Douglas Merrill, chief executive of ZestFinance, jumps up, stares at the computer monitor on the wall and says, “Holy crap, that can’t be right.”
For five years, Merrill has harnessed oceans of online data to screen applicants for the small, short-term loans provided by his Los Angeles-based firm. Improvements in default rates have come in fractions of a percentage point. Now, on this July day, his researchers are claiming they can improve the accuracy of their default predictions for one category of borrower by 15 percentage points.
As sightseers stroll along Hollywood Boulevard below his second-floor office, Merrill, who has a PhD in cognitive science from Princeton University, approves accelerated tests of the finding, which concerns borrowers who make initial payments on time and then default. It is based in part on new data about those who pay their bills electronically.
“It’s hard to model what somebody’s going to do in six months or to even know which data are relevant,” he says. “That’s the subtlety, the artistry of what we do.”
Merrill, 44, sees himself as a rebel in the world of finance. He looks the part, with shoulder-length hair, a tattoo with peacock-feather patterns on his left arm and black fingernail polish on his left hand. He’s one of dozens of entrepreneurs tapping the vast new storage and analytical capabilities of the Internet in a quest to modernize — and perhaps take over — the credit-scoring decisions at the heart of consumer finance.
The flood of undigested information that flows online — or “big data” — has been harnessed most successfully in business by Google to match its advertising with users’ search terms. In finance, big data makes high-frequency trading possible and helps the “quants” in the hedge-fund industry spot trends in stock, bond and commodities markets.
Commercial banks, credit card companies and credit bureaus have dived into big data, too, mainly for marketing and fraud protection. They’ve mostly left advances in the field of credit scoring to upstarts such as ZestFinance, which collects as many as 10,000 pieces of data about the poor and unbanked, then lends them money at rates as high as an annual 390 percent.
“Consumer finance is changing at a pace not seen before,” says Philip Bruno, a partner at McKinsey & Co. and author of a February report on the future of retail banking. “It’s a race between existing institutions and new non-bank and digital players.”
Three of the most-digitized credit scorers for low-income borrowers are ZestFinance, LendUp and Think Finance. Advances in computer science allow these firms to collect thousands of facts on each loan applicant in a matter of minutes. That compares with the few dozen pieces of basic data — mostly a borrower’s debt burden and payment history — that Fair Isaac Corp. requires to compile the FICO score that is the basis of 90 percent of U.S. consumer loans.
ZestFinance’s Merrill, who was chief information officer at Google from 2003 to 2008, compares his job to hydraulic fracturing — that is, blasting through shale until oil embedded in the rock starts to flow. His staffers, several of whom are also PhDs, sort their data using machine learning, or algorithms that can invent their own new analytical tools as the data changes, rather than just following preprogrammed instructions.
The firm’s machines quickly organize individual facts about a loan applicant, including data that FICO doesn’t use, such as annual income, into “metavariables.” Some metavariables can be expressed only as mathematical equations. Others rank applicants in categories, including veracity, stability and prudence.
An applicant whose stated income exceeds that of peers flunks the veracity test. A person who moves residences too often is considered unstable. Someone who doesn’t read the terms and conditions attached to the loan is imprudent.
One peculiar finding: People who fill out the ZestFinance loan application in capital letters are riskier borrowers than those who write in upper- and lowercase. Merrill says he doesn’t know why.
Venture capitalists are betting that the new credit scorers will thrive. Since 2011, ZestFinance has attracted $62 million in venture financing, plus $50 million in debt financing from hedge fund Victory Park Capital Advisors. In 2013, a group led by PayPal billionaire Peter Thiel invested $20 million. LendUp has raised $64 million.
Big banks and credit card companies are watching closely what Merrill and his fellow big-data crunchers are up to. For now, the new firms operate on the outer margins of the financial system — as an alternative to payday lenders, which make loans secured by borrowers’ paychecks, or, in the case of small businesses, factors, who buy receivables from merchants at a discount.
After rejecting two-thirds of applicants, ZestFinance approves loans that average $600 for those that make the cut. The borrower pays $400 in interest for a six-month $600 loan. That computes to an annual percentage rate, or APR, of 390 percent — at least four times the subprime credit card rates offered by some banks. Borrowers have average annual incomes of $30,000, and many have poorly documented credit records or a history of defaults.
Merrill says he only breaks even on the loans, since he spends far more than prime lenders do for capital, marketing and the write-off of defaulted debt. He draws a sharp distinction between his operation and payday lenders, whose loans are an advance on future salary payments.
Annual interest on payday loans runs as high as 521 percent, according to the Consumer Financial Protection Bureau. Because many payday borrowers repeatedly roll over the principal, they wind up paying the interest several times.
Merrill doesn’t allow rollovers. ZestFinance borrowers pay off the full loan in installments and aren’t eligible for a new loan until they pay off the first one. If they qualify for a second loan, the interest can drop to 240 percent.
Critics say such rates are not defensible.
“If your APR is north of 30 percent, even for a month, it’s hard in my mind to justify,” says Mark Pinsky, CEO of Opportunity Finance Network, a nationwide consortium of banks and credit unions that make federally subsidized loans at rates of 8 to 13 percent.
Persis Yu, an attorney with the National Consumer Law Center in Boston, says Merrill’s loans are overpriced and his underwriting methods still unproven.
“I’m concerned,” Yu says, “about using all these low-income consumers as guinea pigs.”
For business borrowers, Kabbage, an Atlanta-based lender, offers its customers loans at a more sustainable 40 percent APR. In exchange, however, the lender demands permission to monitor a company’s bank and credit card accounts and keeps a constant digital eye on cash flow from operations. Kabbage also scours the Internet to keep track of its borrowers’ reviews on Facebook and Yelp.
By year’s end, the three-year-old firm will have placed loans worth $600 million, CEO Rob Frohwein says, and he projects a further $1 billion next year.
Kabbage is one of several new data-savvy lenders trying to displace factors, which buy up small-business receivables for quick cash.
High-interest lenders are under close scrutiny by state and federal regulators. Eighteen states and the District have outlawed payday loans, according to the Consumer Federation of America.
The Consumer Financial Protection Bureau is due to announce the first federal payday-loan guidelines by early next year, and ZestFinance could be caught in the crossfire. Merrill says he could benefit if the bureau focuses on loans that include rollovers.
ZestFinance is licensed to lend money in nine states. Since 2009, the company has made 100,000 loans through a Web site called ZestCash and another called Spotloan. Spotloan is owned by the Turtle Mountain band of the Chippewa Indian tribe of North Dakota, which asserts it isn’t subject to state laws.
Investors find ZestFinance’s proprietary software of as much interest as its loan rates. Merrill says using his models to separate good credit risks from bad has resulted in a 15 percent default rate — half the average of payday lenders.
“ZestFinance has reduced a payday default rate that hadn’t budged in 30 years,” says Arjan Schütte, founder of Los Angeles-based Core Innovation Capital. “That tells me machine learning can do a radically better job of underwriting.”
Schütte, who invests in financial-services start-ups, says he’s passed on ZestFinance so far.
Merrill’s long-term goal is to license his computer models to lending institutions that target the middle class. ZestFinance is running tests with top credit card providers, says Mike Armstrong, the company’s marketing director, though he declines to name them. Since the firm started in 2009, he says, it has had similar talks with banks, auto lenders, cable TV companies, hospitals and debt collection agencies. ZestFinance, he says, could do all the underwriting for a regional bank.
The banks are not knocking down the door. In the United States, the Consumer Data Industry Association hasn’t updated the core data its member banks send to credit bureaus since 1999, says Trevor Carone, senior vice president at Experian, the Dublin-based company with dossiers on 600 million people. The data collected focuses on a consumer’s total debt and repayment history.
Banks and credit bureaus often supplement FICO data with their own proprietary reports. In June, for instance, Experian launched a “risk stability index” that shows whether a consumer’s credit rating is getting better or worse over time. If it’s improving, the borrower is a less of a credit risk.
Yet most banks and credit bureaus have been slow to innovate on credit scoring for low-income borrowers, says Raj Date, managing partner at Fenway Summer, a Washington firm that invests in financial start-ups. The default rate on prime-rated credit cards is 2.9 percent, Date says.
With so few deadbeats, and low-cost capital from depositors, banks have little incentive to buy into Merrill’s complex algorithms.
“Banks don’t care if they can cut defaults among prime or superprime borrowers by a quarter of a point,” says Jeremy Liew, a partner at Lightspeed Venture Partners, a ZestFinance investor since 2011. “But at the bottom of the credit pyramid, if you cut defaults in half, then you radically change the economics.”
Not just any credit analyst can do it. “This is a hard problem,” Liew says. “You have to come from a place like Google or PayPal to have a chance of winning.”
Merrill was born for the role of iconoclast. He grew up in Arkansas and was deaf for three years before surgery restored his hearing at age 6. He didn’t realize he was dyslexic until he entered high school. These disabilities, he says, taught him to think for himself.
At the University of Tulsa and then Princeton, his concentration in cognitive science — the study of how humans make decisions — eventually morphed into an interest in finance. Merrill worked at Charles Schwab, PricewaterhouseCoopers and Rand Corp. before Google, where, among other responsibilities, he directed efforts to compete with PayPal in electronic payments.
Today, Merrill and his 60 ZestFinance employees use a smorgasbord of data sources to evaluate borrowers, starting with the three-page application itself. He tracks how much time applicants spend on the form and whether they read terms and conditions. More reflection, he says, indicates a greater commitment to repay.
Merrill says he doesn’t scan social-media postings. He does buy data from third-party researchers, including Atlanta-based L2C, which tracks rent payments. One red flag: failure to pay mobile or smartphone bills. Someone who is late paying a phone bill will be an unreliable debtor, he says.
Once he’s organized his initial data sets into metavariables, he activates an ensemble of 10 algorithms.
An algorithm called Naive Bayes — named for 18th-century English statistician Thomas Bayes — checks whether individual traits, such as how long applicants have had their current bank account, help predict defaults.
Another, called Random Forests, invented in 2001 by Leo Breiman at the University of California at Berkeley and Adele Cutler at Utah State University, places borrowers in groups with no preset characteristics and looks for patterns to emerge.
A third, called the “hidden Markov model,” named for 19th-century Russian math wizard Andrey Markov, analyzes whether observable events, such as lapsed mobile-phone payments, signal an unseen condition such as illness.
The findings of the algorithms are merged into a score from zero to 100. Merrill won’t say how high an applicant must score to get approved. He says that in some cases where the algorithms predict a default, ZestFinance makes the loans anyway because the applicants’ income suggests they will be able to make up missed payments.
Merrill’s customers don’t necessarily know how thoroughly ZestFinance has scoured public records to learn everything about them. At small-business lender Kabbage, the company practically becomes the borrower’s business partner.
Frohwein, 46, makes loans averaging $5,000 in all 50 states, with the typical client, he says, borrowing a total of $75,000 over three years.
His computers monitor their bank, PayPal and Intuit accounts, which provide real-time updates on sales, inventory and cash flow. Kabbage might hike up the interest rate if business is bad or ply borrowers with new loan offers if they are doing well but are short of cash.
Frohwein considers his 40 percent APR reasonable, considering the risk he takes. Unlike factors, he doesn’t buy receivables. And he doesn’t ask business owners to pledge any property as collateral. Instead, he depends on his algorithms to find good credit risks. He says his customers increased revenue an average of 72 percent in the six months after signing up with Kabbage.
“If you’re using the loan to produce new and profitable revenue, you should do that all day long,” he says.
Jason Tanenbaum, CEO of Atlanta-based C4 Belts, says he turned to Kabbage after SunTrust Banks asked him to wait up to 60 days for approval of a loan. He got the go-ahead on a $30,000 credit line from Kabbage in seven minutes.
Tanenbaum, 28, who has five employees, sells brightly colored plastic belts online.
“If this solution didn’t exist,” he says, “we would have closed our doors.”
Like other purveyors of high-interest debt, Kabbage has attracted the attention of Wall Street. As of mid-September, Frohwein says, he had securitized and sold to investors $270 million of his loans, providing an annual return in the mid-single digits.
Merrill says he needs more years of successful underwriting to open Wall Street’s securitization spigot; he now relies on venture capitalists and hedge funds. He says his goal is to create a more-accurate and more-inclusive credit system.
“People shouldn’t be abused by unfair and opaque rates simply because we don’t know how to underwrite them,” he says, referring to payday lending.
Like other big-data aficionados, Merrill is hoping his credit-scoring breakthroughs will be adopted by mainstream financial players. In the meantime, he risks getting stuck in the payday-lending swamp he says he is seeking to clean up.
The full version of this Bloomberg Markets article appears in the magazine’s November issue.
In a 2012 patent application, Douglas Merrill’s ZestFinance gives examples of how it scours the Internet, collecting as many as 10,000 pieces of data to draw portraits of loan applicants. The nurse and prison guard are hypothetical.
|2 in 10 years||Number of addresses||7 in 5 years|
|1||Social Security numbers||4|
|15 in 45 minutes||Number of ZestFinance Web pages accessed while completing application||3 in 7 minutes|
(1) Lower rent shows higher income-to-expense ratio, quicker recovery after default.
(2) Fewer addresses indicate more stability.
(3) Reading the fine print indicates applicant is a careful consumer.
(4) Fails veracity test as prison guards living nearby report income of $35,000 to $40,000.