An AI system used to predict how much Kenyans can afford to pay for access to healthcare, has systemically driven up costs for the poor, an investigation has found.<\/p>\n
The healthcare system being rolled out across the country, a key electoral promise of President William Ruto, was launched in October 2024 and intended to replace Kenya\u2019s decades-old national insurance system.<\/p>\n
Billed as \u201caccelerating digital transformation<\/a>\u201d, it aimed to expand access to care to Kenya\u2019s large informal economy: the day labourers, hawkers, farmers and non-salaried workers that make up 83% of its workforce.<\/p>\n \u2018No Kenyan will be left behind,\u2019 William Ruto, Kenya\u2019s president, said during the 2023 election. Photograph: AFP\/Getty<\/p>\n \u201cNo Kenyan will be left behind,\u201d Ruto told a crowded stadium in Kericho during his 2023 presidential campaign, announcing that every citizen would soon have access to affordable healthcare.<\/p>\n But his solution has instead sparked protests and anger, as healthcare contributions for millions of people are now calculated via a formula described as \u201cflawed\u201d and which sources have said has almost no transparency.<\/p>\n That solution, which Ruto has described as AI-powered, does not rely on the recent advances in artificial intelligence which underpin large language models such as ChatGPT \u2013 instead it uses a predictive machine learning algorithm.<\/p>\n It now determines healthcare contributions for millions of people through a means-testing process.<\/p>\n Through months of investigation, reporters at Africa Uncensored<\/a>, in collaboration with Lighthouse Reports<\/a> and the Guardian, were able to obtain key details of this system and audit how it worked. The findings reveal how, from the start, it was systematically overcharging the poorest Kenyans, overestimating their incomes, while undercharging the wealthiest by underestimating their incomes.<\/p>\n Every day, Grace Amani* sits in people\u2019s homes to ask them questions from the odd to the intrusive. What type of toilet do you use? What is your roof made of? Do you own a radio?<\/p>\n She helps the occupants answer dozens of these questions \u2013 pit latrine, iron-sheet roof, no radio \u2013 on a digital questionnaire on their phones. People are often confused; some fear they are under investigation. When the form is complete, a number comes back as the algorithm calculates the sum the household must pay that year for public health insurance.<\/p>\n double quotation mark<\/p>\n People are dying, people are suffering. They thought it was something that would help <\/p><\/blockquote>\n SHA volunteer<\/p>\n The mother of 10 is also among those who claim the system is not working as it should and is punishing the least well-off.<\/p>\n The people Amani registers are some of the poorest in Nairobi, Kenya\u2019s capital, yet most are charged fees they cannot afford. She has watched families struggling to feed themselves charged a premium far beyond their means, many facing a sum of between 10% and 20% of meagre incomes.<\/p>\n Amani has also seen critically ill people who cannot get treatment because they have not been able to pay the amount the AI system says they should.<\/p>\n \u201cPeople are dying, people are suffering,\u201d she said.<\/p>\n A train runs through Kibera, Nairobi. Home to about 250,000 people, it is Africa\u2019s largest slum and one of the biggest in the world. Photograph: Donwilson Odhiambo\/Getty<\/p>\n The people she sees are exactly those the government promised would benefit most from the AI-driven health reforms. Those with the lowest incomes were supposed to be charged the minimum premium, or have their costs covered entirely. \u201cThey thought it was something that would help them,\u201d Amani said.<\/p>\n Since its launch, the Social Health<\/a> Authority (SHA) has been met with a barrage of criticism for misclassifying people, and setting unaffordable or incomprehensible premiums.<\/p>\n Kenyans without private insurance who do not pay their SHA premiums risk being turned away from health facilities or presented with steep hospital bills. For some, this has meant they can no longer access treatment. \u201cPeople are dying at home,\u201d Amani said. \u201cMany 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?\u201d<\/p>\n On social media, Kenyans have flooded comment sections with accounts of charges they cannot pay. \u201cFrom struggling to pay 500 Kenyan shillings [\u00a32.90] previously to being billed 1,030 Kenyan shillings,\u201d one wrote.<\/p>\n \u201cGod have mercy on me,\u201d wrote one single mother, after her monthly contribution was set at 3,500 Kenyan shillings.<\/p>\n David Khaoya, a health economist who advised Kenya\u2019s health ministry, said that when faced with the known flaws in the SHA\u2019s formula, a choice was made.<\/p>\n The system\u2019s constraints meant that it could either correctly assess poor households, or correctly assess rich ones. Khaoya said the government chose to prioritise accurately evaluating the wealthy, even if that meant overcharging the poor.<\/p>\n \u201cIf you identify a richer person as poor and therefore ask him to pay less, this person will never own up and say, \u2018I\u2019m actually supposed to be paying more,\u2019\u201d he said.<\/p>\n A patient gets weighed at a Kibera health centre. The new system aimed to expand the state\u2019s services to people who have historically gone uncounted. Photograph: Brian Otieno\/Global Fund<\/p>\n Kenya\u2019s algorithmic healthcare system is structured on a decades-old World Bank bugbear<\/a>: proxy means testing (PMT), a way of estimating the incomes of the poor based on their possessions and other life circumstances, such as how many children they have or whether they live alone.<\/p>\n PMT has been used in World Bank-funded programmes \u201call over Africa<\/a>, all over Asia and the Pacific\u201d, said Stephen Kidd, a development economist. It has often been set as a condition for a government to receive a loan.<\/p>\n In Kenya, this has meant deploying government volunteers such as Amani to households across the country to register their roofing materials, livestock and children \u2013 and feeding those details into an opaque algorithm to decide how much they earn and how much they must pay.<\/p>\n The audit tested the system against thousands of real households. For family after family, the system overestimated their means. For two farmers, their income was predicted as twice what it actually was 0 based on the fact that they have electricity and own their house.<\/p>\n Systems similar to the one built by SHA have been quickly spreading around the world in recent years \u2013 often pushed by the World Bank or other international donors.<\/p>\n double quotation mark<\/p>\n It\u2019s a really poor tool for identifying poor households \u2013 it\u2019s a great tool for helping the government run away from responsibility<\/p><\/blockquote>\n Dr Brian Lishenga<\/p>\n Across Africa, Asia and Latin America, PMT algorithms have become popular in determining which households are \u201cpoor enough\u201d to receive cash transfers, food subsidies and other benefits. These systems aim to expand the services of the state to people who have historically gone uncounted; the informal workforce whose inconsistent earnings do not fit neatly into income-based healthcare schemes.<\/p>\n But Kidd and other researchers have found that these systems simply do not work. In attempting to categorise a population as \u201cpoor\u201d or \u201cnot poor\u201d, most make significant errors. One poverty-targeted scheme in Indonesia that Kidd tested excluded 82% of the population it aimed to serve; another in Rwanda had an error of 90%.<\/p>\n Health volunteers carry out medical tests in Nairobi. Of 20 million people registered for the Social Health Authority, only 5 million regularly pay their premiums. Photograph: Sopa\/LightRocket\/Getty<\/p>\n In Kenya\u2019s case, the SHA system appears to overcharge more than half of poor households, according to the investigative audit by Africa Uncensored and Lighthouse. The incomes of higher-income households are underestimated.<\/p>\n There is not a single reason for these inaccuracies, said Kidd. Poverty<\/a> is a fluid category \u2013 and using factors such as an iron roof or a pit toilet to estimate a family\u2019s wealth is an intrinsically imprecise undertaking.<\/p>\n But means-testing algorithms such as Kenya\u2019s introduce a separate problem: they are opaque, and reduce a population\u2019s faith in government services.<\/p>\n \u201cIt feels like a lottery,\u201d said Kidd. \u201cThe lottery is not a great way of building trust.\u201d<\/p>\n In Kenya, the system has led to widespread frustration. Yet its failings appear to have been anticipated by a report, authored by the international data consultancy IDinsight, and shared with the government before the system was implemented.<\/p>\n A therapist teaches a mother how to provide physiotherapy in Nairobi\u2019s Mathare slum. Some have predicted that the SHA will collapse soon. Photograph: Tony Karumba\/AFP\/Getty Images<\/p>\n That report, obtained by reporters, found SHA\u2019s system was flawed and \u201cinequitable, particularly for low-income households\u201d. Its basis for determining wealth \u201cover-represents middle-income households and has very few data points from poverty pockets\u201d. It was also \u201cout-of-date with the current socioeconomic condition\u201d in Kenya given the \u201cmultiple economic shocks\u201d that had affected the country.<\/p>\n Despite this, Kenya deployed the SHA system anyway. Of more than 20 million people registered for SHA, only 5 million are regularly paying their premiums. Some hospitals are reporting large deficits as promised reimbursements from SHA remain unpaid.<\/p>\n