What it Takes to Build AI Health Tools That Work in African Care Settings
This article synthesises key insights from our webinar, Building AI Health Tools That Work in African Settings, co-hosted with AI HealthBeat Podcast. The conversation explored what it takes to build AI-enabled healthcare solutions that solve important clinical challenges and get adopted in real care settings.
AI is increasingly showing up in African healthcare as the technology becomes more capable and more accessible - showing up in diagnostic tools, patient engagement platforms, and clinical decision support systems.
But technical capability doesn't always translate to real-world adoption and sustained use by health workers and patients.
To understand why this happens and explore how health innovators can build AI health tools that earn trust and adoption in Africa, we hosted a webinar featuring a healthtech founder - Steve Adudans, Founder, Gear Health, a health practitioner - Dr Stephen Ojwang, Consultant Radiologist & Head of Department, AAR Hospital; and Immanuel Momanyi, Head of Acceleration, Villgro Africa - who has been evaluating healthtech companies and making investment decisions.
Here are the key takeaways from the discussion.
The shift that changes how we design AI health products
The way we design AI health products has changed fundamentally. For a long time, the dominant question was technical: does the model perform? Can it detect, predict or classify? These questions still matter. But they are no longer sufficient.
When AI enters a clinical workflow, it does not simply automate a task. It redistributes attention, judgement, work, intervention, and accountability across the care system.
A tool that routes a diagnosis recommendation to a clinician has also decided who is responsible if that recommendation is wrong. A tool that sends appointment reminders to patients has assumed they have a phone, connectivity and the literacy to act on what they receive.
These design consequences must be planned for before deploying an AI-enabled health tool.
Builders should move from asking, "Can AI do this?" to “Can this be responsibly adopted in a real care setting?” Because value is created or lost at the moment of use.
1. Start with the care problem, not the technology
Before asking whether a solution fits the care environment, builders need to make sure they're solving the right problem.
The most technically impressive AI health tools can still fail for a reason that has nothing to do with the technology - the problem was never clearly defined to begin with.
"Adding technology to a poorly understood problem creates a more complicated problem - not a sophisticated solution." - Maureen Macharia, CEO, Spindle Design
Dr Steve Adudans, Founder of Gear Health, learned this firsthand. In the early days of building Gear Health, the team started with what the technology could do: a faster way for patients to order medicine. The app was built. But uptake was low not because the technology failed, but because the team hadn't clearly defined the problem they were solving.
They stepped back and mapped the patient journey from when someone felt unwell to when they received care. This helped them identify the friction - patients were uncertain about their conditions and what information to bring to an appointment. Then they rebuilt around that flow: patients could book appointments, receive preparation guidance, and ask questions about their condition virtually before arriving at a clinic.
Dr Adudans’ experience surfaced an important lesson: start with the decision, task, or moment in the care journey that needs to improve. Map the existing workflow, understand what AI will change - and for whom before building the product.
For builders in African healthcare, this means spending enough time in the care environment to understand the real problem.
2. Reduce burden at the point of care
After understanding the problem, the next question is whether the solution actually improves the experience of delivering and receiving care.
At the point of care, the quality of human interaction between a health worker and a patient is important.
"Doctors shouldn't be looking at a screen half the time they are with a patient. AI should reduce clinical burnout, not create digital burnout." - Dr. Steve Adudans, Founder, Gear Health
This reframes what a successful AI health tool looks like in practice. It does more than improve a clinical support outcome. It reduces administrative burden to give the clinician more presence with the patient - more time for examination and conversation that builds trust which determines whether a patient follows through the care journey.
This principle also surfaces a risk that is easy to overlook: introducing AI tools without designing for workflow fit can generate new burdens for clinicians rather than reducing existing ones. An additional screen or a separate platform that does not talk to the existing system is a small tax on a health worker's attention, can cause burnout and undermine adoption.
Spindle's framework for assessing whether an AI health tool can survive contact with a real care setting reflects this:
A tool that works technically but fails on workflow fit, or cannot be understood or trusted by the clinician will not earn adoption.
3. Design for what happens after the AI output
An AI output is not the end of the design problem. You need to consider what happens after an AI system flags a risk. Who sees the alert? Who follows up? What action should the clinician take? Where is the patient referred? Can the patient afford the next step? What happens if the recommended service is unavailable?
These questions matter because a recommendation only creates value when it connects to what happens next in the care process.
"Healthcare workers should be able to follow up on patients - those who are diabetic or hypertensive need medication and should be linked to pharmacies. There needs to be holistic care." - Dr. Stephen Ojwang, Consultant Radiologist & HOD, AAR Hospital
For builders, this means understanding where an AI output enters the existing care workflow and designing the actions, handoffs and information needed to move from insight to appropriate care.
4. Design for uncertainty and failure
Designing AI health tools for the full patient journey also means planning for when things go wrong. What happens when data is missing, connectivity drops, a model produces an uncertain recommendation, or a clinician disagrees with its output? Where does human judgement take over and what safeguards prevent an error from becoming a patient safety issue?
These scenarios should be part of the design from the beginning - not something addressed after deployment. Responsible products make failure visible, provide clear ways to intervene, and continue to work when the conditions around them are imperfect.
In practice, this means showing uncertainty clearly, making override easy, defining escalation pathways, and creating ways for clinicians to report errors without adding more burden to their workflow.
5. Build AI health tools for sustainability beyond the pilot
A working pilot is a good start, but shouldn’t be the goal.
"Demonstrate that your solution is accurate, useful, and safe. And that people have both the incentive and the ability to pay for it." - Immanuel Momanyi, Head of Acceleration, Villgro Africa
In healthcare, "building with the user in mind" is more complex than it sounds. There isn’t just just one user. A clinician uses the tool. A hospital administrator pays for it. A patient benefits from it. A regulator determines whether it enters the market. A founder who designs only for the most visible user, typically the clinician, will discover the others when the tool fails to gain institutional buy-in, or when a payer declines to fund it, or when the product fails to meet a regulatory requirement. Understanding all of these stakeholders is part of the design work.
For funders and investors, sustainability means adoption holds when the grant cycle ends. That means the solution has demonstrated it is worth paying for - and patients, institutions, insurers, or governments are paying for it.
Moving from pilot to practice
The sentiments shared by these four speakers point to an important lesson: moving from pilot to practice is not just an engineering challenge. It is a product design challenge, a systems design challenge and an implementation challenge.
We cannot control every condition within a care system. But we can understand those conditions, make our assumptions visible, design for failure as well as success, and build AI health tools that are usable, safe and trusted by the people who rely on them.
Watch the full webinar recording
Building or funding an AI-enabled healthcare solution? Reach out, let’s explore how to move beyond a technically sound model to a solution that works in real clinical environments and delivers care.