How Fintech Serves The ‘Invisible Prime’ BorrowerKnowledge Wharton
For decades, the main recourse for cash-strapped Americans with less-than-stellar credit has been payday loans and their ilk that charge usury-level interest rates, in the triple digits. But a slew of fintech lenders is changing the game, using artificial intelligence and machine learning to sift out true deadbeats and fraudsters from “invisible prime” borrowers — those who are new to credit, have little credit history or are temporarily going through hard times and are likely repay their debts. In doing so, these lenders serve people who don’t qualify for the best loan deals but also do not deserve the worst.
The market these fintech lenders are targeting is huge. According to credit scoring firm FICO, 79 million Americans have credit scores of 680 or below, which is considered subprime. Add another 53 million U.S. adults — 22% of consumers — who don’t have enough credit history to even get a credit score. These include new immigrants, college graduates with thin credit histories, people in cultures averse to borrowing or those who mainly use cash, according to a report by the Consumer Financial Protection Bureau. And people need access to credit: 40% of Americans do not have enough savings to cover an emergency expense of $400 and a third have incomes that fluctuate monthly, according to the Federal Reserve.
“The U.S. is now a non-prime nation defined by lack of savings and income volatility,” said Ken Rees, founder and CEO of fintech lender Elevate, during a panel discussion at the recently held “Fintech and the New Financial Landscape” conference held by the Federal Reserve Bank of Philadelphia. According to Rees, banks have pulled back from serving this group, especially after the Great Recession: Since 2008, there has been a reduction of $142 billion in non-prime credit extended to borrowers. “There is a disconnect between banks and the emerging needs of consumers in the U.S. As a result, we’ve seen growth of payday lenders, pawns, store installments, title loans” and others, he noted.
One reason banks are less keen on serving non-prime customers is because it is more difficult than catering to prime customers. “Prime customers are easy to serve,” Rees said. They have deep credit histories and they have a record of repaying their debts. But there are folks who may be near-prime but who are just experiencing temporary difficulties due to unforeseen expenses, such as medical bills, or they haven’t had an opportunity to establish credit histories. “Our challenge … is to try to figure out a way to sort through these customers and figure out how to use the data to serve them better.” That’s where AI and alternative data come in.
“The U.S. is now a non-prime nation defined by lack of savings and income volatility.” –Ken Rees
A ‘Kitchen-sink Approach’
To find these invisible primes, fintech startups use the latest technologies to gather and analyze information about a borrower that traditional banks or credit bureaus do not use. The goal is to look at this alternative data to more fully flesh out the profile of a borrower and see who is a good risk. “While they lack traditional credit data, they have plenty of other financial information” that could help predict their ability to repay a loan, said Jason Gross, co-founder and CEO of Petal, a fintech lender.
What exactly falls under alternative data? “The best definition I’ve seen is everything that’s not traditional data. It’s kind of a kitchen-sink approach,” Gross said. Jeff Meiler, CEO of fintech lender Marlette Funding, cited the following examples: finances and wealth (assets, net worth, number of cars and their brands, amount of taxes paid); cash flow; non-credit financial behavior (rental and utility payments); lifestyle and background (school, degree); occupation (executive, middle management); life stage (empty nester, growing family); among others. AI can also help make sense of data from digital footprints that arise from device tracking and web behavior — how fast people scroll through disclosures as well as typing speed and accuracy.
But however interesting alternative data can be, the truth is fintechs still depend heavily on traditional credit information, supplementing it with information related to a consumer’s finances such as bank records. Gross said when Petal got started, the team looked at an MIT study that analyzed bank and credit card account transaction data, plus credit bureau information, to predict defaults. The result? “Information that describes income and monthly expenses actually does perform pretty well,” he said. According to Rees, lenders gets clues from seeing what a borrower does with money in the bank — after getting paid, do they withdraw it all or transfer some money to a savings account?
Looking at bank account transactions has another perk: It “affords [lenders] the ability to update [their] information frequently because it’s so close to real time,” Gross said. Updated information is valuable to lenders because they can see if a consumer’s income suddenly stops being deposited into the bank, perhaps indicating a layoff. This change in circumstance will be reflected in credit scores after a delay — typically after a missed or late payment or default. By then, it might be too late for any intervention programs to help the consumer get back on track.
Data gathered through modern technology give fintech companies a competitive advantage, too. “The technology we’re talking about significantly reduces the cost to serve this consumer and lets us pass along savings to the consumer,” Gross said. “We’re able to offer them more credit for less, higher credit limits, lower interest rates and no fees.” Petal offers APRs from 14.74% to 25.74% to individuals who are new to credit, compared with 25.74% to 30.74% from leading credit cards. It also doesn’t charge annual, international, late or over-the-limit fees. In contrast, the average APR for a payday loan is 400%.
“We think it’s prudent to focus first on financial information — there’s plenty that is not yet factored into mainstream credit decisions.” –Jason Gross
Alternative data and AI also are good for detecting fraud. “Alternative data is extremely powerful in fraud-decisioning,” said Al Goldstein, CEO of fintech lender Avant. Fraudsters can be sophisticated crooks that steal identification, come from organized crime, commit family fraud such as assuming an ex-spouse’s identity, or perpetrate ‘soft’ fraud like forging pay stubs.
Earlier this year, FICO officially gave its stamp of approval to the use of alternative data. It unveiled its UltraFICO score, which uses this data to rate people who could not previously qualify for loans because they had thin or no credit histories. In a 2015 study, FICO found that using alternative data increases the predictability of future behavior among no-score consumers by nearly 10-fold. “A model combining alternative data with bureau data sufficiently differentiates risk within traditionally unscorable segments of consumers, enabling responsible credit decisions,” FICO said.
While FICO acknowledged that consumers with no scores are generally more risky — the default rate is triple that of people who can be scored — there are different types of borrowers lumped together in this group. Using alternative data, FICO was able to give a score to about half of the no-score group. The report said more than a third of those who were newly scored have a rating of 620 or above. Moreover, the majority kept or raised their scores two years later.
Social Media, Privacy and Banking
What fintech executives say they don’t heavily rely on is social media data. “When you start talking about things that are further and further attenuated from the consumers’ ability to pay and actual financial position [such as] how many friends they have on Facebook or contacts on their cell phone, there you really do run into some tricky challenges” about assessing the risk of lending to them, Gross said. “We think it’s prudent to focus first on financial information — there’s plenty that is not yet factored into mainstream credit decisions.” Moreover, he added, some companies such as Facebook don’t allow using their members’ social media information for credit underwriting in the first place.
“Fifty sets of rules isn’t just a loss for [lenders]. It’s a loss for consumers.” –Jeff Meiler
Meiler said his company is committed to using alternative data safely and responsibly. He noted that the ‘no-action’ letter Upstart Network received from the Consumer Financial Protection Bureau was a “hopeful and encouraging sign” that regulators would allow the use of alternative data in credit decisions. In the November 2017 letter, the agency said it had no present intention to supervise or start an enforcement action against Upstart regarding its use of alternative data, provided that it regularly reports lending and compliance information. It also said it was looking at ways alternative data may be used to improve the decision-making process for loans.
According to Goldstein, Avant founded a startup called Spring Labs that is building a blockchain-based decentralized credit and identity network to protect consumer privacy. The startup counts among its board members former FDIC chair Sheila Bair, former Goldman Sachs President Gary Cohn, former HSBC North America CEO Bobby Mehta and Capital One co-founder Nigel Morris. Spring Labs is creating a network where the consumer owns the data and can choose to selectively share information to a lender, with no other third parties involved without permission.
But the fintech lenders at the conference said they tread carefully when handling alternative data lest they violate privacy laws. To that end, they asked regulators for more clarity and streamlining of rules around the use of this data in making credit decisions. Meiler said it’s not that fintechs don’t want to be regulated on privacy like in Europe or California. Rather, “the concern … is potentially to have 50 different [state] rules” and complicate compliance, he said. “Fifty sets of rules isn’t just a loss for [lenders]. It’s a loss for consumers.”
Article by Knowledge@Wharton