Could machine learning replace the credit score?
Petal has received a $13 million funding round from Valar Ventures, a New York-based venture capital fund that specializes in financial technology, to use artificial intelligence to fill holes in legacy risk vetting.
"The problem is not that people have a history of bad credit, but have no history of credit at all," said Jason Gross, Petal's CEO. "They're young or have lacked access to financial services."
Petal will use the funds to add scale for its formal launch. The company has been signing up prospective users since the fall and currently has about 40,000 consumers who preordered its alternative payment cards. It uses "cashflow underwriting" as a substitute for credit scoring.
Since traditional scoring relies mostly on borrowing history—a metric that many young people lack—Petal instead turns to a consumer's payments history for recurring bills and other obligations, along with information such as savings and income.
As Petal tracks people's payments over time, it gets a better picture of creditworthiness through its use of artificial intelligence, which Gross said further reduces the risk and enables more people to be brought out of "subprime" status. By focusing on these factors, Petal hopes it can find low-risk consumers who were overlooked by more traditional card issuers.
"The playing field is not level for some folks that are just setting off in their lives," Gross said.
Petal did not name its card issuing bank partner or its planned launch date, and Valar did not comment by deadline. Petal does not charge fees if the consumer pays his or her card bill on-time each month, and charges interest only if the consumer does not pay the full monthly balance each month. It does not charge interest on new purchases and makes most of its income from merchant interchange.
Petal also plans to add marketing and loyalty programs similar to other card brands, and offers benefits through its partnership with Visa, such as theft coverage and roadside dispatch.
The model may face a challenge reaching the target consumer base, which has found alternate ways to obtain credit. "Typically banks can balance people with underqualified credit with the rest of their portfolio," said Brian Riley, director of the credit advisory service at Mercator.
Riley's colleague, Tim Sloane, Mercator's vice president of payments innovation and director of the emergency advisory service, said machine learning can enable analysis on huge data sets. This allows more relationships to be teased out of existing data and provides a deeper analysis on small data sets if the data can be connected through other relationships such as monthly debit transactions from thousands of people along with demographic information, Sloane said.
But traditional credit scores have value because issuers trust them and they are statistically proven, according to Sloane.
"If the date fed into the machine-learning tool doesn't include data elements related to creditworthiness outcomes, then a statistically valid correlation to creditworthiness is impossible to discover," Sloane said.
"Cashflow underwriting" has been used successfully for other types of credit, Gross said, noting Square and Kabbage use a similar model for small-business lending. Square Capital, for example, bases its lending decisions on past and current payments made through Square's platform. And VC funding is flowing to other alternative credit companies, Gross said, noting Affirm just received a $100 million infusion to grow its financing product.
"Petal doesn't require any score at all," Gross said. "It's based on economic fundamentals."