Hi all - Nate here,
I recently co-authored an article in Lancet Primary Care with the Community Health Impact Coalition (CHIC) Research Group. The article discusses a survey of 28 NGOs running CHW programs across 18 countries. The survey found that 61% are piloting AI tools. Only one reports meaningful integration within a national health system.
Of the AI initiatives already underway, 48% are led by NGOs or private entities with government approval, 19% operate with no government involvement, and 22% are co-led with government partners. None are solely government-owned.
The mapping I conducted of 38 AI-in-CHW programs across 27 countries shows the same pattern. 68% of programs are active pilots. Only one program, Afya-Tek in Tanzania, is integrated into a national system.
The article identifies two problems.
The first is a governance gap. 57% of surveyed NGOs rated their own AI capacity as solid. Nearly two-thirds judged the systems they operate within as not ready or only minimally ready. That gap, between implementer confidence and system readiness, is where tools tend to stall.
The second is a sequencing problem. Whether an AI tool works in practice depends as much on the surrounding infrastructure as on the technology itself. Deploying AI where the required referral pathways, workforce capacity, or clinical infrastructure don't yet exist tends to produce poor results. And those poor results often go undetected, because success gets measured by whether a pilot launched, not by whether patient outcomes improved.
The mapping adds one more data point here. 66% of programs have only organizational reports or no published evidence. LLM and generative AI tools account for 58% of all programs, but not one has been evaluated in a randomized trial. This lack of rigorous evaluation means activity gets documented and outcomes don't.
What would it take to change this? The article argues for three things: governments owning and governing AI tools as public infrastructure rather than receiving them as donor projects; shared learning systems so that evidence compounds across implementers instead of sitting in individual pilot reports; and investment in the workforce and data systems that AI assumes already exist. Those aren't technical problems. They're political and financial ones, and they require decisions from governments, donors, and NGOs before the next wave of tools arrives.