How fintech startups are using big data to solve China’s huge credit gap

October 02, 2017.

 It was almost Angel Zhang’s 22nd birthday. She wanted to do something special, spend a little extra on herself. But instead of asking her parents for more money – they give her a US$300 monthly stipend – she used Huabei, a virtual credit card run by Ant Financial.

Written by Eva Xiao - Originally published at this link.

“Almost every shopping app has their own credit service these days,” says Zhang, a fourth-year university student in northern China. Like many of her classmates, Zhang doesn’t earn any income. “As long as you pay back your loans on time, you can increase your credit line.”

She pays for all kinds of things with credit now: clothes, makeup, toiletries, hotels, train tickets, even her phone bill. The time it takes to save up for something – say a face mask or a new pair of shoes – has been shortened, she tells Tech in Asia.

China has been a very cash-based, non-debt-focused consumer base. This is changing.

University students like Angel Zhang have only recently been able to take out monthly micro-loans in China. According to World Bank estimates from 2014, only 10 percent of China’s adult population had ever borrowed from a financial institution, despite 79 percent having an account. That’s partly because consumer credit scoring is relatively new. China’s banking regulators didn’t develop a consumer credit database until 2006. In contrast, US credit scoring company FICO launched its scoring system in 1989.

The national credit system also has limited coverage. Though the People’s Bank of China (PBOC) had data on roughly two-thirds of the population as of 2015, only about a third had a credit history.

Thanks to big data, however, China’s fintech companies are rising to the challenge.

“China has been a very cash-based, non-debt-focused consumer base. This is changing,” says Zennon Kapron, director of Kapronasia, an Asia-focused financial industry research and consulting firm. “The uptick in consumer credit in terms of borrowing, peer-to-peer platforms, consumer lending, credit cards, and mortgages has increased significantly.”

Still, the PBOC’s credit database is not very robust, he says. Not all lending companies have access to it either. “They have to create their own credit rating system or use third-party scoring companies.”

Faster loans

To gauge someone’s creditworthiness, you have to answer the following question: what is the borrower’s ability and willingness to pay? And in the era of big data, there’s an added dimension: how do you train computer systems to make that decision – with as little human intervention as possible?

“It’s all about the data. The more data we get, the richer the data, the more diverse the data, the better the model,” says Ren Ran, vice president of Dumiao, a credit lending unit under Beijing-based fintech services company PINTEC. According to the company, Dumiao processes about 3 million loan requests per month.

“It’s like cooking. To cook dishes, you need to have the ingredients first. If you don’t have the ingredients, then it doesn’t matter how good a chef you are,” he emphasizes.

To compensate for the dearth and inaccessibility of official lending data, Chinese fintech companies often have to use reams of consumer data from a variety of sources: search results, social media, ecommerce purchases, online travel data, location, phone records – even social connections. Then there’s the more standard set of user information, such as education level, salary, employer, and national ID number.

Compiling as much as possible is especially essential for consumers with little to no credit history. According to Ren, about 85 percent of Dumiao’s user base is between 22 and 35 years old, and less than 15 percent live in major cities like Shanghai and Beijing. These are digital natives who live on their smartphones and are very open to using credit, he says.

It’s like cooking. If you don’t have the ingredients, then it doesn’t matter how good a chef you are.

Relying on big data to assess credit-worthiness has another advantage: speed and cost.

“One of the biggest problems we had in banks was all the transactions had at least one face-to-face interaction,” says Simon Loong, who worked at financial institutions like Citibank and Standard Chartered before founding WeLab, a Hong Kong-based fintech startup.

That means hiring staff and maintaining physical branches to vet potential borrowers. Wolaidai, a credit lending startup under WeLab, on the other hand, has a completely mobile onboarding process. Through technology like facial recognition, text and image analysis, and monitoring users’ online behavior, the fintech startup is able to make credit decisions in one to two seconds and process more than a million applications per month, says Simon.

According to the company, Wolaidai’s loan delinquency rate – where borrowers are late on payments – is two to three percent. Ticket sizes range from US$450 to US$7,500.

“There was a bank that came to me and said that they couldn’t do lending less than [US$30,000] per ticket size, because they couldn’t break even. It’s just too human intensive,” he tells Tech in Asia. “So they can’t do small ticket-sized lending for the mass market.”

Filtering out noise

Of course, using big data to calculate how much users can borrow comes with its own set of challenges. Machine learning models have to be updated regularly to take into account new data and changes in the market.

“Sometimes we find blind spots in our model,” says Ren, explaining that Dumiao’s data scientists are constantly analyzing their system for improvements. “All models are wrong, but some are useful,” he adds, quoting world-renowned statistician George Box.

All models are wrong, but some are useful.

Fintech startups also have to automate fraud detection, like catching people who steal personal IDs to apply for loans. That’s usually done by tracking user behavior, like the way someone types out their ID number, or their physical location, if users agree to share that information.

For instance, Fenqile, an installment-based ecommerce site for young professionals, might raise an alert if a large number of users – located in roughly the same area – buy the same product simultaneously.

It could be a case of fraud, or it could be an anomaly, explains a company spokesperson. Once, the system identified a group of users buying the same Canon camera. It turned out that a local photography club was recommending the camera to its members.

Not all data is created equal either. In fact, the best data comes from the PBOC’s credit report, which is harder to access, says Sync Shan. Previously at Chinese search giant Baidu, Shan now runs the big data team at Dianrong, a Chinese online lending marketplace co-founded by Lending Club co-founder Soul Htite.

“If we can get our hands on a personal credit report, we’ll definitely use it,” says Shan. It’s very efficient data, which means its ability to differentiate and classify good lenders from the bad is high.

He shows me a diagram of a triangle – a hierarchy of data for analyzing creditworthiness. At the top, there’s data from the national credit database, which includes housing loans and credit card information. The level underneath is consumption or purchasing habits, then mobile phone data, social connections, and user behavior on the lending app itself.

“The higher you go, the more effective the data is,” he explains. “But it covers less people.” Conversely, the further down you go, the weaker the correlation, but the wider the coverage.

Then there’s the cost of data. It has to be balanced with its performance in Dianrong’s model, says Shan, which classifies users into different tiers of credit risk. A cheaper data point will be about 15 cents, while others can be much more expensive, he says, declining to specify.

Data crunch

At scale, the cost of data can quickly escalate, especially for startups without multiple consumer-facing platforms like Ant Financial, which can easily access ecommerce data from Alibaba. Relying on third parties to supply data – or in some cases, process it – also requires a fair bit of due diligence. This summer, a Chinese data broker was reportedly shut down by the police for investigation after failing to protect consumer privacy.

The onus is also on fintech startups to attain permission for data collection – usually through lengthy terms and conditions statements.

Fintech startups can barter their technology for trusthworthy data.

“Although the data has been transferred to third-party data brokers for data processing, the main responsibility of protecting the data is still the financial services institute,” says Samuel Sinn, cybersecurity and privacy services partner at PricewaterhouseCoopers, who advises companies on technology risk management.

It doesn’t mean that they can release their liability after transferring the data, he emphasizes.

That’s why partnerships are crucial for fintech startups, who can barter their technology for trustworthy data. This is especially true for niche demographics, like students or farmers. Wolaidai, for instance, processes loans for Ule, an ecommerce startup that focuses on rural shops and villages. In exchange for its credit assessment technology, Wolaidai gets data on China’s hinterlands.

Foxconn, another one of Wolaidai’s partners, is helping the startup tackle China’s blue collar worker segment. The manufacturing giant is one of the largest employers of blue-collar workers in China, says Loong.

Black box

To be sure, the challenges to big data-driven credit assessment go beyond technology. When online data is collected en masse and used to determine the financial livelihood of people – especially vulnerable populations – ethics plays a huge role.

In the US, for instance, major data brokers have come under fire for selling ‘sucker lists’ that identify individuals who are “old, in financial distress, or otherwise vulnerable to certain types of marketing pitches,” writes investigative journalist Julia Angwin in her 2014 book Dragnet Nation: A quest for privacy, security, and freedom in a world of relentless surveillance.

China doesn’t have a data privacy act as of today.

In October 2012, the US’s Federal Trade Commission fined Equifax – now infamous after its massive data breach this year – for selling lists of people who were late on recent mortgage payments to fraudulent marketers.

In China, laws of data privacy and personal data collection are only just starting to emerge. In June, China enacted its latest cybersecurity law, which details different principles and guidelines on cross-border data transfers, data collection, information infrastructure, and more.

“China doesn’t have a data privacy act as of today,” Sinn tells Tech in Asia. “It’s covered in the cybersecurity law under personal information protection.” But it’s just the beginning – new laws are expected to be finalized in the future and companies need to stay tuned, he adds.

Predatory loan sharks are also an issue that Chinese regulators are battling. In particular, some lending companies will target college campuses in China and charge students exorbitant interest rates, or even hire thugs to take back loan payments. Overwhelmed by debt and hounded by debt collectors, some students have committed suicide.

Finally, China’s consumer credit scoring landscape still lacks standards. In addition to the PBOC’s credit database, eight other companies have been tasked with developing a consumer credit scoring system, including Tencent and Sesame Credit, Ant Financial’s proprietary credit system.

That was in 2015. Today, none of the eight companies have received an official credit scoring license, as regulators remain dubious of each system’s fairness and independence, given their ties to the company’s main business. This year, the Chinese government set up a national clearinghouse for online payments, which would force different payment systems – Alipay, WeChat Pay, and others – to share transaction data. Perhaps that could pave the way for a national credit system.

For now, China’s fintech startups will have to continue poring through different caches of data and various credit scores from Sesame Credit and other companies, trying their best to tie everything together to form a complete picture of a prospective borrower.

“In China, the data is not as aggregated as in the US,” says Ren, who used to work at Capital One in the US. “It’s very sparse, but that’s actually creating a very good atmosphere for people – data guys like me – who want to play with data,” he says.

Converted from Chinese currency. Rate: US$1 = RMB 6.64.