Investigation Finds Kenya's AI Healthcare System Overcharges Poor Households

An investigation by Africa Uncensored, Lighthouse Reports, and the Guardian found that Kenya's AI-powered healthcare system systematically overcharges the poorest households by overestimating their incomes, while undercharging wealthier ones by underestimating theirs.

The system, managed by the Social Health Authority (SHA), was introduced in October 2024 to replace the country's previous national insurance framework. It targets Kenya's informal workforce, which comprises 83% of the workforce, including day laborers, hawkers, and farmers.

Government volunteers conduct household surveys asking about possessions such as toilet types, roof materials, and radio ownership. These details are input into a predictive machine learning algorithm that calculates annual healthcare premiums based on proxy means testing, a method estimating income from life circumstances.

The algorithm does not use large language models but relies on machine learning to perform means testing. An audit of thousands of real households showed the system overestimates incomes for more than half of poor households, leading to premiums between 10% and 20% of their actual incomes.

Residents in Nairobi's poorer areas have reported being charged amounts they cannot afford, resulting in denied access to treatment. For example, some individuals previously paying 500 Kenyan shillings now face bills of 1,030 Kenyan shillings, while a single mother was assessed at 3,500 Kenyan shillings monthly.

Grace Amani, a government volunteer registering households, stated that families struggle with these premiums and that some critically ill people cannot receive care. She added that people are dying at home due to inability to pay.

“People are dying, people are suffering,”

David Khaoya, a health economist who advised Kenya's health ministry, said the system prioritizes accurate assessments for wealthier households, even if it means overcharging the poor. He explained that richer individuals would not voluntarily pay more if undercharged.

Stephen Kidd, a development economist, noted that proxy means testing systems often exclude large portions of intended beneficiaries, with error rates up to 90% in some cases. He described such systems as opaque, reducing public trust. A pre-implementation report by IDinsight, shared with the government, found the system flawed and inequitable for low-income households, citing outdated data and over-representation of middle-income groups.

Despite this, the system was deployed.

Similar proxy means testing algorithms are used in programs across Africa, Asia, and Latin America, often supported by the World Bank. In Kenya, of over 20 million registered for SHA, only 5 million regularly pay premiums. The reforms were part of President William Ruto's 2023 campaign promise to provide affordable healthcare to all Kenyans.

During a 2023 rally in Kericho, Ruto stated that no Kenyan would be left behind.