Mining Data to Uncover Worthy Borrowers Shut Out by Credit Scores
For decades, the arbiters of creditworthiness have been two powerful groups: the Big Three credit bureaus, which keep files on roughly 200 million consumers, and score creators like FICO, which turn that raw data into a three-digit key to credit cards, car loans, mortgages and more.
But with tens of millions of consumers left out of traditional credit scoring and the pandemic exposing potential problems in the current system, established players and slick start-ups alike are collecting and crunching all manner of other data to determine who ought to get a loan and how much they should pay.
This so-called alternative credit scoring could have profound effects for consumers, many of them minorities or low-income individuals, who can be asked to hand over more intimate personal information — like their spending habits and details of their college degree — in hopes of getting a loan.
“The box for who gets a conventional credit score is pretty small, and that box hasn’t been updated in a while,” said Silvio Tavares, the chief executive officer of VantageScore, an established credit scorer that is owned by the big bureaus and is working on adding alternative data to its models. “Data is really a big equalizer.”
The efforts to better understand potential borrowers increasingly take two forms, which sometimes overlap. The first involves obtaining cash flow and transaction data from users’ bank accounts, a practice that lenders including Kabbage have used. The second involves applying artificial intelligence to broad swaths of information — which may include items already in your credit report, new details such as the mileage on the used car you’re buying or perhaps behavior gleaned from your debit accounts — to assess applicants’ ability to pay.
Regulators have recently begun discussing both issues, considering and collecting input from the financial industry and others. Officials at the Consumer Financial Protection Bureau have warned that artificial intelligence could amplify risks, including by perpetuating biases against certain borrowers, charging some of them too much or simply making inaccurate predictions. The bureau’s director, Rohit Chopra, recently said the new algorithms became “black boxes behind brick walls” when left unchecked.
A deeper understanding of potential borrowers’ finances is valuable intelligence for lenders. The roughly 45 million people who have a thin or nonexistent credit history — more than 15 percent of the country’s adult population — are a lucrative untapped market.
“FICO is more than 30 years old,” said Dave Girouard, chief executive of Upstart, which uses nonfinancial data including the type of job you hold and your level of education to help make credit decisions on personal and auto loans. “It leaves millions of people out in the cold and millions more who pay more for credit than they should.”
Upstart’s platform is growing rapidly. It has more than 30 lending partners, including Cross River Bank, which made more than 360,000 new loans totaling $3.13 billion in the third quarter, up 244 percent from a year earlier. At least four of those lenders dropped their minimum FICO score requirement altogether.
The company is also something of a regulatory guinea pig: Upstart was the first business to receive a no-action letter from the Consumer Financial Protection Bureau. The letter essentially said the bureau had no plans to take any regulatory action against the company in return for detailed information about its loans and operations.
Though the bureau didn’t recreate Upstart’s results on its own, it said the company had approved 27 percent more applicants than the traditional model, while the average interest rates they paid were 16 percent lower. For example, “near prime” customers with FICO scores from 620 to 660 were approved about twice as frequently, according to company data. Younger and lower-income applicants also fared better.
Upstart, which also agreed to be monitored by two advocacy groups and an independent auditor, takes into account more than 1,000 data points inside and outside a consumer’s credit report. It has tweaked its modeling at times — it no longer uses the average incoming SAT and ACT scores of a borrower’s college — but includes the person’s college, area of study and employment history. (Nurses rank well, for example, because they’re rarely unemployed, Mr. Girouard said.) The amount that borrowers are asking for may also be a factor: If they are seeking more than Upstart’s algorithms believe is appropriate, that may work against them.
Other companies work in a similar way, although the methods and data they use vary.
TomoCredit, for example, will issue a Mastercard credit card to applicants — even those with no credit score — after receiving permission to peer at their financial accounts; it analyzes more than 50,000 data points, such as monthly income and spending patterns, savings accounts and stock portfolios. Within two minutes, consumers are approved for anywhere from $100 to $10,000 in credit, to be paid off weekly. On-time payments help build users’ traditional credit files and scores.
Zest AI, a Los Angeles company that already works with banks, auto lenders and credit unions, is also working with Freddie Mac, which recently began using the company’s tools to evaluate people who may not fit squarely inside traditional scoring models.
Jay Budzik, Zest AI’s chief technology officer, said the company went deep into applicants’ credit reports, and might incorporate information from a loan application, such as the mileage or potential resale value of a used car. It can also look at consumers’ checking accounts.
“How frequently are they getting close to zero?” Mr. Budzik said. “Those things are helpful in creating an additional data point on a consumer that is not in the credit report.”
The same methods can also be applied to those who already have a robust credit history, filling out their profiles in real time. Such information became more valuable during the pandemic because credit scores alone may not have picked up signs of stress when borrowers could pause payments on student loans and mortgages.
It can take months for some information to filter into credit scores, said Kelly Thompson Cochran, deputy director of FinRegLab, a nonprofit that tests new technologies in the financial industry. “This can make it particularly difficult for lenders to predict default risk accurately both for applicants who have recently experienced financial difficulties and for applicants who are rebounding from past income or expense shocks,” she said.
Established credit scoring and reporting companies are increasingly offering consumers ways to add additional information. The credit bureau Experian’s Boost feature allows consumers to pipe in payments on bills from a services like Netflix, Disney+ and their mobile phone provider. The average customer’s FICO 8 score — the formula currently used by most lenders — rises 13 points, executives said.
And FICO is piloting a new score, UltraFICO, which augments its traditional model by taking into account — with users’ permission — their cash on hand, history of positive balances, and recentness and frequency of banking transactions. FICO estimated the new score can reach 15 million more people.
More information, such as income data or whether you have a 401(k) plan, could be included in future iterations, said FICO’s chief executive, Will Lansing. “I think the future of the industry is the consumer taking more control of their data,” he said, “and deciding when it will be used and what it will be used for and for what purpose.”
Consumer advocates say that’s a crucial issue.
While the growing use of transaction data could be a boon to many borrowers, checking and debit accounts contain all sorts of revelatory information, and access to it must remain voluntary, advocates said. Lenders may be looking largely at the broad strokes of your cash flow now, but will they eventually glimpse at where you shop and what types of doctors you visit?
“Credit invisibility is a problem, but some of the solutions or cures can be worse than the disease,” said Chi Chi Wu, a staff attorney at the National Consumer Law Center. “It’s a high-wire act to make sure this helps more than it hurts.”
Tara Siegel Bernard covers personal finance. Before joining The Times in 2008, she was deputy managing editor at FiLife, a personal finance website, and an editor at CNBC. She also worked at Dow Jones and contributed regularly to The Wall Street Journal. @tarasbernard