Building a Credit Scoring Engine for Rwanda's Unbanked Population

Traditional credit scoring excludes 83% of Rwandan adults. Here's how alternative data and machine learning are closing the gap.

The Credit Gap in Rwanda

Rwanda has made extraordinary strides in financial inclusion — mobile money penetration exceeds 80%. Yet formal credit remains out of reach for the majority. Banks rely on payslips, collateral, and credit history. Most Rwandans have none of these.

Alternative credit scoring uses the data people do generate: mobile money transaction patterns, airtime top-up frequency, utility bill payments, and social graph signals. Combined with gradient-boosted ML models, these signals can predict default risk with accuracy comparable to — and in some segments exceeding — traditional FICO-style scores.

Our Approach

GetRwanda built a modular credit scoring API for a Kigali-based microfinance institution. The pipeline ingests anonymised MTN MoMo transaction data, applies feature engineering across 90-day windows, and serves a risk score in under 200ms.

In a six-month pilot, the institution increased loan approval rates by 41% while maintaining its non-performing loan ratio below 4% — better than the industry average.

Responsible AI in Lending

We apply fairness constraints at training time to prevent proxy discrimination by location or phone model. All scoring decisions are explainable via SHAP values, allowing loan officers to understand and override the model when necessary.