You may not think the number of words in an email subject line says anything about you, but at least one company is betting that the metric can help determine your likelihood of paying back a loan.
LenddoEFL, based in Singapore, is one of a handful of startups using alternative data points for credit scoring. Those companies review behavioral traits and smartphone habits to build models of creditworthiness for consumers in emerging markets, where standard credit reporting barely exists.
In addition to analyzing financial-transaction data, Lenddo’s algorithm takes into consideration things such as whether you avoid one-word subject lines (meaning you care about details) and regularly use financial apps on your smartphone (meaning you take your finances seriously). Lenddo also looks at the ratio of smartphone photos in your library that were taken with a front-facing camera, since selfies indicate youth, helping the company divide people into customer segments.
The data points are unconventional, but Darshan Shah, Lenddo’s managing director for South Asia, says the company’s overall algorithm is a reliable predictor of creditworthiness for the so-called underbanked. For those who lack formal credit histories, Lenddo and others say artificial intelligence can help sort through a variety of data points that, in sum, indicate financial responsibility.
“The potential this technology has is massive,” said Arjuna Costa, a partner at the Omidyar Network, which was founded by billionaire entrepreneur Pierre Omidyar.
The Omidyar Network has invested in alternative-lending companies including LenddoEFL and Boston-based Cignifi. Costa estimates that new forms of loan scoring could help provide credit access to 1 billion people worldwide, especially in emerging economies in Africa, Asia and Latin America.
If such new methods work well in emerging economies, they could be employed in the U.S., where more than 30 million individuals are underbanked, said PayPal Holdings Inc. CEO Dan Schulman. The underbanked lack access to common financial products such as checking accounts, savings accounts, credit cards and loans.
“For most people, it’s not that their expenses are greater than revenues, it’s that their cash flows are uneven,” Schulman said. “And that’s what gets them into trouble.”
In the U.S., creditors rely on FICO scoring, which takes into account borrowing history when gauging whether you’re likely to pay back a loan. The system works well enough for those who’ve already gotten loans and paid them back on time, but it excludes those who aren’t in the digital financial system and can’t get loans in the first place.
From cash to credit
Credit access is worse in emerging markets, where most people conduct cash-based transactions, lack bank records and don’t have assets to serve as collateral. Until recently, financial-services providers have deemed it too risky and costly to provide loans to the underbanked, but the wealth of data available from smartphones is changing that. Alternative data points are providing a new means for credit access beyond traditional micro-loan programs.
For developing countries, easier access to credit could jump-start economic progress. Shivani Siroya, chief executive officer of Santa Monica, Calif.-based Tala, told the story of a woman working in marketing in the Philippines who took out a loan through the Tala platform to buy a glazing machine for her doughnut business, a venture she pursued on the side. Eventually, she was able to quit her marketing job and hire an additional employee for her doughnut store.
Renée Hunter, a researcher at the Centre for Financial Regulation and Inclusion, a South African think tank, said that while consumers might use initial loans of small amounts to deal with day-to-day consumption, they develop a “shift in mindset” when they take out larger loans and find ways to purchase items that could ultimately yield profits.
“Lending is core to business development,” said Avi Goldfarb, a marketing professor and artificial-intelligence (AI) expert at the University of Toronto’s Rotman School of Management. Better prediction models may increase the willingness to lend, he added, which “could be exciting for these countries’ economies.”
Unmet financing needs
About 65 million micro-, small- and medium-sized businesses in the developing world have unmet financing needs that amount to $5.2 trillion in total, according to a joint report from the International Finance Corp. and the SME Finance Forum. That number represents the amount of financing that would be possible if small businesses in developing countries had the same credit access that companies in developed countries have.
Traditional emerging-market lenders have considered moving down-market for decades, but it was often time-consuming and costly to identify potential borrowers, assess risk and collect money. New credit systems that make use of AI, coupled with rising smartphone penetration, cut out some of those impediments: Mobile phones make it easier for financial-services providers to find customers and collect their money, while algorithms reduce the costs of risk assessment, said Omidyar Network’s Costa.
The data points that get fed into alternative credit models vary, but many involve information from mobile devices. The GSM Association, a telecommunications trade group, predicted last year that smartphone penetration in developing markets could increase to 62% in 2020 from 47% in 2016. That compares with a 65% level of smartphone penetration, considered to be “mature,” in developed markets. So the number of people with access to alternative credit is rising by the day.
Smartphones, satellites
Smartphone data can be useful for small-store owners who’ve traditionally needed to pay upfront for inventory. Sokowatch, a Kenya-based company that also operates in Tanzania, uses mobile records of shopkeepers’ orders and payment patterns to extend purchase credit. The company counts Unilever NV
UN,
“Shopkeepers have issues of cash-flow management, so to buy products on credit is really a lifesaver,” said Maelis Carraro, a senior associate at Somerville, Mass.-based BFA, a consulting firm focused on financial inclusion.
Another startup, Newark, Calif.-based Harvesting, employs satellite technology and artificial intelligence to scan small farms for size, crop type, harvest progress and weather effects. Using its own algorithms, the company helps eligible farmers get loans.
Matteo Marinelli, CEO of Myanmar-based microfinance institution Maha Agriculture, is beginning to employ Harvesting data alongside his company’s more traditional credit-scoring model. He is aiming to double the number of active Maha borrowers from less than 15,000 today to 32,000 by March, in part because the Harvesting tools help improve predictive capabilities and enable weather monitoring.
“In the abstract, having access to credit is better than not having access to credit and certainly better than having access to really predatory credit at extremely high interest rates,” said Stephen Rea, a fellow at the Institute for Money, Technology, and Financial Inclusion at the University of California, Irvine.
Still, he cautions that while increased credit access has the potential to meaningfully improve the standard of living in emerging markets, companies and consumers must tread carefully.
Before AI-based credit models, underbanked consumers could seek out credit opportunities within their own communities, through micro-loan programs that allowed them to borrow money locally. Neighbors would pool money and borrowers would each get chances to make use of the communal pot. Those offerings helped expand credit access, but they were known to carry high interest rates, prompting criticism that they only served to trap people deeper in a cycle of poverty.
Interest rates on newer loan options vary widely, but Costa, of the Omidyar Network, notes that annual percentage rates (APRs), which are sometimes in the triple digits, can be misleading since many loans cover relatively short durations, sometimes just a few days.
Artificial intelligence
AI-driven credit models also rely on data-scraping technology, which introduces privacy concerns. Users give credit-scoring apps permission to access things like text and call logs, GPS data, address books and digital transactions, but some companies are more explicit about their privacy policies than others. Consumers also have to trust that lending companies are properly encrypting and safeguarding their data.
Another concern is that artificial intelligence will introduce new kinds of biases into the lending process. The University of Toronto’s Goldfarb warned that prediction models may make use of data that’s discriminatory, such as zip-code information that’s highly correlated with ethnicity.
“If certain ethnic groups get more loans than others because a machine predicts it, that’s something that should be audited and corrected,” he said.
Alternative credit-scoring models are beginning to take off in emerging markets, but they’ve been so far muted in the U.S. and other developed countries, partly due to already higher rates of financial inclusion. Still, the algorithms hold promise, especially for those who are still relying on payday lenders that charge “insane” interest rates, said Camilo Tellez, head of research and innovation at the Better Than Cash Alliance in New York.
Though they’re more focused on business loans than consumer loans, companies such as Square Inc.
SQ,
PayPal’s Schulman has suggested that consumer applications could be next, especially in the U.S.
“If you can really understand an individual or small business through advanced ways of looking at data and information, then you can responsibly lend to individuals who might otherwise be forced out of the financial system,” he said.
Emily Bary is a reporter covering technology for MarketWatch.
