it’s absolutely possible. Our view is you would measure the outcome. You should supervise the system and make sure that it’s not introducing bias.
What kind of alternative data does Upstart use to underwrite its loans?
We use more than 1,000 variables, the majority of which come from credit bureaus. However, we supplement that with a lot of alternative data, such as work history, educational background, how they found us, what size loan they’re requesting and their awareness of their credit score. That creates a very unique credit model that allows us to approve a lot more people at lower interest rates. FICO is extremely limited and backward-looking; it does not identify millions of creditworthy people.
A lot of companies are using technology to improve the loan underwriting process. How is Upstart going to differentiate itself?
I think we have a headstart because we began doing this six-plus years ago. People at Goldman Sachs Marcus claim to be building similar technologies to us. I’m not positioned to say whether they are or they aren’t, but we expect a large number of banks to either develop this technology themselves or look to companies like Upstart for a partnership.
Fintechs are constantly rolling out new products to create lifelong customers and become more bank-like. What is the end goal for Upstart?
We are going to be a partner to banks. Our model is not to become a bank; we don’t have an interest in that. We want to be a technology company that helps banks succeed in this technology transition.