Credit scoring — FICO, CRB Kenya, alternative data — Credit Module 4

A credit score is a single number that compresses years of borrower behaviour into a predictor of future repayment. The maths is statistical: regress historical defaults against borrower characteristics, find the features that predict default, weight them, and produce a number. FICO popularised it in the US in the 1980s; the same methodology now powers credit decisions globally.

What FICO actually weighs

text

FICO score components (300–850 scale):
  Payment history          35%   Have past payments been on time?
  Amounts owed             30%   How much of available credit is used?
  Length of credit history 15%   How long have accounts been open?
  New credit               10%   Recent applications and new accounts?
  Credit mix               10%   Variety: cards, mortgage, installment
  ──────
  100%

Thresholds:
  300–579   Poor       Most lenders decline
  580–669   Fair       Subprime rates
  670–739   Good       Standard rates
  740–799   Very good  Best rates from most lenders
  800–850   Exceptional Best rates anywhere

The FICO algorithm has been refined since 1989 but the weightings are stable. Payment history dominates because it's the strongest predictor of future payment behaviour.

Credit bureaus in Kenya

Metropol Corporation: largest CRB by data coverage in Kenya. Produces a Credit Reference Bureau (CRB) score on a 200–900 scale.
TransUnion (Kenya): subsidiary of the US giant; major SACCO and bank coverage.
Creditinfo: smaller; specialises in microfinance and mobile-money data.
All three are licensed by the Central Bank of Kenya under the Credit Reference Bureau Regulations 2020.

The 'CRB blacklist' problem

Until 2020 reforms, a single missed payment of any size — including KES 100 — could 'blacklist' a Kenyan borrower with a CRB for years, locking them out of formal credit entirely. The 2020 amendments required minimum thresholds (KES 1,000) and proper notice, but the cultural memory remains. Many Kenyan borrowers still avoid formal credit out of fear of CRB listing. Lenders who explain CRB reporting clearly at origination earn enormous trust.

Alternative data — mobile money as credit signal

M-Shwari (2012) was the first product to demonstrate that mobile-money transaction patterns are powerful credit signals. The model: 6 months of M-Pesa inflows + outflows + balance variability + counterparty diversity predicts repayment of a microloan with accuracy comparable to a traditional credit bureau. The implications are profound for emerging markets:

Thin-file or no-file borrowers (the majority in many African markets) become scorable.
Lending decisions get made in seconds instead of days.
Underwriting cost per loan falls from $50+ to under $1 — making KES 500 loans economical.
Default rates on these algorithmically-scored products are typically 5–10% — comparable to or better than human-underwritten unsecured personal loans.

The dark side of algorithmic scoring

Algorithmic scoring's strength is speed and scale. Its weaknesses are: (1) opacity — borrowers don't understand why they were declined; (2) bias amplification — if historical data encodes discrimination, the model reproduces it; (3) feedback loops — declined borrowers can't build the credit history that would qualify them later; (4) data extraction — borrowers' transaction data is the lender's asset, with little compensation flowing to the borrower. These are open governance questions in 2026, not solved problems.

Exercise

A Nairobi-based startup builds a micro-lending app. They have three options for scoring: (a) buy CRB Kenya scores from Metropol; (b) build their own model from M-Pesa transaction data they obtain via API; (c) use a combination. What are the tradeoffs? Which would you recommend for a first product?