People are dying at home. Will they pay SHA, or pay for food?
In Kenya, a government promise to bring healthcare to all has collided with the quiet violence of a flawed algorithm. The Social Health Authority's AI-driven system, launched in October 2024 to cover millions of informal workers, instead systematically overcharges the poor while undercharging the wealthy — an outcome that was warned against, and chosen anyway. It is a story as old as reform itself: the machinery built to lift the vulnerable is calibrated, in the end, to protect the comfortable. What was sold as digital transformation has become, for millions, a barrier between illness and care.
- Millions of poor Kenyans are being assigned healthcare premiums that consume up to a fifth of their annual income, forcing an impossible choice between medical coverage, food, and shelter.
- People are being turned away from hospitals and dying at home because they cannot pay bills generated by an algorithm that was never accurate enough to be trusted.
- An investigation confirmed the system's core flaw: it consistently overestimates poor households' incomes and underestimates wealthy ones, with the government knowingly prioritizing the latter.
- The algorithm's foundation — proxy means testing — has documented failure rates of 82 to 90 percent in comparable programs across Indonesia, Rwanda, and beyond, yet Kenya deployed it over explicit warnings.
- Of more than 20 million people registered, only 5 million pay regularly, hospitals are reporting large deficits, and a former deputy president has predicted the entire system will collapse within months.
- What began as a flagship reform and a campaign promise of universal coverage is now widely described as a failed experiment — one that has left the people it was designed to help most exposed and most harmed.
When President William Ruto campaigned in 2023, he promised that no Kenyan would be left behind on healthcare. After taking office, he moved to deliver: in October 2024, his government launched the Social Health Authority, a digitally powered system designed to finally bring coverage to the 83 percent of Kenya's workforce in the informal economy — day labourers, street vendors, farmers — whose incomes had always slipped through the cracks of traditional insurance.
The system's engine was a machine learning algorithm tasked with estimating what each household could afford to pay. Government volunteers visited homes across the country, asking families about their roofing materials, their toilets, whether they owned a radio. Those details were fed into a formula that produced a number: the annual premium each family owed. Grace Amani, a mother of ten who became one of those volunteers, watched the results appear on her tablet screen in Nairobi's poorest neighbourhoods. Families already struggling to eat were told they owed between 10 and 20 percent of their meager incomes. Some could not pay. Others, critically ill, were turned away from hospitals. "People are dying at home," Amani said. "Will they pay SHA, or pay for food?"
An investigation by Africa Uncensored, Lighthouse Reports, and the Guardian found the algorithm was doing the opposite of what it promised. It overestimated the incomes of poor households and underestimated those of the wealthy — overcharging more than half of poor families in testing. Two farmers, for instance, were assessed at twice their actual income simply because they owned their home and had electricity. This was not a technical accident. Health economist David Khaoya, who advised the ministry, explained that the system's constraints forced a choice: accurately assess the poor, or accurately assess the rich. The government chose the wealthy. The poor, he noted, have little recourse to object.
The algorithm's underlying method — proxy means testing, a World Bank tool that estimates income from possessions rather than earnings — has been tested across the developing world and found to fail at extraordinary rates. An Indonesian program excluded 82 percent of those it was meant to serve. A Rwandan scheme had a 90 percent error rate. Before Kenya's system launched, the consultancy IDinsight warned the government in writing that the algorithm was flawed and inequitable, particularly for low-income households. The government deployed it anyway.
The consequences are now visible across the country. Of more than 20 million people registered, only 5 million pay regularly. Hospitals are reporting large deficits as reimbursements go unpaid. On social media, Kenyans have shared their stories in waves — premiums doubled overnight, a single mother posting "God have mercy on me" after her monthly contribution was set at 3,500 shillings. The chair of Kenya's Rural and Urban Private Hospitals Association now calls it "an experiment that has failed." A former deputy president predicted in March that SHA would collapse within six months. The promise of universal healthcare has become, for millions, a lottery that feels rigged against the very people it was meant to protect.
President William Ruto stood in a stadium in Kericho during his 2023 campaign and made a promise that resonated across Kenya: no citizen would be left behind. Healthcare, he said, would soon be affordable and accessible to everyone. When he took office, he moved quickly to deliver. In October 2024, his government launched a new Social Health Authority system, designed to replace Kenya's aging national insurance scheme and finally bring coverage to the 83 percent of the workforce laboring in the informal economy—the day labourers, street vendors, farmers, and others whose incomes don't fit neatly into traditional salary structures.
The system was billed as a digital transformation, powered by artificial intelligence. What it actually deployed was a predictive machine learning algorithm tasked with a deceptively simple job: estimate how much each Kenyan household could afford to pay for healthcare based on their circumstances. The algorithm would ask questions through government volunteers who visited homes across the country. What kind of toilet do you have? What's your roof made of? Do you own a radio? These details—pit latrines, iron sheets, the absence of possessions—would be fed into a formula that would spit out a number: the annual healthcare premium each family owed.
Grace Amani, a mother of ten, became one of those volunteers. She sits in the homes of Nairobi's poorest residents, helping them answer dozens of questions on a digital form. When the questionnaire is complete, a number appears. Families she visits—people struggling to put food on the table—are often told they owe between 10 and 20 percent of their meager annual income. Some are charged sums so high they cannot pay. Others, critically ill, are turned away from hospitals because they have not settled their accounts. "People are dying, people are suffering," Amani said. "People are dying at home. Many people have been unable to go to hospital. Will they pay SHA, or pay for food, or pay for the small house they live in?"
An investigation by Africa Uncensored, Lighthouse Reports, and the Guardian examined how the algorithm actually performed. The findings were stark: the system systematically overestimated the incomes of poor households while underestimating those of the wealthy. Two farmers, for instance, were told their income was twice what it actually was, based solely on the fact that they owned their house and had electricity. The audit tested the algorithm against thousands of real households and found it overcharged more than half of poor families. The wealthy, by contrast, were consistently undercharged.
This outcome was not accidental. David Khaoya, a health economist who advised Kenya's health ministry, explained the choice that was made. The system's constraints meant officials could either correctly assess poor households or correctly assess rich ones—but not both. The government chose to prioritize accurate evaluation of the wealthy. "If you identify a richer person as poor and therefore ask him to pay less, this person will never own up and say, 'I'm actually supposed to be paying more,'" Khaoya said. The poor, by contrast, have little recourse.
The algorithm's foundation is a decades-old World Bank tool called proxy means testing, or PMT—a method of estimating income based on possessions and life circumstances rather than actual earnings. It has been deployed across Africa, Asia, and Latin America, often as a condition of receiving loans. But researchers have found these systems simply do not work. An Indonesian poverty program tested by economist Stephen Kidd excluded 82 percent of the population it aimed to serve. A Rwandan scheme had a 90 percent error rate. Kenya's system appears to have similar problems.
Before the SHA system launched, an international data consultancy called IDinsight produced a report warning the government that the algorithm was flawed and "inequitable, particularly for low-income households." The report noted the system had too few data points from poor areas, was out of date given Kenya's recent economic shocks, and over-represented middle-income households. The government deployed it anyway. Of more than 20 million people registered, only 5 million regularly pay their premiums. Hospitals report large deficits as promised reimbursements remain unpaid. On social media, Kenyans have flooded comment sections with their stories: one person wrote of jumping from paying 500 Kenyan shillings to 1,030; a single mother posted "God have mercy on me" after her monthly contribution was set at 3,500 shillings. Dr. Brian Lishenga, chair of Kenya's Rural and Urban Private Hospitals Association, now calls the system "an experiment that has failed." A former deputy president predicted in March that SHA would collapse within six months. The promise of universal healthcare has instead become a lottery—one that feels rigged against those it was meant to help.
Citas Notables
People are dying, people are suffering. People are dying at home.— Grace Amani, government healthcare volunteer
This is an experiment that has failed. It's a really poor tool for identifying poor households.— Dr. Brian Lishenga, chair of Kenya's Rural and Urban Private Hospitals Association
La Conversación del Hearth Otra perspectiva de la historia
Why would a government knowingly choose an algorithm that overcharges the poor?
Because the alternative was worse for the wealthy. If you misidentify a rich person as poor, they'll never correct you. But poor people have nowhere to appeal.
So this wasn't a technical failure—it was a deliberate trade-off?
Yes. The health economist was explicit about it. They knew the system couldn't be accurate for both groups, so they picked accuracy for the people with power.
How did this get past the warning signs? There was a report before launch, right?
There was. IDinsight told them the system was inequitable and flawed. They launched it anyway. Sometimes the machinery of government just keeps moving.
What happens to someone who can't pay?
They don't get treatment. Or they get a hospital bill they can't afford. The system was supposed to expand access. Instead it's created a new barrier for the people who need help most.
Is this unique to Kenya?
No. These proxy means testing algorithms are everywhere—pushed by the World Bank, used across Africa and Asia. They fail consistently. But Kenya's version is particularly stark because the overcharging is so systematic.
What does Ruto say about it now?
The source doesn't say. But the system is hemorrhaging money, hospitals are in deficit, and only a quarter of registered people are actually paying. The promise from that stadium in Kericho looks very far away.