The Promise and the Problem
In 2024 President Ruto introduced a plan to broaden healthcare access through a newly designed digital system. The government replaced parts of the old insurance arrangement with a predictive, machine learning approach to set household health contributions. The stated aim was to target subsidies better and expand coverage. Instead the rollout triggered widespread backlash as many low-income households were charged higher contributions and some were denied care because they could not pay.
How the System Determined Contributions and What Proxy Means Testing Is
The system used a form of predictive means testing known as Proxy Means Testing, or PMT. Rather than using direct income measures, PMT estimates household economic status from observable proxies such as ownership of assets, housing characteristics and local indicators. A machine learning model mapped those proxies to a score that determined contribution levels.
How the System Failed the Poorest
The model systematically misclassified many of the poorest households as wealthier, raising their required payments. At the same time, some wealthier families were scored lower and asked to contribute less. The outcome was regressive: the greatest burden fell on those with the least ability to pay. Reports describe people unable to afford treatment, delayed care and rising out of pocket expenses for basic services.
Ignored Red Flags
An independent IDinsight report flagged these flaws before national implementation, showing PMT would be less accurate for the lowest-income groups and risk inequitable outcomes. Those warnings were not acted on at scale, which meant a predictable harm was introduced into a public service affecting millions.
Critical Lessons for AI in Healthcare
This case underlines several lessons for any health AI project: make data quality and representativeness a priority; test models on the specific populations they will affect; provide transparent methods and independent audits; build accessible appeal mechanisms and human oversight; and align incentives so safety and equity are measured before deployment. When healthcare access is at stake, algorithmic tools must be subject to strong governance and clear accountability.
