Creditors want to sneak into your phone before they give you a loan

As digital lending starts to rely increasingly on phone records, messages and social media, questions about reliability of proxy indicators remain unanswered

If you're in your early 30s, applying for a loan at 10 a.m. in the morning and using a lot of backspace to delete your answers while filling an online application for a loan from a digital lender, chances are you won’t get the money. This is the not some new regulation, but just how self-learning algorithms work. Deployed by lenders to vet a person’s ability and intent to pay, computer algorithms are now replacing human intervention and making lending decisions on a company’s behalf in a matter of 30 seconds to seven minutes.
 
These lenders come in different avatars. Their offerings include consumer-durable lending, which allows a person to get an online purchase financed at one click and pay back in monthly installments, peer-to-peer lending allowing borrowers to get their loans funded by retail investors or conventional personal loans which can now be received in less than half an hour directly into the bank account. What holds all these fledgling industries together, however, is the new-to-credit consumer.

 
A new-to-credit consumer is someone who hasn’t taken a loan yet and never used a credit card. Since traditional credit bureaus like CIBIL rely on data about someone’s credit repayment habit to create a risk score, new-to-credit customers have a CIBIL score of -1, which means no credit history.
 

To bridge this divide, companies have started using alternate data to create a virtual blueprint of an applicant’s life and their typical day. A user is asked to download a mobile app in most cases which then asks for permissions to scrape through each available data point in the phone ranging from battery levels to contents of text messages in some cases. All this data collected is routed to the company’s servers in real time where algorithms analyse this data and match it with their ‘learning’ to deduce if a person’s eligible for a loan or not.

 
What’s more, these algorithms also instruct companies using them about the interest rate a person should be charged. For instance, consider i2ifunding’s algorithm which the company uses to evaluate the eligibility of a borrower on its peer-to-peer platform.
 
“Our data algorithm uses a couple of alternative data sources. For instance, we use phone and text messages to see what a person is communicating. If there’s a use of words such as collections, loans and delay in repayments in text messages, then it raises an alarm for us,” said Raghavendra Singh, co-founder of i2ifunding, which lends Rs 10 million a month.

 
Singh added that the company also looks at call data to decipher if a person is in constant touch with family and friends or if they are avoiding calls made by banks and lending companies. Another interesting metric being tracked by lending companies including i2i is number of digital lending apps on the phone which indicates customer’s potential leverage outside the banking system.
 
Similarly, another company, Zestmoney uses data scraping from phone to track people’s financial position through a variety of proxy indicators such as the time of the day when a customer applies for a loan, location history as well as consumer behaviour through their previous e-commerce purchases.

“For example did they use a search engine to find us or respond to an ad? What time of day did they apply and how did they interact with our application. Behavioural data is very useful in understanding a customers intent to repay or not,” said Lizzie Chapman, CEO and co-founder, Zestmoney.

Chapman said that the company collects all data only with user consent. The app also tracks social media profiles of a customer and records IP address of their computers or phone to reduce fraud.

While the company does use some credit bureau data to make a lending decision, it deploys five artificial intelligence models in order to determine creditworthiness.

“ZestMoney does not rely on a credit bureau score to make approval decisions - not only do less of our customers have a score (i.e are “new to credit”) but we have also found that credit scores of our customers are not completely predictive of their behaviour when it comes to a small ticket EMI style loan,” Chapman said, adding that the company lends Rs 500 million a month.

Meanwhile, not everyone is convinced that using alternate data is the best way to lend even small accounts because of sheer reach and volume of these lenders and the largely unregulated space that they operate in. An industry insider who works on developing algorithms for these companies spoke on condition of anonymity and said that machine learning models require years to reduce fraud and even then a customer can learn how to game the system.

“It’s like betting with your eyes closed, you are seeing what a customer chooses to show you. There’s no double check of things because all data like calls, messages and social media can be easily faked while bank account statements or credit bureau scores aren’t easy to manipulate,” he said.

Credit bureaus like CIBIL, however, are realising their deficiency when it comes to new-to-credit customers and they are trying to include a host of different indicators to their database so as to allow banks and NBFCs make better lending decisions.

“We are speaking to the regulators about getting data from utility and telecom bill payments into our database for building the database,” said CIBIL COO Harshala Chandorkar.

Chandorkar added that almost 18-20 per cent of all loan seekers are new-to-credit meaning that there’s no credit history. She, however, said that new age models need a lot of evaluation before being adopted at a mass scale.

“These models are in vogue but everyone is unsure about their large scale applicability, we are closely watching the space to see if things develop in a big way that can help big lenders also. As of now, the situation is that some people are claiming that their models work but we are yet to see concrete evidence,” she said.

Even as companies claim that alternate data has helped them lend better, the official numbers of delinquencies for any of these lending companies are yet to be seen. For instance, i2i is a regulated peer to peer lender which claims it brought down NPAs to less than 2 per cent but RBI hasn’t released any numbers so far. Those looking for loans, however, have no way but to allow companies to sneak into their phones and lives.

“We have a manual verification option too when customers decline to share data on their phone which involves looking at their bank statement, visiting them etc but we often don’t do this and reject the loan. We want to be 100 per cent sure before approving credit,” said Singh from i2i.