Evidence and deployment

What happened when it ran on real loan applications

A 60-day production pilot with Jiweze Improved Productivity Ltd in Kiambu County, Kenya, operating alongside the lender’s existing process.

The pilot at a glance

Historical pilot · Kenya · 60 days

previously unbanked borrowers approved

loan applications processed
average processing time
auto-approval rate
observed default rate over 60 days
infrastructure uptime

These figures relate to a historical deployment in Kenya using a predecessor engine. They do not constitute validation of the current engine.

How the pilot was conducted

  1. In parallel with the lender

    The 60-day Jiweze pilot operated alongside the lender’s existing system. Jiweze retained control of lending decisions throughout the evaluation.

  2. Real loan applications

    The predecessor scoring engine processed real loan applications in Kiambu County, Kenya, in a lending context with an average loan size of approximately US$40.

  3. Beyond the initial evaluation

    The engagement subsequently developed into a strategic partnership with Jiweze, extending the relationship beyond the initial pilot.

Pilot operating results

Location
Kiambu County, Kenya
Manual review rate
30%
Historical decision alignment
84% agreement with Jiweze’s historical approval decisions
Peak daily volume
120 applications
Average loan size
Approximately US$40
Women applicants
Approximately 43%

The broader opportunity

40%
of adults in Sub-Saharan Africa held a mobile-money account in 2024, up from 27% in 2021
24%
of adults in low- and middle-income economies borrowed formally in 2024

Sources: AfriScore × Jiweze pilot report, 2025; World Bank Global Findex 2025. The default rate reflects observed outcomes within a 60-day window. Agreement with historical lending decisions is not equivalent to predictive accuracy or proof of long-term credit performance.

Understanding the evidence

A limited observation period

The reported default rate reflects defaults observed during the 60-day pilot window. It should not be interpreted as a lifetime default rate or definitive measure of portfolio risk. Longer repayment cycles and extended observation are required.

Historical decision agreement is not model accuracy

The reported 84% measures agreement between AfriScore assessments and Jiweze’s historical approval decisions. Historical decisions are a reference point, not ground truth. The figure does not establish predictive accuracy, default prediction performance or causal impact.

Results are market-specific

The results were observed in Kenya using a predecessor engine and a particular lending environment. They do not establish the performance of the current engine in CEMAC. Local data coverage, borrower behaviour, lending practices and market conditions require independent validation.

From historical results to local validation

The Kenyan pilot provides an initial operational reference, not a performance guarantee for the current engine. Our next deployment focus is CEMAC. With a pilot institution, the current engine will be evaluated using authorized local data through a controlled process covering data coverage, score distribution, ranking performance, probability calibration, robustness, historical decision comparison and repayment outcomes as they mature.