Paul Gu on How AI is Making Lending Smarter

What’s one of the biggest problems with consumer credit in the United States? 

“An incredible amount of error”, says Paul Gu, Co-founder and Chief Technology Officer at lending company Upstart [UPST]. With traditional credit metrics, “only about 50% of Americans are considered prime credit.” However, “something like 80% of Americans, if you actually were to do kind of a pure back-test, actually repay their loans on time with no problems,” Gu explains. “There’s a whole other third of the country that would have been able to pay you back, but you would erroneously disqualify.”

Upstart has aimed to address this margin of error from the get-go. 

Before founding Upstart with CEO Dave Girouard and Anna Counselman, Paul Gu studied computer science and economics at Yale University, participated in the Thiel 20 Under 20 business incubation fellowship, and worked as an analyst at hedge fund DE Shaw. During his time as an analyst, Gu noticed how much time, effort and talent was being dedicated to arbitraging securities prices — a relatively narrow field. When looking to start a business, he wondered “if you could bring the same amount of intelligence to bear on a problem that would actually affect people’s lives in some real way.” 

With Upstart, he says, he settled on a pervasive problem: “access to money”. By looking for a smarter approach to lending, Upstart aims to “make it so that everyone can access consumer credit and access it instantly, easily and ultimately with a good rate.”

To achieve this, Upstart employs a range of artificial intelligence (AI) tools and processes. “We see ourselves first and foremost as a technology company,” Gu says. “But of course, we actually are also importantly a consumer-facing company.”

In the latest episode of OPTO Sessions, Gu sits down with us to explain how the company balances customer and regulatory needs with machine-learning efficiency gains, and discusses how AI could change the consumer loans market. 

Product Portfolio

In addition to personal loans, Upstart offers auto refinancing loans, auto purchase loans and home equity line of credit, or HELOC. 

Upstart also works with other companies to deploy its automated loan solutions. “A bank can come to us and work with our technology and offer all of those benefits to their customers and to future ones,” Gu explains. The company also works with a number of auto dealerships, offering its products to consumers looking for auto financing. “We’ve built out our own lending platform that dealerships use.”

Upstart has improved on the traditional loan approval process by looking at non-traditional variables such as education and employment history to predict creditworthiness. This creates what Gu calls “better separation … there will always be some people that you didn’t think would pay back that pay back. And there’s going to be some people that you thought would pay back that don’t pay you back. There’s errors in both directions.”

Essentially, by improving separation — determining who will pay back the loan and who will not — “you can get dramatically higher approval rates and lower APRs,” ultimately lowering the cost of credit. Additionally, from the customer’s perspective, one of the biggest gains from automating the approval process is time. “When they need credit, people aren’t generally looking to spend days or weeks filling out paperwork,” Gu says.

Critics may point to the potential limitations for using AI to differentiate between similar lenders. For that, Gu has a punchy response: “Only humans are capable of treating two similar people different ways. Computers are really good at just following the data. If the data says two people are the same, then the model is going to treat them the same.”

This, Gu says, is a strength of using AI, not a weakness. The company has invested significant time and capital into determining the factors that indicate an individual’s ability to repay a loan, and how to measure them using AI models. Ultimately, he says, “the dimension of similarity that really matters is willingness and ability to repay a loan … if two people are both similarly able to repay a loan and you treated them differently, then there’s something wrong.”

Improving Performance

Perhaps one of the most exciting prospects for Upstart is expanding its total addressable market, Gu notes. The company has built its model on unsecured consumer loans, which, in the grand scheme of consumer credit, is a small enough part of the market to constitute “a rounding error”.

However, Gu expects the company’s model to scale as it expands into the comparatively larger spaces of auto and home loans. “They’ve got all the same problems, where you either need to apply AI to get wins in credit decisions or you need to apply it to get wins in verification and automating the process.”

Within the wider financial space, however, the potential is even greater, Gu believes. “Lending is an enormous part of the market,” he says. “We think it’s in some sense the entire source of profits for the whole financial sector.”

In a company as reliant on technology as Upstart, the devil is in the details. “The single biggest source of growth for the company comes from wins in AI, from improvement to the meta model architecture,” Gu explains. 

Those improvements have paid off, however. In Q2 2025, the company originated 372,599 loans, up 159% year-over-year, with a 23.9% conversion rate. Total revenue of $257m, up 102% year-over-year, allowed the company to achieve GAAP profitability a quarter sooner than expected. 

As investors are increasingly worried about macroeconomic factors, from tariffs to stock market bubbles, Upstart has fine-tuned its models to respond accordingly. Rather than try to predict “black swan” events, like the Covid-19 pandemic, the company has focused on “how we can make it so that we are just the fastest to respond and the most precise in responding.” This way, even with macro headwinds, the company can continue to screen and track for delinquency and respond accordingly. 

Eventually, Gu believes that even traditional creditors will have to face the music. “No matter what you think about AI, whether you’re an enthusiast or not, you’re just going to have to look at the numbers and come to terms with the reality that this is the new way lending is done.”

What end result does Gu see from this process? “You’re going to asymptotically approach the real levels of access to credit that should exist in this country,” he says. “Something like 80% of Americans should have access to bank-quality credit. And the other 20% should have a clear and defined path of how to get from where they are to where they want to be.”