The Wall Street Journal
To judge potential borrowers, lending app Yongqianbao collects more than 1,200 data points, from the usual credit information like a bank-card number to what phone a person uses and how many calls go unanswered.
Financial-technology upstarts like Yongqianbao are on the leading edge of credit scoring, using artificial intelligence to analyze swells of data, much of it unconventional. "While banks only focus on the tip of the iceberg above the sea, we build algorithms to make sense of the vast amount of data under the sea," says Jiao Ke, founder and chief executive of Yongqianbao, a product of Beijing Smarter Finance Information Technology.
China is an ideal laboratory. The country lacks reliable credit-history systems for individuals, while data-privacy expectations are minimal and smartphones are everywhere and used by young Chinese for nearly everything—including personal finance.
Some 160 million Chinese went online to take out loans worth 1.2 trillion yuan ($173.9 billion) in 2016, according to research firm iResearch. That puts China way ahead of any other country, including the U.S., where online lenders provided more than $36 billion of loans in 2015, according to KPMG. And iResearch forecasts China's total will grow at an annual rate of about 50% for the next three years.
In the U.S., credit reports include bill-payment and loan histories, current debt, even criminal records. In China, mortgages were barely available before 2000. Americans average 2.6 credit cards each; Chinese, a third of a card. Plus, many Chinese are still paid in cash.
Fintech in China is grappling with fraud and defaults, a result in part of fast growth and spotty regulation. The iResearch report says overdue rates over the past four years ranged from 10% to 20%—the "main factor that prevents online lending from becoming a mainstream channel in China's financial industry."
Fintech champions say the power of artificial intelligence is bringing default rates down by finding correlations between smartphone behavior and risk and using them to create tools that can analyze creditworthiness in an instant.
Yongqianbao crunches its 1,200 data points to generate more than 100,000 risk scenarios in a few seconds. The result: a loan whose terms are tailor-made.
Fintech startups elsewhere are also using alternative data such as mobile usage and online-shopping records to assess credit risk. But China's scale—695 million smartphone users and huge amounts of online lending—allows AI programs to get better faster, say people in both the internet and finance industries.
Affirm, the San Francisco-based online lender led by PayPal co-founder Max Levchin, last fall projected its 2016 loan volume would hit $300 million. Yongqianbao, by comparison, approved 1.2 million loans worth 1.8 billion yuan ($270 million) in February alone.
Yongqianbao, whose name means "use-money pal," mostly issues short-term "payday" loans of 500 yuan to 5,000 yuan ($72.50 to $725). About 80% of its borrowers are younger than 30. Loans past the 60-day due date stood at 2.8% in February, according to Mr. Jiao, a former project manager for search engine Baidu.
A competing platform, Dumiao, averages 30,000 personal-loan applications a day. It makes cash loans of 1,000 yuan to 50,000 yuan, mostly to people under 35. As of January, 1.4% of loans were 30 to 89 days past due.
Among the two startups' findings from running hundreds of millions of loan applications: iPhone users tend to have lower late-payment rates than Android phone users, and people who don't answer calls or whose outgoing calls go unanswered represent a higher default and fraud risk. Other red flags: making many changes when filling in the application, letting batteries run down and changing phones frequently. Multiple applications from a single Wi-Fi hot spot is a danger sign. And users who borrow in one city but spend in another are poorer risks.
These startups, which aim to become the credit bureaus of digital finance, differ on how much decision-making power to entrust to their fast-learning machines.
Dumiao, whose executives hail from traditional financial institutions, approves only 5% to 8% of first-time cash-loan applicants because they're "overkilling" for safety purposes, says Ren Ran, who left Capital One Financial last year to join Dumiao's parent, Pintec Group, and leads Dumiao's data-science unit.
Yongqianbao, whose approval rate for first-time applicants is 20% to 30%, is run by engineers.
"We don't hire any risk-control people from traditional financial institutions like the other fintech companies do," Mr. Jiao says. "We don't need human beings to tell us who's a good customer and who's bad. Technology is our risk control."
This article was licensed through Dow Jones Direct.
Tagged in: industry news
Learn how FIS can help you stay on top of industry trends and address your business challenges.Contact us