Background

Banco Popular (Popular) sought to offer the unbanked population an alternative to predatory lenders charging 60% to 300%. These solutions would create genuine customer value, foster economic growth and benefit the bank. Popular also wanted to mainstream clients via credit reporting and a migration path of right-fitting products and services.

At the time, Popular’s connection with the unbanked was mainly through non-bank offices offering payroll, check cashing, bill pay and money transfers. Charting new territory, Popular was prepared to learn and adapt towards a solution.

Challenge

The SMB journey to digitalization had already been happening prior to the earliest reports of what was to be a global pandemic. We can now compare that progress with the latest Small and Medium Business Trends report compiled by Salesforce which includes responses from 2,500+ SMB owners and leaders across North America, South America, Europe, and Asia Pacific. The study found that 71% of SMBs survived the pandemic by going digital and 66% reported that their businesses couldn’t have survived using technology even a decade old.

The good news is that their efforts mostly paid off. As eCommerce sales and online shopping spiked to record highs, those SMBs that adapted accordingly, weathered the storm. Today, 83% of SMBs have reported to have at least some of their operations online, with nearly all having migrated part of their operations online in the past year alone.

Solution

We assessed all available data sources associated with non-bank office customers. The centers had five major data collection and storage points.

  1. ID verification- image recognition & government ID lookup systems
  2. Main point of sale - associated verified IDs with all direct transactions
  3. Money transfer & third-party POS - ID, time & amounts passed to bank
  4. Office security system - time, location & image recognition
  5. Loan applications

Models were created by using these data sources to assess work, financial, life behavior patterns, accountability, community connectedness and strength of references. Models used a combination of ML and expert modeling to rapidly evolve performance and improve loan applications. Credit approval was instant, subject to reference verification. Pre-approvals were presented at bank POS. First time loans were small. Subsequent types included unsecured and vehicle secured. Customers were extended custom solicitations for traditional bank relationships.

Results

  • The initial model for clients with no lending history had an average PD of 6% (this is 3 to 4 times smaller than PD at predatory lenders)
  • Models quickly improved to an average PD under 3%
  • Unsecured interest rates were between 13% - 28%
  • Even with small loans and short terms, the program operates profitably due to high automation, low operating costs and advanced analytics

‍