A growing number of fintechs are becoming increasingly reliant on using alternative data like cash-flow analysis as a key underwriting tool, arguing the traditional bank process doesn't cut it anymore.
In theory, analyzing the money flowing in and out of a potential customer's bank account sounds like a natural way to evaluate creditworthiness. But banks have historically avoided that route, preferring to look at credit report data, credit scores and debt-to-income ratios.
But leading nonbank online lenders like Upstart, Kabbage and Betterfin say cash-flow data is critical to their business. Kabbage wouldn’t be able to lend to small businesses without cash-flow data, said Kathryn Petralia, the company’s co-founder and chief operating officer.
“Lenders that use cash-flow data see actual transactions that are happening in their core banking accounts,” she said. “Credit bureaus don't have that information.
There's data supporting her argument. A
Petralia also points to the credit card company Advanta, which in the 2000s was one of the largest small-business credit card lenders in the U.S.
“By 2010 they were out of business,” she said. “And the reason is because they were only using credit bureau data. They were using basically consumer information to underwrite small businesses for credit cards, and they had no idea how those businesses were performing. If they had been able to see cash-flow data, they would have known that things weren't going well.”
BetterFin executives feel similarly.
“Cash flow is objectively the best [data to use in lending decisions] for a lot of reasons,” said Eric Griego, CEO of BetterFin, a lender and provider of data visualization software to small businesses (previously, he was the product manager for OnDeck's data team). “The main reason is because everything quantitatively needed to predict repayment is inherently layered in there. If all you're doing is pulling credit reports for businesses, you're missing out on a huge signal.”
Cash-flow data can let a lender see two full years or more of transactions, he noted. Also, critically, cash flow data helps lenders give people loans they can afford.
Paul Gu, co-founder of Upstart, says cash-flow data is more useful than debt-to-income ratio.
“What we've seen is that that is better than doing nothing, but by itself it’s a pretty weak predictor, because there there's a lot else that's happening besides your debts,” he said. “The big one obviously is your expenses — your rent, for instance. In our borrower population, around 80% of borrowers are renters, so that’s not going to show up as debt in a DTI ratio.”
Then there’s where an applicant lives — a $75,000 income is going to go a lot further in Nebraska than it is in San Francisco or New York. Upstart’s underwriting model takes into account the local cost of living.
“Those are the kinds of things that the traditional DTI measure completely misses,” Gu said. “If you can get more granular data on a person's cash flow, you can improve the predictive power of that measure.”
Cash-flow data can also help verify information about an applicant, including income.
“A person can tell you that they make $75,000 all day long, but you sort of need to know if they actually do,” Gu said. “Cash-flow analysis is a great way to do that in a way that's very non defraudable. It's very easy to upload a fake pay stub. You can go on Google and find a hundred websites that will make you a fake PDF income statement or fake PDF pay stub. But if you actually get access to someone's bank transactions and you're looking at the deposits and withdrawals and a pattern of that across many months, it's much harder for a fraudster to fake that, almost impossible.”
Petralia said this fraud risk is why the bank account data used by a lender has to be received in real time and in an automated way that comes directly from the source.
“Manual processes are rife with fraud, especially for small-business lending,” she said. “Access to data is really important and that's where this whole notion of open banking comes in. Because if I can't get access to the data that businesses have about themselves in their checking accounts, then I can't make the decisions about them and I can't deploy capital safely to them.”
Kabbage uses the data aggregators Yodlee, Plaid and Finicity to pull in bank account data.
“But until we have some sort of structured process in the U.S. that is government mandated, it makes it really hard and it gives both businesses and consumers unfair limitations to access to capital,” Petralia said.
A U.S. open banking law would make lending safer and democratize access to financial services, she said.
Hangup 1: Investors’ demands for traditional underwriting
One hindrance to the use of cash-flow data in lending decisions is that the investors fintechs sell their loans to insist they meet traditional underwriting guidelines.
“Until the alternative nonbank lending community can really convince the capital markets that transaction data is the new genre of underwriting and they can successfully sell these loans, alternative lenders are going to be mostly tied to traditional underwriting using credit reports,” Griego said.
Gu says this is particularly true for startups. Now that Upstart has been lending for almost six years, “we have seen an evolution in the reactions of the investor community,” he said. “I think justifiably when a new lender starts out claiming that they've got a new method, investors are very skeptical and there's no shortage of lenders that have failed. So there’s a very reasonable skepticism and preference for the things that are better understood, like FICO and DTI.”
But investors have come to accept Upstart’s model, according to Gu, because Upstart’s loss rates are about half that of a traditional lender.
Kabbage appends credit bureau or FICO data to loans, though it doesn’t use that data for underwriting. “The purpose of it is more for benchmarking so they can explain what they think that loan looks like to their investors,” Petralia said.
Hangup 2: Regulators
One big reason that banks fear doing anything different is concern regulators will scrutinize or reject something new.
The Consumer Financial Protection Bureau announced several years ago that it was studying the use of alternative data in lending, but it hasn’t come out with any conclusions yet.
The CFPB sent a no-action letter to Upstart a few years ago, allowing it to continue its processes while sending data to the agency to monitor its lending activity. Upstart runs quarterly tests of its loan decisions that show the impact on protected groups across race, age and gender. So far, Gu said, there have been no signs of disparate impact.
Upstart’s models include many types of alternative data besides cash flow. But the company says it tests every data type it uses.
“None of it would be included if it wasn't improving the strength of the overall results,” Gu said.
