Financial regulators’ recent statement cautiously encouraging banks and credit unions to use alternative data in their lending decisions is a big step forward for the use of nontraditional data and artificial intelligence in loan decisions that could significantly expand credit to those who normally don't qualify, according to bankers and industry representatives.
“It's going to give banks that were already piloting or looking at these types of things validation that they're heading in the right direction,” said Marc Butterfield, senior vice president, new product development, innovation and digital strategy for First National Bank of Omaha. “So they're going to keep moving forward with that. And it's going to make banks that weren't planning on using it start looking at it. So it's an announcement that's going to lift all boats.”
The federal banking and credit union regulators
For Butterfield, whose $23 billion-asset bank has been testing the use of alternative data in loan decisions for several months, the news was well-received.
“Having them come out and say that you can do it, but you’ve got to do it a prudent way and be very transparent with where it's sourced from, it's great,” Butterfield said. “They gave some high-level guidance around strong model governance. All the right things were said, and I think it's helped to bring it out of what was largely in the shadows.”
Gilles Gade, CEO of the $2 billion-asset Cross River Bank, said the statement "validates, finally, what has been our strategy for the greater part of the last seven years."
Cross River does not use alternative data in pricing decisions, but does use it to expand the number of customers deemed creditworthy, he said.
Melissa Koide, CEO and director of FinRegLab, a nonprofit innovation center that tests new technologies and data to inform public policy and push the financial sector toward responsibility and inclusion, noted the cautious tone the regulators took.
“They are trying to send an important signal that they are aware that there are certain data that can be quite useful, especially in assessing credit risk for an underserved population,” she said. “But the alternative data label includes a lot. I think they're trying to send a caution to banks and nonbanks: 'Be very attentive to your consumer protection obligation, your safety and soundness needs, and help consumers understand this new type of data that may be used in credit underwriting.' ”
What data is 'alternative'
For years, lenders have ascertained borrowers’ creditworthiness by analyzing a combination of FICO score, credit history and debt-to-income ratio.
Consumers who are young or new to this country tend to have a low or no credit score because they haven’t taken out many loans yet. Others may have a low score due to a temporary setback that disqualifies them for a loan.
So-called alternative data represents additional data points that can be used to help paint a clearer picture of a person or a business’s behavior, including cash-flow data — the money flowing in and out of a potential borrower's bank accounts.
FinRegLab analyzed the use of cash-flow transaction data in loan decisions made by five online lenders and found that such data can be helpful in assessing risk.
Koide said there is reason for regulators' caution.
“There is still a lot to be attentive to: How much does the consumer understand? What data are being used? How are you making sure that the data are sufficiently protected and secure?” she said.
Cash-flow data has been used by human loan officers for small-business underwriting for decades.
Today, "the flow of the data can be automated so that it is coming in much more easily and quickly,” Koide said. “And we also now see payment processors able to glean really helpful insights about the small business in this electronic way and use that type of data for the small-business evaluation for a credit.”
Some types of alternative data can help decipher a customer's intent. For instance, First National Bank of Omaha and many online lenders gather data on how a potential borrower fills out the loan application. If someone spent 30 seconds completing an application that takes most people several minutes, the applicant might be a bot or copying and pasting synthetic identity information.
“There's this whole ocean of alternative data about where did that ‘customer’ come from?” Butterfield said. “Is this really a person? What site did they come from? Did they come directly to the application from the bank’s website or did they come from an IP address that we don't know about? Those are all alternative data points that help understand that willingness to repay a loan and original intent.”
W-2s and employment data could help a lender better predict how a borrower will repay. Utility and phone bill data can show a person's ability to pay their monthly bills.
Other alternative data types are more creative. A lender might collect an Uber driver’s driving record and the number of stars they’re getting in the Uber app. Or for an Amazon shopper, they might gather data on how often the person uses Amazon and whether or not their buying seems lavish.
Why it’s controversial
Regulators, consumer advocates, lawmakers and others have questioned the use of alternative data in the past, arguing that the outcomes are unknown.
For instance, could considering college education data cause roadblocks for people who for socioeconomic reasons can’t afford higher education? Could the use of employment data favor people in certain professions?
There’s the concern that alternative data could be used by risk-averse organizations to decline people.
And the use of new kinds of information may trigger privacy and security concerns.
The bringing in of new information “means we have to all lift our heads up and say, how does this work in light of existing rules and regulations?” Koide said. “And then how are we also making sure that the customer — a consumer or a small business — really understands how this type of information about them may be used?”
Guardrails
In their statement, the regulators said banks need to have tools and policies in place to make sure the use of new data streams won’t violate consumer protection and fair-lending laws.
Cross River has had many conversations about this with its regulator, the Federal Deposit Insurance Corp., which has insisted the bank have the right guardrails in place.
“We've followed the various guidances to make sure of the integrity of the lending process,” Gade said. For instance, the bank conducts fair-lending checks “constantly,” he said. “The idea is to be mindful that this is something that's new and needs to be monitored all the time. That's the role we play as the bank on the back end.”
The Fort Lee, N.J., bank partners with several fintechs including Affirm, Circle, Coinbase, Stripe, Upstart and TransferWise.
Gade thinks of his bank as an industry sandbox.
“We've been prepared to take on that risk and to help our partners to venture into new territories, provided that they follow our very strict compliance guidelines," he said.
The regulator has asked the bank to have “perfect knowledge” of its fintech partners’ compliance programs.
The bank complies with FDIC's financial institution letters about third-party loan origination and third party businesses in general. (Its latest letter points to the financial regulators' statement on alternative data.)
The FDIC expects the bank to have “impeccable compliance management,” Gade said. “It's fine and dandy to have alternative data access or utilization, provided that we preserve the integrity of the lending process and provided that we are not discriminating against any class. So we have to constantly verify that there is integrity in the data.”
The goal: financial inclusion
Alternative data can help fill in what’s missing from traditional underwriting data.
“FICO is extremely imperfect in determining risk,” Gade said. “First of all, a lot of people have access to your Social Security number. Secondly, FICO is a very poor predictor of what's happening in the future. Let's say you have a 725 FICO score, but you're coming from 780. That's a negative trend, but the algorithm is not going to pick that up.”
Alternative data can help lenders extend credit to people who have lost their job or had a medical emergency, Gade noted.
“The alternative data will help reintegrate that individual,” he said. “It's a big positive because it broadens the credit spectrum as opposed to diminishing it.”
Butterfield said the new data should be used for financial inclusion.
“If you've tried all of your traditional methods and you can't approve an individual, what additional data can you have to provide a more complete picture on their financial well-being?” he said. “That’s the type of attitude banks need to have.”
But even with pure motives, there’s still a need for model governance testing to ensure the data is inclusive so a lender won’t unintentionally look at protected classes differently.
“One of my favorite sayings is, 'The road to hell is paved with good intentions,' ” Butterfield said. “You can have that right mindset and if you don't have rigor and testing, then you could find yourself in a really bad spot.”
