Our client had a bold mission: to provide solar technology to off-grid African villages. The products were financed by leasing. However, traditional banking models completely fail in African markets. You cannot underwrite a loan using a bank statement that doesn't exist.
The initial approach relied on manual guesswork, leaving the client highly exposed to credit risk. They needed a rigorous, scalable way to identify who pays back. Even if traditional financial data was completely absent.
Our Strategy: All Data is Credit Data.
We transitioned the client away from intuition and built a custom Credit Risk engine. Instead of looking for formal income, we engineered the model to hunt for "Digital Proxies of Wealth", analyzing agricultural, social, and geographic variables to identify hidden indicators of financial reliability.
The Credit Risk Engine
To process this unconventional data, we developed a credit risk model that evaluated over 220 distinct features. The model transformed customer intuition into hard, quantitative risk metrics:
- The "Cocoa Index": We quantified agricultural liquidity by counting literally everything: cows, goats, chickens, and bags of cocoa beans or yams. We proved that a farmer's agricultural yield and livestock count act as highly reliable, informal collateral.
- Social Trust as a Credit Score: We integrated reputation as a data point, By asking, "Would you lend your neighbor money?" in a village of 300 people, we quickly established the social reputation and community trust of potential customers, which proved to be great proxies for a “performing” client.
- Hygiene Proxies: The model uncovered that families with school-aged children were 1.25x more likely to repay on time, acting as a proxy for long-term planning. Furthermore, geographic risk was identified: districts with low soap availability or shared toilets showed significantly higher default rates.
- Sector-Specific Risk Profiles: The algorithm successfully isolated high-risk lending environments, revealing that SME loans were 3.44x more likely to be non-performing, and workers in the mining sector carried a 3.41x higher default risk.
The Business Impact: Profitable Financial Inclusion
By turning messy data into a world-class predictive model, we proved that unbanked does not mean unreliable; it just means you need a smarter algorithm.
- Unprecedented Accuracy: credit risk model achieved a staggering 96.9% Accuracy and an AUC of 0.991, providing the client with high confidence in their lending decisions.
- Risk Mitigation: We successfully crushed default risks, keeping them under 5% in a highly volatile emerging market.
- Scaling the Unscalable: Armed with a reliable credit risk engine, the client was able to safely and profitably scale their lease-to-own solar systems to over 8,000 households, transforming solar energy from a luxury into a scalable financial product.
We don't just find data where there is none; we find the story within the data.
If you want to know who is going to pay you back in a market with no credit scores, you don't need a bank. You need a firm that knows how to count cocoa beans.
