Thick file customers vs thin file: AI helps level the playing field
October 6, 2020
Founder & CEO
Other Popular Articles
According to global statistics, 1.7 billion people are unbanked today. Credit-challenged and underbanked customers are not necessarily risky customers. They simply lack credit history data used by lenders to assess their creditworthiness.
Technology can help identify non-risky behaviors that are typical of 'thick-file' customers and find similar micro-behavioral patterns in credit-challenged, underbanked, and 'thin-file' customers, even those with a FICO score under 500.
Many of these thin file customers are creditworthy, but lack the traditional credit record needed by financial institutions to approve their requests for a loan. This lack of accessibility and ease of using financial services across the globe was made even worse with the COVID-19 pandemic. Some previously thick file customers may have become thin file customers by circumstances resulting from the pandemic.
Banking and non-banking financial institutions need to make better credit decisions by redefining the way these players view and assess the creditworthiness of individuals. We do this by giving access to, perhaps, the best source of behavioral data in the world: smartphone and web behavioral metadata.
Built on over 30 million loan applicants, CredoLab's AI-based algorithm crunches millions of features from first party, opt-in/privacy-consented, anonymous smartphone and web behavioral metadata to find the most predictive behavioral patterns before converting them into bank-grade digital scorecards. These credit scores enable any lender to make the most granular assessments possible of their applicants.
By leveraging smartphone and web behavioral metadata, we make it possible to underwrite the credit applications of individuals including thin-files, millennials, self-employed, and gig economy workers in real time, with high predictability, great accuracy, and full privacy protection. As a result, clients have seen 20% higher new-to-bank customer approvals, a 15% reduction in non-performing loans, and a 22% dip in delinquency.
Initially focused on empowering the unbanked and the underbanked, we strive to democratize financial services more broadly and make them more affordable, more accessible and easier to use. Our machine learning-based scoring algorithm can plug into any existing system to make the entire process digital. This drives easy access to credit, no matter the background, and without having to visit the physical branch.
What's the score?
Bank-grade digital scorecards can help the underbanked convert from thin-file to thick-file customers. For example, in the Philippines, 77% of the population is underbanked, so the hurdle to get a bank account is quite high. The fees and bank charges make it impossible for daily wage earners to get access to financial services.
We worked with a local bank to reach customers who lack documentation or financial history and used our AI-powered smartphone metadata-based credit scoring solution. The scoring algorithm uses about 50,000 data points from customers' smartphones to underwrite them - thus converting a thin-file customer to a thick-file one. We only use anonymized data, meaning it would never read customer emails or texts. Instead, it analyzes the number and type of applications installed, the number of new contacts added recently vs. old ones, etc. The algorithm then takes these data points and looks for correlations with responsible lending habits.
Consequently, the bank we were working with in the Philippines was able to acquire one million customers within 10 months of its launch. We've also seen similar successes with clients in other parts of Southeast Asia, Africa and India.
Technology can help "risk-ridden" customers to a healthier financial standing in the following ways:
Mobile and web intelligence
The algorithm analyzes data that the individuals are already generating by simply using their smartphones or filling in a loan/credit card application form online. Individuals are not asked to access to their bank accounts, utility bills, rent history, or social media accounts. They also don't have to fill in a lengthy psychometric questionnaire.
We only collect first party, privacy-consented, anonymous, and non-intrusive metadata to calculate the alternative credit score. No private or sensitive information of the applicant is accessed or moved out of the mobile phone.
The AI-based algorithm learns and becomes more accurate and predictable as more data is fed into it. Therefore, the larger the set of data observed, the better the scoring.
Furthermore, we don't proceed with any credit analysis unless the privacy consent and the necessary mobile operating system's permissions are given. Once these are received, the SDK and the white label apps capture only metadata, defined as data of other data, not private or sensible data. Once the data is analyzed, the statistical outcomes are shared only with the lender tagged to a unique, randomly generated identifier, and not shared with any third party.
All this translates to results like a 20% higher new to bank customer approvals, a 15% reduction in non-performing loans, and a 22% dip in fraud rate.
The time to act in the U.S. is now. Because of payment holidays, in conformity with the Fair Credit Reporting Act, lenders cannot report late payments as delinquent to the credit reporting agencies. As such, lenders will not be able to rely on credit scores to discern who didn't pay back because they are delinquent and who didn't because of payment holiday.
With traditional credit scores becoming unreliable for the next 6 to 9 months, a behavioral solution to determine an alternative credit score becomes instrumental to improve the risk decision-making process.
Using today's technology, it is possible to help an unbanked customer become a customer worthy of the risk.
Read the original article in Mobile Payments Today.