Kenya’s AI Health System Overcharges Poor, Undercharges Wealthy

A Digital Health Revolution with Unintended Consequences

Kenya‘s AI-driven healthcare reforms, introduced in October 2024 under President William Ruto‘s administration, aimed to expand access to healthcare by replacing the country’s decades-old national insurance system. However, an investigative audit by Africa Uncensored, Lighthouse Reports, and The Guardian exposed a major flaw: the algorithmic system charges the poorest Kenyans more while undercharging the wealthy, contradicting the government’s promise of fair healthcare access. This case highlights the dangers of deploying complex algorithms without proper oversight, revealing how well-intentioned digital reforms can widen economic gaps. The reforms, targeting Kenya‘s 83% informal workforce—day laborers, farmers, and hawkers—ended up deepening inequality, with the poorest households bearing the brunt of a faulty design.

The Flawed Foundation of Proxy Means Testing

“People are dying, people are suffering”

The SHA system uses proxy means testing to estimate household income based on factors like roofing materials, livestock ownership, and family size—criteria used in World Bank-funded programs across Africa, Asia, and the Pacific for decades. However, its effectiveness is questionable. Stephen Kidd, a development economist, found that PMT systems often exclude large portions of target populations: 82% in an Indonesian poverty-targeted scheme and 90% in a Rwandan program. These failures highlight a broader trend in international development: relying on opaque, data-poor algorithms to allocate resources, often with harmful results.

The IDinsight report, obtained by reporters, explicitly warned that SHA’s system was ‘flawed and inequitable, particularly for low-income households.’ The report criticized the system’s basis for determining wealth as ‘over-representing the wealthy and under-representing the poor,’ with only 16 percent of the poorest households being predicted correctly. This stark statistic shows the algorithm’s inability to capture the real socioeconomic realities of its target population, worsening the divide between the rich and poor.

The Human Toll of Algorithmic Errors

The consequences of these errors are clear. Grace Amani, a health volunteer in Nairobi’s Kibera slum, described families being charged premiums they can’t afford, forcing them to choose between food and healthcare. ‘People are dying, people are suffering,’ she said, recounting cases where critically ill patients couldn’t access treatment due to unpaid premiums. The system’s opacity makes the crisis worse. As Kidd explained, ‘It feels like a lottery’ for households trying to navigate SHA’s requirements. With only 5 million of the 20 million registered users regularly paying, hospitals face severe financial strain. Some facilities report deficits as promised reimbursements remain unpaid, creating a cycle of underfunding and reduced access.

The audit found the algorithm overestimates poor households’ incomes, charging premiums that exceed 10-20% of their meager earnings. Wealthier households are undercharged due to inflated income estimates, as seen in cases where two farmers were assessed as earning twice their actual income simply because they owned their homes and had electricity. This two-tiered approach, noted by health economist David Khaoya, prioritizes accuracy for the wealthy over fairness for the poor, deepening economic divides.

A Global Experiment in Algorithmic Social Governance

Kenya’s experience reflects a growing global trend of using algorithms to manage social welfare. Systems similar to SHA’s have been adopted in other developing nations, often with World Bank backing. However, these implementations frequently face the same challenges: data inaccuracies, algorithmic bias, and lack of transparency. Kenya’s case mirrors these failures, with the SHA system’s PMT approach failing to accurately reflect the realities of its poorest citizens.

Dr. Brian Lishenga, head of Kenya’s Rural and Urban Private Hospitals Association, called the system ‘a really poor tool for identifying poor households—it’s a great tool for helping the government run away from responsibility.’ This critique highlights the systemic risks of relying on opaque methodologies that obscure accountability. The World Bank’s role in promoting PMT systems is a key factor in this global trend. Despite warnings from experts like Kidd, who noted PMT systems ‘exclude large portions of target populations,’ the Bank continued funding such initiatives. This institutional inertia shows a broader issue: overreliance on algorithmic solutions to complex social problems without safeguards.

“a really poor tool for identifying poor households—it’s a great tool for helping the government run away from responsibility”

Reimagining Digital Governance in Public Health

Kenya’s AI-driven healthcare reforms offer critical lessons for future digital governance. First, any algorithmic system must prioritize transparency and accountability, ensuring the poorest aren’t penalized for systemic flaws. Second, more granular, real-time data collection methods are needed to better reflect poverty’s fluid nature. As the IDinsight report warned before SHA’s launch, the system’s methodology was ‘out-of-date with Kenya’s current socioeconomic condition,’ given the country’s ‘multiple economic shocks.’ Despite these warnings, the government proceeded, underscoring political and institutional inertia.

Former Health Minister Rigathi Gachagua predicted ‘SHA will collapse in another six months,’ reflecting deep skepticism about the system’s viability. This isn’t just speculation—it’s rooted in the system’s design flaws. SHA’s reliance on PMT, which has a documented history of exclusion and inaccuracy, creates a self-fulfilling prophecy of failure. The system’s inability to adapt to Kenya’s economic realities—like rising inflation and unemployment—means it’s ill-equipped to serve its intended beneficiaries.

After reading this, the reader understands Kenya’s AI-driven healthcare reforms aren’t just a local failure but a symptom of a global trend: overreliance on opaque algorithms to solve complex social problems. The case serves as a warning for nations modernizing social services. The government must now face the uncomfortable truth that its digital revolution has, in many ways, deepened the divide between the rich and poor. The path to equitable healthcare reform may lie not in algorithmic complexity, but in simpler, more inclusive approaches that truly serve all citizens. The Kenya case underscores the need for rigorous evaluation of algorithmic systems before deployment, emphasizing that digital governance must prioritize human needs over technical novelty.