Former Google CIO Douglas Merrill is rolling out today the fifth generation of his Google-like underwriting platform for loans to the underbanked.

Merrill started his Hollywood, Calif., company, ZestFinance, after his sister in law, a single mother of three who has a full-time job and is a full-time student, needed money to pay for a new tire. "Like 60 million Americans, she is underbanked," Merrill says. "She has a bank account but no access to traditional credit. I said, what would you have done if I hadn't answered the phone? She said she would have taken out a payday loan."

Merrill didn't know what that was. Research told him payday loans are a form of subprime lending where a person (usually without access to credit) borrows against a future paycheck, typically in small amounts and over a short period of time. There are 25,000 payday lenders in U.S. These types of loans typically cost 400 percent annual interest (APR) or more, and the finance charges range from $15 to $30 on a $100 loan, according to the Consumer Federation of America.

Nearly 19 million households in the U.S. use payday loans; industry analysts say this adds up to more than $30 billion in short-term credit every year.

"I thought that wasn't fair — is there a way to provide loans to these people that's not so expensive?" Merrill says.

Payday lenders aren't evil, but they don't know how to underwrite, Merrill believes. "They have to charge a lot because they have to assume people won't be able to repay," he says. He considers this a machine learning/Big Data problem. Traditional underwriting methods use logistic regression, feeding 10 variables into that regression, he says. "If any of those variables are incorrect, people get slammed. If you could get access to thousands of signals, you could correct that."

A lot of data for the underbanked is inaccurate or missing, Merrill points out. "A heavy weight is placed on a small number of signals - bankruptcy, credit card default. Traditional math doesn't work."

Merrill's group of ex-Googlers and ex-Capital One people have developed a new underwriting model named Hilbert. "We name all our models after famous dead statisticians," Merrill explains.

Hilbert takes 70,000 signals, runs them through 10 separate underwriting models, each able to consume hundreds of thousands of variables. "The 10 models vote in a way, it's like getting your 10 smartest friends around a table and asking their opinion about something," Merrill says. Results are produced in 250 milliseconds to deliver results. "We're about 50% better than our previous model on approval rate and default. And [the previous model] Hollerith was 50% better than industry average." (The models are evaluated by looking back at loan scores and noting which went bad and which didn't.)

ZestFinance's data sources include ten "alternative" credit bureaus that payday lenders report to. Its new model weaves more human understanding into the mathematical calculations. For instance, there are good bankruptcies and bad bankruptcies, Merrill says. "You have to understand the implications of bankruptcy on the underbanked," he says. "Sometimes it doesn't matter, sometimes it does matter."

Merrill describes Hilbert as "ridiculously stunning." "What I want to do is use math, science and art to give more credit to the underbanked." The goal is to help people save money and develop relationships with loan providers. A $500 loan made through ZestFinance's platform would cost the borrower $400 in interest and fees. A comparable payday loan would rack up $900 in costs, Merrill says.

"The ultimate win will be when major banks realize this is a huge win for them," Merrill says. "They don't know how to underwrite and they're scared."

ZestFinance has one public partner, SpotLoan, that makes first-term installment loans. It also works with banks that prefer not to have their names used publicly.

BillFloat in San Francisco has a similar program for underwriting loans to the underbanked, and it also works with banks. The company derives its data from utilities and telecom operators, who provide information about how well people pay their bills. Atlanta-based credit reporting company Equifax is also starting to provide payment data for the underbanked.