From the newsletter
Caroline Mbindyo, CEO of Amref Health Innovations in Kenya, says artificial intelligence has great potential to strengthen health systems in Africa, especially in last-mile healthcare delivery. However, one of the biggest setbacks in deploying these technologies is focusing on developing the technology without understanding the problem it is meant to solve.
“At Amref, we do not begin by asking how AI will help. We begin by asking what the priority problems in communities are, and then consider whether AI can help solve them,” says Mbindyo. “Many AI tools are built elsewhere and brought into African countries looking for problems to solve. That approach does not make sense.”
She argues that communities must be involved in developing AI solutions because health challenges differ across regions. The way a maternal health problem appears in one county may look very different in another, making it unrealistic to expect a single technology to work everywhere.
More details
Q: How can AI realistically improve last-mile healthcare delivery in rural Africa?
AI is the definitive technology of our time, but we must remember that it is still an innovation, and innovation is supposed to solve problems. At Amref, we do not begin by asking how AI will help. We begin by asking what the priority problems in communities are, and then we consider whether there is an opportunity for AI to help solve them.
Many AI tools are built elsewhere and then brought into African countries looking for problems to solve. That approach does not make sense. The people who understand the problem best are those experiencing it, and they must be involved in defining both the problem and the solution.
So the starting point has to be communities themselves. We work with people to define the priority issues they face and then design solutions with them. The maternal health challenges in one county in Kenya may look very different from those in another part of the country or elsewhere in Africa. Communities need to define what the problem is for them and what they would want a solution to look like.
There is also a broader conversation about whether AI should replace certain roles. For example, if it were technically possible to replace a nurse with a digital device during pregnancy care, the question is not only whether it can be done, but whether it should be done. A pregnant woman in a rural part of Africa may still want to interact with a nurse in a health facility rather than a digital assistant.
As Africans, we should have the agency to define what AI should look like in our context. We should be producers of these solutions, not simply consumers of whatever technologies arrive from elsewhere.
Q: There is also concern that AI tools could replace some tasks currently performed by health workers. How should that be approached?
Across sectors we are seeing AI used to perform tasks that humans previously did. For example, tools can now draft speeches when given instructions. In healthcare we are seeing similar shifts, where tasks that nurses used to perform may now be done by AI tools.
We are also seeing devices such as AI-enabled stethoscopes where a button is pressed and the device interprets the result. The question we must ask is what happens if we adopt these tools wholesale. Are we unintentionally de-skilling health workers?
If a health worker previously needed to understand a clinical process but now relies entirely on a tool, how will they verify whether the tool’s output is correct? These are the trade-offs we must consider carefully before adopting new technologies. We need to think critically about what we gain and what we might lose.
Q: What specific AI use cases is Amref prioritising right now?
When we examine the health system, we first look at what is already in place so that we can build on existing systems rather than starting from scratch.
In Kenya, the Ministry of Health has rolled out the electronic Community Health Information System (eCHIS). Community health workers use this digital tool to provide services and collect data from households across the country.
The question we ask is how that data can be used more effectively at every level of the health system. A community health worker collects data during household visits. Could an AI tool help them interpret that data and identify signals that support decision-making? For example, should they prioritise visiting one household over another?
Supervisors receive data from multiple community health workers. Instead of reviewing each report individually, an AI-supported dashboard could summarise key trends or highlight areas of concern. This allows supervisors to focus their attention where it is most needed and ask the right questions.
At the health facility level, staff receive aggregated data from community health promoters in their catchment areas. AI tools could help interpret this information to guide decisions on issues such as immunisation coverage or where additional interventions may be needed.
Ultimately, this type of system can support decision-making from the community level all the way up to the national policy level.
Currently we are building an AI assistant linked to the electronic community health information system. It supports supervisors in managing and supporting community health promoters.
The tool is being deployed in Machakos County across the entire county. We built it together with community health workers, supervisors, the county health team and the national Ministry of Health.
We spent time with community health promoters to understand how they use the system and the challenges they encounter. We asked what type of support they would find useful and used that immersion process to co-design the tool.
As we deploy it, we continue to evaluate how it is used. We analyse backend data to understand whether the tool is being adopted and whether it changes decision-making compared to previous practices. That evidence helps us determine the next steps.
The tool does not yet have a formal name because it is still a proof of concept. For now we refer to it as an AI assistant for the electronic Community Health Information System.
Q: How does Amref ensure AI tools are clinically safe, evidence-based and validated before deploying them in communities?
Although AI itself has existed for some time, the ways it is currently being used are relatively new, especially in the African context. Even in high-income countries where these technologies have been used more extensively, evidence is still emerging and earlier assumptions are often being challenged by new data.
For us, a key priority is generating evidence that is relevant to African settings. We do this not only within Amref but together with other stakeholders.
We work with institutions such as the Digital Health Authority and the Ministry of Health. We also collaborate with communities where solutions are deployed. Within Amref we have Amref International University, which functions as an academic and research institution.
Through research and evaluation we examine how these tools work in practice. That allows us to bring evidence when we say what we are observing and what we propose should happen next.
Q: With AI regulations in Africa still evolving, how is Amref navigating compliance across multiple countries?
There are already several regulatory frameworks that apply to digital health and AI. In Kenya, for example, the Ministry of Health and the Ministry of ICT have policies and guidelines that govern digital technologies.
Kenya also has a Data Commissioner responsible for issues related to data governance and protection.
Countries across Africa are at different stages of regulatory maturity. At the continental level, the African Union has developed frameworks for data governance. There are also global frameworks from organisations such as the United Nations and the Organisation for Economic Co-operation and Development. In the health sector specifically, there are international standards such as Health Insurance Portability and Accountability Act and Fast Healthcare Interoperability Resources.
So there is a multiplicity of frameworks, guidelines and standards. The key question is how countries create an enabling infrastructure that allows new technologies to be introduced responsibly.
Kenya is making significant progress in this area. Through the Digital Health Authority and other partners, the country is establishing processes that developers must follow before introducing digital tools into the health system.
This includes standards and certification processes, particularly for clinical decision-support tools that are AI-enabled. However, these efforts are still in early stages and the regulatory environment is continuing to evolve.
Q: How do you address data protection gaps when deploying AI solutions?
Health data is extremely sensitive, and the use of AI often relies on analysing large amounts of data to identify patterns and trends.
In Kenya, there has been debate about issues such as bilateral agreements involving data sharing. These debates are important because they raise questions about what countries may lose or gain when negotiating such agreements.
When negotiations happen from a position with little leverage, it becomes easier to compromise on issues such as data sovereignty. That is why these matters are increasingly being challenged in court and debated publicly.
At the same time, the Office of the Data Commissioner has introduced safeguards. For example, anonymised and aggregated data may be considered safe for sharing, while identifiable data requires stronger protections.
Consent is also central. The Data Commissioner has emphasised that individuals must give consent before their data is used, even in cases that do not involve health data directly, such as the use of personal photographs.
Another principle is that organisations should collect only the data they are authorised to collect.
However, there is always a tension between regulation and innovation. The challenge is balancing the need to protect individuals while still allowing innovation to move forward. Navigating that balance is not straightforward, and it is something policymakers and organisations continue to work through.