The current SHA algorithm relies on a flawed logic: using housing materials to guess a citizen's income. This results in massive overcharges for the most vulnerable. We have built a data-driven replacement that fixes this.
Don't take our word for it. Pick one of the real household profiles below — or enter your own numbers — and compare what you'd pay under the current algorithm versus the proposed fix. Two minutes. Your data never leaves your browser.
Try the Calculator →To assess the informal sector, SHA uses a Proxy Means Testing (PMT) algorithm. Because informal workers lack formal paystubs, the algorithm looks at their housing materials—like whether they have a tile roof, stone walls, or basic electricity—and uses that to estimate their wealth. It fails on the ground.
A stone wall adds KSh 120,000 to a household's wealth score. An elderly pensioner living with zero income in an inherited stone house is incorrectly classified as wealthy and charged high premiums.
The system values land purely by acreage without considering location or arability. A citizen with 5 acres of dry, non-productive land in Turkana is scored as wealthier than someone owning 1 acre of prime real estate in Nairobi.
The algorithm treats gross M-Pesa transaction volume as personal income. A community treasurer (Chama) handling KSh 80,000 a month in group funds is penalized as a high-income earner.
The algorithm looks at gross metrics but ignores the realities of urban survival. A KSh 20,000 income in Nairobi is largely consumed by high rent and transport, yet it is treated the exact same as KSh 20,000 in a rural village where costs are vastly lower.
Having access to electricity automatically adds KSh 80,000 to a wealth score. In modern Kenya, this punishes the vast majority of citizens for simply having basic infrastructure connected.
A Bodaboda motorcycle is a rider's sole source of daily income and is often heavily financed by debt. The algorithm blindly treats it as a luxury vehicle, adding KSh 150,000 to their wealth score.
The system treats a few goats kept for subsistence milk or dire emergencies as liquid cash wealth. It assumes a poor rural family with zero cash flow can simply sell their safety net to pay monthly health premiums.
The current system does not reveal how it arrives at your premium. Citizens receive a flat charge with zero mathematical explanation of how their assets or income proxies were tallied, preventing any appeals or corrections.
These 8 examples are just the surface. Our comprehensive audit uncovered dozens of critical algorithmic failures driving the 80% overprediction rate and causing the system to haemorrhage KSh 11 Billion in fraud.
We didn't just find the 18 flaws—we engineered the exact solution to fix all of them. The AGI Model abandons housing guesswork entirely. Instead, it relies on verifiable digital footprints to accurately calculate household ability-to-pay. It preserves the statutory 2.75% rate, meaning it can be deployed immediately without legislative amendment.
Housing Materials Excluded: The model entirely abandons roof type and wall materials as wealth proxies. Inheriting a stone house no longer arbitrarily spikes an elderly citizen's premium.
Productivity over Acreage: Acreage alone is discarded. The model evaluates commercial agricultural output and KRA-declared revenue, ensuring dry ASAL land isn't conflated with prime fertile real estate.
80% Fiduciary Exemption: The system actively identifies group and community treasurers (Chamas) and applies a massive 80% exemption to their gross M-Pesa throughput, taxing only personal consumption.
3-Tier Cost of Living Deductions: Calculates an "Adjusted" Gross Income by systematically deducting urban rent and transport costs. KSh 20K in Nairobi is heavily adjusted downwards compared to KSh 20K in a rural village.
Decoupling Infrastructure: Having basic utility connections like electricity or piped water is no longer taxed as a luxury asset, protecting citizens who simply have access to basic modern infrastructure.
50% Commercial Exemption: Bodaboda motorcycles and similar debt-financed equipment are automatically recognized as essential operational assets, not luxury wealth, preventing immense premium spikes.
Commercial vs. Subsistence Farming: The model differentiates between commercial livestock sales (via triangulated market data) and subsistence farming, ensuring emergency safety nets aren't treated as liquid cash.
Transparent SHAP Receipts: Every citizen receives a plain-language deduction receipt detailing exactly how their premium was calculated, down to the shilling. Full transparency prevents "Error by Design" and allows for verifiable appeals.
We ran representative Kenyan households through both the current SHA PMT algorithm and our proposed AGI Model.
This is a fully functional, mathematically sound prototype. By fixing the core logic of the means-testing algorithm, we can increase voluntary compliance, lower premiums for the poorest, and ultimately raise net revenue for the health system.
We are not asking for immediate policy adoption.
We are asking for an introduction. We need to get this prototype in front of the right technical stakeholders within the innovation department, the Ministry of Health, or KIPPRA who can rigorously review this model and help guide it from a prototype into a viable policy alternative.
If you work in human rights, constitutional law, or investigative journalism: everything here — the methodology, the code, the data — is public. Cite it, test it, or take it further than we did. If you find something we got wrong, that's useful too.
If you're a citizen and the numbers above got your attention, sharing this is the most useful thing you can do. If you work in policy or health financing and can make that introduction — pmkaulani@gmail.com.