The promise was a digital savior for a collapsing medical system. In 2024, the Kenyan government launched the Social Health Authority (SHA), a massive overhaul designed to replace the aging National Hospital Insurance Fund (NHIF). At the heart of this transition sits a predictive algorithm designed to calculate the financial worth of every citizen. It was pitched as a tool for equity, a way to ensure the wealthy paid their share while the poor received a safety net.
But for the millions of Kenyans living in the informal economy, the reality is a predatory mathematical error. Instead of identifying the vulnerable, the SHA’s "proxy means testing" algorithm is systematically overestimating the income of the poorest households, trapping them in a cycle of debt or forcing them to abandon medical care entirely.
The Ghost in the Machine
The mechanism at play is a machine learning model that relies on observable household attributes. When a Kenyan registers for the new Social Health Insurance Fund (SHIF), they are not asked for a bank statement—most do not have one. Instead, the system asks about the material of their roof, the type of toilet they use, the number of livestock they own, and their household size.
The algorithm then cross-references these answers against a 2020 household survey to "predict" an annual income. This prediction is the final word. A percentage of this theoretical income is automatically set as the household’s mandatory health insurance premium.
The fundamental flaw is that the algorithm cannot see the volatility of poverty. A family might own a cow or live under a corrugated iron roof, but that does not equate to a steady monthly cash flow. If a drought kills the cow or a casual laborer loses their contract, the algorithm doesn’t care. The premium remains fixed, based on a snapshot of assets that may no longer provide liquidity.
Loading the Burden onto the Vulnerable
In neighborhoods like Huruma and Kibera, the disconnect between the digital prediction and the physical reality is stark. Investigative data reveals that the poorest Kenyans are frequently being assigned premiums that consume 10% to 20% of their actual monthly earnings. In contrast, wealthier citizens with stable, formal salaries often find their contributions capped at much lower effective rates.
Consider the case of a single mother in Nairobi’s outskirts. After the system processed her "assets"—basic household items and a small rental space—it assigned her a monthly premium of 3,500 Kenyan shillings. For a woman earning less than 15,000 shillings a month in the informal sector, this is not an insurance premium. It is an eviction notice.
When these individuals attempt to appeal the automated decision, they enter a secondary bureaucratic nightmare. The automated appeal lines often return denials without explanation. There is no human auditor to explain why the machine thinks a family is "richer" than they are. The "black box" of the SHA algorithm has replaced the local knowledge of community health workers who once understood the nuance of a family’s financial distress.
The Two Tier Reality
While the algorithm squeezes the informal sector, the government has simultaneously protected its own. The Public Officers Medical Scheme Fund (POMSF) exists as a separate, more generous pot of money for civil servants. These officials draw from a taxpayer-funded pool that provides comprehensive coverage, only falling back on the standard SHIF benefits once their premium funds are exhausted.
This creates a brutal irony. The very people designing and implementing the algorithmic "means testing" for the masses are themselves insulated from its errors. They have secured a "comprehensive" tier of care while the rest of the population is left to the mercy of a machine that treats a tin roof as a sign of middle-class wealth.
The technical breakdown of the SHA rollout also highlights a massive data gap. The training data for the current model is based on 2020 statistics—data collected before the massive inflation spikes and economic shifts of the mid-2020s. By using outdated "ground truth" to train a predictive model for 2026, the government has guaranteed that the error rates will be highest among those whose lives are most impacted by economic instability.
A System Built to Fail the User
The Digital Health Act of 2023 was supposed to provide the legal guardrails for this transition. It spoke of data sovereignty and interoperability. However, the rush to migrate millions of records into the new Digital Health Agency (DHA) infrastructure has prioritized administrative speed over accuracy.
The gatekeeping mechanism is another friction point. Every household is now mapped to a Primary Care Network (PCN). This "hub and spoke" model works for a stationary population, but Kenya’s economy thrives on mobility. Truck drivers, construction "fundis," and domestic workers who move between counties for work find themselves locked out of care because they are physically distant from their assigned network. To the algorithm, they are a data point at a fixed coordinate. In reality, they are mobile workers who need a flexible system.
The High Price of Efficiency
The Kenyan government argues that AI is the only way to scale healthcare to 50 million people without an army of manual auditors. Efficiency is the justification for the automation. But when efficiency is prioritized over empathy, the result is a system that views the poor not as patients to be treated, but as a revenue source to be optimized.
The systemic overcharging isn't just a glitch; it is a feature of a model that lacks real-time economic feedback. Without a massive overhaul of how the SHA calculates "proxy" income—including the integration of real-time social welfare data and a transparent, human-led appeals process—the Digital Health Act will be remembered as the moment Kenya’s healthcare became a debt trap.
If a mother has to choose between her child’s antibiotic and the digital premium required to see the doctor, the system has already failed. The machine is running perfectly, but the patients are staying home.
