Double discrimination: Algorithmic amplification of gender bias in African fintech credit scoring—a 10-algorithm audit reveals 37% underfunding penalty against women-led SMEs
African women-run small and medium-sized companies (SMEs) need $42 billion more in capital each year. Even if they pay back their loans just as well as male-led enterprises, women may only access 7% of formal credit. This is the first research to look at gender bias in African fintech lending algorithms in depth. It achieves this by using both gender entrepreneurship theory and algorithmic fairness indicators. It analyzes 10 standard credit scoring models from conventional banks and fintechs in Nigeria, Kenya, and South Africa using 1,200 synthetic SME profiles that have the same financial fundamentals but distinct gender indications in ownership signals, sector coding, and networking patterns. The audit demonstrates that female entrepreneurs are systematically punished with a 37% underfunding penalty. This shows that AI algorithms change digital lending from a promised equalizer into an engine of prejudice. This is shown by closely measuring approval rates, interest spreads, and collateral demands. These numbers show that women are punished with hidden proxies, such as sector-based risk misclassification (for example, calling beauty services high-risk even though they are known to be profitable), network analysis that favors male-dominated affiliations, and linguistic bias against communal leadership language. These findings demonstrate that lawmakers need to act right now to make algorithms accountable. They also demonstrate that machine learning, which is supposed to be fair, instead makes human prejudices worse, creating self-perpetuating cycles of financial exclusion in Africa's digital lending environment.
Paper
Full text
Double discrimination: Algorithmic amplification of gender bias in African fintech credit scoring—a 10-algorithm audit reveals 37% underfunding penalty against women-led SMEs
Semantic Scholar · 2025
Abstract
African women-run small and medium-sized companies (SMEs) need $42 billion more in capital each year. Even if they pay back their loans just as well as male-led enterprises, women may only access 7% of formal credit. This is the first research to look at gender bias in African fintech lending algorithms in depth. It achieves this by using both gender entrepreneurship theory and algorithmic fairness indicators. It analyzes 10 standard credit scoring models from conventional banks and fintechs in Nigeria, Kenya, and South Africa using 1,200 synthetic SME profiles that have the same financial fundamentals but distinct gender indications in ownership signals, sector coding, and networking patterns. The audit demonstrates that female entrepreneurs are systematically punished with a 37% underfunding penalty. This shows that AI algorithms change digital lending from a promised equalizer into an engine of prejudice. This is shown by closely measuring approval rates, interest spreads, and collateral demands. These numbers show that women are punished with hidden proxies, such as sector-based risk misclassification (for example, calling beauty services high-risk even though they are known to be profitable), network analysis that favors male-dominated affiliations, and linguistic bias against communal leadership language. These findings demonstrate that lawmakers need to act right now to make algorithms accountable. They also demonstrate that machine learning, which is supposed to be fair, instead makes human prejudices worse, creating self-perpetuating cycles of financial exclusion in Africa's digital lending environment.