Banking

U.S. Bank, Suncoast say AI helps take loans from ‘no’ to ‘yes’


Front door of U.S. Bank branch

U.S. Bank has made loans to 2,000 customers who would not qualify under traditional underwriting methods, through a partnership with fintech Pagaya.

Banks have become more willing to lend to people who don’t meet traditional underwriting criteria, with the help of so-called alternative data and artificial intelligence.

Two cases in point: U.S. Bank, which revealed today that it has made 2,000 such loans in a matter of months with fintech partner Pagaya, and Suncoast Credit Union, which has been working with Zest AI to lend to underserved communities.

The motivations: inclusion and efficiency

At U.S. Bank in Minneapolis, which has $587 billion in assets, the aim is to deepen relationships with existing clients by giving them unsecured personal loans. 

“We continue to see that there is a segment that we have not been able to serve through traditional credit underwriting,” said Mike Shepard, head of consumer lending partnerships at U.S. Bank, in an interview. “We don’t want to force that consumer to have to go outside of our four walls to get that. They trust us with their deposit account; they trust us with other solutions. How do we then continue to be on that journey with them through their financial life?”

These borrowers don’t qualify under traditional loan underwriting methods for different reasons, he said. For instance, some have low credit scores. 

The 2,000 customers to whom U.S. Bank has granted loans through the Pagaya partnership are all long-term, established U.S. Bank customers. 

“They have multiple products with us, so they trust us with those, and now we’ve been able to deliver the ‘yes’ message as opposed to a ‘no’ when they came to us looking for a personal loan,” Shepard said.

Among all borrowers for whom Pagaya facilitates loans for its 29 bank partners, 50% have a credit score above 660, according to Leslie Gillin, chief growth officer at Pagaya. About half have low to moderate income and half are women. About 35% are Black and Latinx. 

“All banks have that mandate to bring more Black and Latinx consumers into the mainstream economy and make sure that more deserving consumers are able to get credit access,” Gillin said. 

Suncoast Federal Credit Union, a community development financial institution based in Tampa, Florida, with $10 billion of assets, has a diverse membership, especially for the branches it has deliberately placed in neighborhoods in underbanked communities, where it’s understood they are not going to be profitable. Some of these potential borrowers have no FICO score.

“Somebody has to be their first loan,” said Darlene Johnson, executive vice president at Suncoast, in an interview. “We want to give these new borrowers with zero credit score an opportunity to borrow responsibly and from a responsible lender. The past, in my opinion, doesn’t predict the future and bad things happen to good people. We have to give people an opportunity to rebound. If I made a mistake five years ago because I had a medical condition, or I’m going through a divorce, or the great recession hits and I have to change professions and I’m out of work for six months, a FICO score doesn’t rebound. It doesn’t really indicate whether or not that member has the ability and willingness to repay it in the future. So we have to have a more humane way of trying to get consumers the goods that they need through our lending channels.”

The bank deployed Zest AI software a couple of years ago. It takes the more basic decisions away from the human loan analysts, giving them more time in the day to concentrate on the most challenging applications, looking for opportunities to serve members and offer coaching and counseling, Johnson said.

The loans start as small as $50. The credit union provides financial counseling and makes sure borrowers understand what it means if they don’t repay.

Johnson was also looking for greater efficiency when she began using Zest AI’s software. 

“We’re always looking to increase capacity for our team and provide better service to our members,” Johnson said. “The one key factor around accomplishing all three of those is to find innovative, robust technology solutions that can help us meet the mark on those. That’s where we started: How do we make faster loan decisions? How do we make them more efficiently?”

She looked for software that could “think like a human, to ensure that we were giving our members credit for how long they’ve been with the credit union, their repayment of loans.” She also wanted software that could bring more consistency with loan decisions made across 300 lenders and 78 branches.

“We do training, we obviously have procedures in place and we train loan analysts on how to make fair and equitable decisions,” Johnson said. “But you’re talking about 300 people, 300 diverse ways of thinking. So bringing in the automation that supported consistency and fairness was at the top of our list.”

How AI helps get to a “yes”

Pagaya operates a two-sided network. On one side are lenders and on the other are institutional investors. Through a partnership with TransUnion, Pagaya brings additional credit variables to bear on loan decisions that aren’t part of a traditional credit score or underwriting model. 

Using AI lets Pagaya bring more data to the decision.

When U.S. Bank runs a loan application through its usual underwriting model and gets a decline, it immediately and automatically sends the application to Pagaya, which runs it through its AI-based model. 

“Where U.S. Bank is leveraging their proprietary scores and FICO and the information on the application, we use hundreds of additional variables in credit data,” Gillin said. “There’s no alternative data. It’s all credit bureau data, but all the variables are being leveraged in production.”

For a borrower whose debt-to-income ratio is higher than U.S. Bank’s threshold, Pagaya’s data might show 10 years’ worth of pristine credit history.

“We’re able to bring them through because we can see that holistic view,” Gillin said. “We’re able to, in milliseconds, come back to U.S. Bank with a ‘yes’ versus a ‘no’ decision, on average about 25% of the time.”

The AI-based model uses U.S. Bank’s credit policy and pricing. It’s not apparent to the customer that anything has changed.

“Because it’s the same product, the same terms, the same pricing parameters, there’s not a waterfall approach where you’re pushing somebody to somebody else with new terms and conditions, a new product, new pricing,” Gillin said. 

The loans are in the same range of $10,000 to $15,000 of the bank’s regular personal loan business. They’re used in the same way – for debt consolidation, home improvements and such, Shepard said. Interest rates are a couple of hundred basis points higher than the bank’s more typical loans, in keeping with the higher risk.

The loans are sold to investors, so U.S. Bank doesn’t take on incremental credit risk.

Investors on Pagaya’s network are mainly looking for return on assets, Gillin said.

U.S. Bank services the loan and monitors its performance, which lets the bank learn and enhance its own models over time. 

Where traditional credit models are built using logistic regression, which can accommodate 15 to 20 variables, a machine learning model can analyze 500 data points, Zest AI CEO Mike de Vere said. 

At Suncoast Federal Credit Union, Johnson points out that there’s a big difference between a 620 FICO score that’s on its way up and a 620 that’s on its way down.

“If I just look at a score, I don’t understand that,” she said. “I have to go beyond just the score and then I have to consider, why is it the way it is and what is their condition today?”

The credit union is considering using credit bureau data on cell phone bills, rent and utility payments in loan decisions. 

Defying the “black box” criticisms of AI-based lending

Bank regulators, including Rohit Chopra, director of the Consumer Financial Protection Bureau, have repeatedly warned that when banks use AI in their lending decisions, the models can’t be a “black box;” they must be explainable, transparent, fair and free of bias.

U.S. Bank applies model risk governance, monitoring and tracking to its own and its partners’ models, Shepard said 

Pagaya’s models have to pass the same model validation that banks’ internal models would have to pass, Gillin said. 

“We like to think of ourselves as a glass box instead of a black box,” she said. “Our model goes through the same fair lending testing that U.S. Bank’s models do. So there’s a lot of transparency.”

Banks and fintechs that use AI in lending decisions say their models are more transparent, explainable and fair than traditional models and FICO scores.

Suncoast, which is a state-chartered credit union, has been examined several times by Florida’s Office of Financial Regulation since it deployed the Zest software. 

“We have the reporting, the monitoring, the tracking, we have it outlined in terms of how we address fair lending and we’re able to provide whatever documentation our examiners look for, and they have not taken issue with it,” Johnson said.



Source link

Leave a Response