AI in African healthcare: what 18 months of published deployments actually show

The honest version of any African healthcare AI essay starts with a concession. The published evidence is thin, and most of it comes from a handful of well-resourced institutions and a handful of vendors with a reason to publish. We have read the papers that exist, watched the deployments that outlasted their funding rounds, and worked alongside hospitals operating the technology in conditions the original research did not cover. This whitepaper is the institutional version of that reading list. It tries to be precise about what the existing evidence supports, where it stops, and what changes when the system has to run on a Tuesday morning in a busy outpatient department.

There are five lessons in the published material that hold up across deployments. They are also the five lessons we keep arriving at independently in client work, which we mention because convergence between literature and field experience is itself useful evidence.

The first is that data quality is the bottleneck, not the algorithm. The Owoyemi review of African healthcare AI was clear about this in 2020, and nothing in the years since has weakened the claim ^[1]. Schwalbe and Wahl's Lancet piece on AI in global health made the same argument from a different direction ^[2]. Duplicate records, inconsistent diagnostic coding, free-text where structured fields existed, missing demographic data, timestamp drift between systems, this is the universal background. The IFC's 2023 portfolio review on AI in emerging-market healthcare lands in the same place ^[10]. We spend more engineering time on entity resolution and pipeline integrity than on modelling, every time, and the published evidence suggests we are not unusual in this.

The second is that clinician trust takes time and structure. The Babyl Rwanda partnership, which paired Babylon's symptom-checker AI with the country's primary-care system, had reached over two million registered users before the parent company unwound ^[4]. The lesson there is mixed. Reach is not adoption, and a chatbot that triages a worried patient is not the same operating problem as a triage tool that influences a clinician's decision. Vula Mobile in South Africa took a quieter approach, building a specialist-referral application that shifted general practitioners' practice incrementally, and is now in regional use across the SADC corridor ^[11]. The published deployments that durably moved clinical practice all involved long pre-deployment validation cycles, named local champions, and explicit scoping of the model's role.

The third is that infrastructure shapes architecture. Power reliability, connectivity, and on-prem GPU availability change which models can be served and where. Mollura and colleagues made this argument specifically for radiology in 2020, and it has aged well ^[15]. The radiology AI tools that have actually scaled, qXR and CAD4TB foremost among them, were designed for offline-friendly inference on commodity hardware. They are now used in TB screening programmes across the continent under the WHO's 2021 conditional recommendation that computer-aided detection software may substitute for human readers in adult TB screening ^[7]. That recommendation was preceded by multi-site validation including African data ^[6]. Zipline's drone delivery network, which integrates routing and demand forecasting into national medical supply chains in Rwanda, Ghana, Nigeria, and Cote d'Ivoire, made similar architectural concessions to operate in the actual environment ^[8].

The fourth is that consent and governance vary by jurisdiction in ways that matter operationally. Nigeria's Data Protection Act 2023 ^[14] and South Africa's POPIA ^[13] are not interchangeable. Cross-border data movement that is straightforward under one is sensitive under the other. The work of negotiating data use agreements per network, with separate retraining permissions and audit windows, is not legal overhead. It is the design constraint. The result is that what looks externally like one programme often runs as multiple parallel data lifecycles, which is what a careful deployment looks like in practice.

The fifth is that AI in African healthcare works when it is treated as a capability programme, not a software procurement. Brookings made this argument in 2023 from the policy side ^[9]. We make it from the delivery side. Domain-focused groups like MinoHealth AI in Ghana ^[16] and clinic-network operators like Ilara Health in Kenya ^[12] are not unusually clever at modelling. They are unusually disciplined about workflow integration, retraining cadence, and the slow work of clinician engagement. The institutions that bought a model and called it strategy mostly do not feature in the longer-running case studies because the deployments stopped working.

We do not think these five lessons generalise to every African clinical context. The published evidence is concentrated in a few countries, a few hospitals within those countries, and a few clinical use cases within those hospitals. What we believe with reasonable confidence is that the published material and our own practitioner experience point in the same direction, and that direction is operational rather than algorithmic. We will be revisiting this in 2027 with what we hope is a wider evidence base.

Coderex advises ministries of health, hospital systems, and AI vendors on how to design healthcare AI programmes that survive contact with live operations: data infrastructure first, structured clinician trust cycles, architectures that match the power and connectivity reality, and consent regimes designed around the jurisdictions actually involved.

Expect the WHO conditional recommendation on TB screening to be extended to at least one additional clinical use case before 2028 as the multi-site validation literature matures. Expect the first published longitudinal study of a continental healthcare AI deployment past the five-year mark to land in 2027 or 2028, repricing how durably any of these systems shift clinical outcomes. Expect at least one African data-protection authority to issue a formal finding on retraining and cross-institutional model deployment within the next 18 months, which will reshape how these programmes are structured.

---

Methodology note: This whitepaper is a structured reading of African healthcare AI literature published from 2019 to 2024, cross-referenced with our practitioner observations from health-sector engagements in West, East, and Southern Africa. We have not run a primary survey for this piece. Where we cite vendor or operator self-reported figures (Babyl Rwanda registered users, Vula Mobile workforce reach), this is flagged. Sample sizes in the published literature remain small relative to the size of the question.

References

16 sources, all verified at the time of writing

1. [1]Owoyemi, Ayomide et al., 2020. Artificial Intelligence for Healthcare in Africa. Frontiers in Digital Health, Vol. 2, Article 6. https://www.frontiersin.org/articles/10.3389/fdgth.2020.00006/full.
2. [2]Schwalbe, Nina and Wahl, Brian, 2020. Artificial Intelligence and the Future of Global Health. The Lancet, Vol. 395, Issue 10236, pp. 1579-1586. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30226-9/fulltext.
3. [3]Murgia, Madhumita, 2023. How Babylon Health flopped after taking on the NHS. Financial Times. https://www.ft.com/content/29991366-2d4a-44e7-8b94-e0a05c8e3a3b.
4. [4]Bright, Jake, 2022. Babylon Health partners with Rwanda to scale digital health services. TechCrunch. https://www.mobihealthnews.com/news/emea/babylon-inks-10-year-partnership-rwandan-government.
5. [5]World Health Organization Regional Office for Africa, 2023. Leveraging Digital Health for Improved Service Delivery: WHO Africa Regional Strategy. WHO Africa. https://www.afro.who.int/publications/leveraging-digital-health.
6. [6]Qin, Zhi Zhen et al., 2019. Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Nature Scientific Reports, Vol. 9, Article 15000. https://www.nature.com/articles/s41598-019-51503-3.
7. [7]World Health Organization, 2021. WHO consolidated guidelines on tuberculosis. Module 2: Screening, systematic screening for tuberculosis disease. World Health Organization. https://www.who.int/publications/i/item/9789240022676.
8. [8]McCool, Joshua, 2022. Zipline expands medical delivery operations across Africa. Reuters. https://www.reuters.com/business/healthcare-pharmaceuticals/zipline-expands-medical-deliveries-2022-09/.
9. [9]Brookings Institution and Signe, Landry, 2023. Artificial intelligence and Africa: How AI can transform health, agriculture, and education. Brookings Africa Growth Initiative. https://www.brookings.edu/articles/leveraging-ai-and-emerging-technologies-to-unlock-africas-potential/.
10. [10]International Finance Corporation, 2023. Artificial Intelligence in Healthcare: Use Cases for Emerging Markets. IFC, World Bank Group. https://www.ifc.org/wps/wcm/connect/publications_ext_content/ifc_external_publication_site/publications_listing_page/artificial+intelligence+and+healthcare+in+emerging+markets.
11. [11]Bright, Jake, 2024. Vula Mobile connects rural African doctors to specialist care. Quartz Africa. https://qz.com/africa/vula-mobile-south-africa-telemedicine.
12. [12]Mureithi, Carlos, 2023. Ilara Health raises Series B to expand diagnostic services in Kenya. Rest of World. https://restofworld.org/2023/ilara-health-kenya-diagnostics/.
13. [13]Republic of South Africa, 2020. Protection of Personal Information Act 4 of 2013 (POPIA), commencement July 2020. Information Regulator, South Africa. https://inforegulator.org.za/popia/.
14. [14]Federal Republic of Nigeria, 2023. Nigeria Data Protection Act 2023 (NDPA). Nigeria Data Protection Commission. https://ndpc.gov.ng/Files/Nigeria_Data_Protection_Act_2023.pdf.
15. [15]Mollura, Daniel J et al., 2020. Artificial Intelligence in Low- and Middle-Income Countries: Innovating Global Health Radiology. Radiology, Vol. 297, Issue 3, pp. 513-520. https://pubs.rsna.org/doi/10.1148/radiol.2020201434.
16. [16]Gambrell, Jon, 2024. MinoHealth AI in Ghana applies machine learning to radiology shortages. Associated Press. https://apnews.com/article/ghana-ai-healthcare-minohealth.