BankMobile has become the first bank to start using online lending software developed by Upstart, which is designed to use artificial intelligence and alternative data to determine the creditworthiness of consumers with thin or no credit files.
BankMobile, the digital-only subsidiary of Customers Bank in Wyomissing, Pa., which is due to be spun off and merged with Flagship Community Bank in mid-2018, is planning to use the software to offer its first credit product to recent college graduates and other young consumers. BankMobile has relationships with 800 universities through which it offers deposit accounts.
“Our customer base is thin-credit-file individuals,” said Luvleen Sidhu, president and chief strategy officer of BankMobile. “One of the huge advantages of banking with us early on is we get to know their behaviors and what sort of majors they have and when they graduate, what sort of cash-flow potential they have, etc. With that sort of information and partnering with Upstart, which has the algorithms and machine learning to decipher dimensions above and beyond the FICO score, they’re really a good match to be able to penetrate our thin-credit-file, near-prime portfolio of customers.”
In addition to using artificial intelligence, Upstart partly relies on education data to assess creditworthiness. The use of education data in loan decisions has been controversial and is still being examined by the Consumer Financial Protection Bureau. In September, the CFPB sent Upstart its
To apply for BankMobile loans, applicants share their highest level of education obtained, the name of their school or university, their graduation year and their area of study. That information is what Upstart’s underwriting software uses to make its credit decision. (Eventually the data will probably be prepopulated from BankMobile’s databases.)
“We hope to figure out what information is the most relevant and correlates to the best risk assessment,” Sidhu said. “We’re not starting from scratch, we do have a relationship with these customers on the deposit front, and we can get more and more information from that deeper relationship over time that might continue to help with the underwriting Upstart is helping us with.”
BankMobile’s customer base is not superprime with high FICO scores, but people at the other end of the spectrum.
“We look at this as a way to say, if you’re a young person who is just coming out of school and you have no credit file, the fact that you took those steps to higher education is meaningful,” said Jeff Keltner, head of business development at Upstart. “We think it can be powerful in expanding access to credit, not just say, ‘hey, you went to a great school, we know you’re not going to default.’ That’s great and true but not the purpose of our program, which is how do we use these things to tackle a set of consumers who have traditionally been locked out of the market through no fault of their own.”
BankMobile chose Upstart based on the solidity of its own loan portfolio, Sidhu said.
“They were one of the few survivors [in online lending], and I think that’s partially because they have a strong model that starts with their algorithms and machine learning and they have a team that comes in with deep expertise to be able to build those models,” she said. “They’ve been able to prove they’re able to get better pricing and able to anticipate defaults better than those that are just using a traditional FICO score to evaluate credit.”
She also believes the software provides a positive customer experience.
“That was very important to us because we’re all about being digital and the digital experience has to be superior to what people are used to in the bank branch and going through traditional models,” Sidhu said.
Upstart was also willing to modify its software to BankMobile’s requirements. For instance, the bank offers its customers a 25 basis point discount on loans if they fund them through their BankMobile account, and provides 1% back after 12 months of consecutive payments. It charges no origination fee and a slightly lower rate than Upstart’s algorithm would normally set. (Loan rates range between 6% and 29.99%.)
Integration of Upstart’s software and BankMobile’s website and app took only a couple of months, according to Keltner. The bulk of that time was spent on compliance reviews, disclosures on the Upstart site, the adverse actions sent to borrowers, and the email communications sent borrowers, he said.
The whole concept of using AI and alternative data to make loan decisions is still relatively new and few banks have jumped into it. Many fear how regulators will view such decisions.
Sidhu said she became comfortable with the concept after spending a lot of time with Upstart to define risk and creditworthiness parameters.
“We have part of the model looking at traditional ways of looking at credit and looking at the FICO score,” she said. “We’re able to add on additional layers to evaluate our customers in nontraditional ways so some of these thin-file or near-prime customers that are part of our customer base have a chance to get credit.”
The methodology Upstart and the CFPB jointly created to test for fair lending and disparate impact will also be applied to the BankMobile loans.
“The nice thing about the methodology we built with the CFPB is it allows us to not only test the book but look at changes to the model and analyze if they would maintain or break compliance with the metrics we laid out with the CFPB,” Keltner said. “The ability to say when we make changes to the model it will still meet a fair lending standard that’s defined and agreed upon by a regulator is powerful.”
BankMobile is starting with basic personal loans offered through the Upstart platform that could be used to refinance credit card debt, home improvement, vacations and the like. Next year, the bank plans to expand into credit cards, student loan refinancing, home equity and car then eventually car loans made through the software.
