Building an Alternative Credit Score from M-PESA Behavioral Data

#fintech #africa #python #mcp

FICO was invented in 1956 and took five decades of credit card data to become the global standard. That data simply doesn't exist for most Kenyan adults. But something else does: M-PESA.

35 million Kenyans use M-PESA. Every paybill payment, every savings deposit, every Fuliza loan — that's a behavioral financial record. The question mkopo-mcp asks is: can you build a 300–850 credit score from it?

The Research Basis

The core insight comes from two bodies of work:

Breza & Kinnan (2021): Mobile money adoption in India measurably improved credit access for previously unbanked households. Transaction regularity was a stronger predictor of repayment than income level.

World Bank Financial Inclusion Database (2022): In Sub-Saharan Africa, 57% of adults who use mobile money have never had a formal bank account. That's the population FICO was never designed to serve.

Central Bank of Kenya Prudential Guidelines (2022): CRBs in Kenya are required to report both positive and negative credit information. Fuliza (M-PESA overdraft) data is already flowing to TransUnion Kenya.

The Score Model

from mkopo_mcp import alternative_credit_score

result = alternative_credit_score(
    monthly_income_kes=45000,
    mpesa_transactions_monthly=28,
    avg_transaction_kes=1850,
    paybill_payments_monthly=4,
    fuliza_usage="occasional",
    savings_behavior="regular_deposits",
    phone_tenure_years=3.5,
    months_of_history=18
)
# → score: 647, tier: "FAIR", loan_capacity_kes: 85000

The score is decomposed into four signals:

Signal	Weight	Proxy For
Income regularity	30%	Capacity
Payment consistency	25%	Willingness
Savings behavior	25%	Financial discipline
Tenure + history	20%	Stability

Why Behavioral Signals Work

The intuition: a person who pays their Nairobi Water paybill on time, every month, for 18 months is demonstrating financial discipline that's directly relevant to loan repayment. It's not income — it's behavior.

This is structurally similar to how FICO evolved to include rent payment data (FICO 10T, 2020) and how Experian Boost works in the US — both extend credit scoring to non-traditional signals.

The difference is that in Kenya, those signals are already flowing through one system (M-PESA), not scattered across utilities and landlords.

What mkopo-mcp Doesn't Do

Trust integrity matters here: mkopo-mcp is DEMO infrastructure. It does not connect to real M-PESA transaction data. It models how a score would be computed if such data were available.

Production deployment requires: