This playbook turns AI ambition into a sequenced, fundable plan for global health organizations operating under tight budgets. It lays out a four-quarter roadmap that moves deliberately from data foundation, to governed pilots, to integrated scale-up, to sustainable operation, with clear gates between phases so that nothing scales before it is safe or affordable. It is written for ministries of health, NGOs and funders facing the 2025 funding contraction, and it stresses sustainability from the first quarter: every phase names its outcome, its governance gate, and how the work will actually be paid for after donor money recedes.
Sequence deliberately, scale only what is proven
The temptation in global health AI is to chase many pilots at once, seeding demos across districts and hoping some take root. The 2025 funding environment makes that approach reckless. With roughly $40 billion in USAID assistance withdrawn and major replenishments under strain, organizations cannot afford tools that never integrate into national systems or never become affordable to run. A roadmap imposes the discipline the moment demands: build the data foundation first, pilot a small number of use cases under governance, integrate only what genuinely works, and plan for sustainability from the outset rather than discovering at grant-end that nothing can continue.
Each phase should end at a gate. If data readiness is red, you do not pilot. If a pilot cannot demonstrate cost per outcome or clear its equity and sovereignty checks, it does not scale. This sounds slow, but it is faster than the alternative of repeated pilotitis, because it concentrates scarce money on the few use cases that will actually survive contact with the field and the funding cycle. The four-quarter roadmap below is a template; real timelines flex with local readiness, procurement rules, and funding windows, but the sequence and the gates hold in every setting. Skipping a gate to move faster is the single most common way these programs waste money. The roadmap works only if the person who controls funding is willing to say no at a gate, so agree in advance who owns each gate decision and what evidence a tool must produce to pass it.
A four-quarter phased roadmap with gates
Move through foundation, pilot, integrate, and sustain. Each quarter has one primary outcome and a gate that must pass before the next phase is funded, so investment always follows proof rather than promise. Treat the quarters as sequence rather than strict calendar: a foundation phase that needs two quarters because the data is not ready is far cheaper than a rushed pilot built on data that cannot support it, and the discipline of the sequence matters more than hitting an arbitrary date.
| Quarter | Focus and outcome | Gate to advance |
|---|---|---|
| Q1 Foundation | Data readiness assessment, DHIS2 integration, governance setup | Data layers green and governance gates defined |
| Q2 Pilot | One or two governed use cases run in shadow then live mode | Local validation passed and cost per outcome shown |
| Q3 Integrate | Embed proven tools into workflows and national systems | Workflow adoption and referral completion sustained |
| Q4 Sustain | Scale-up plan backed by domestic or pooled financing | Funding path secured and monitoring in place |
| Ongoing | Monitoring, bias and drift checks, workforce reskilling | Equity and performance held within agreed thresholds |
Run the roadmap with discipline
- Start with Q1 foundation work: assess data readiness, integrate with the national backbone, and stand up governance gates before building or buying any model.
- Pilot narrowly in Q2, one or two use cases, in shadow mode first, and require local validation and a demonstrated cost per outcome before anything goes live.
- Integrate only proven tools in Q3, embedding them into clinical workflows and DHIS2 rather than leaving them running as disconnected side projects.
- Plan sustainability from Q1, not Q4, by naming the domestic or pooled funding source that will carry recurring cost once the launch grant closes.
- Hold the gates: refuse to scale any tool that fails its equity, sovereignty, validation, or affordability checks, however promising the initial demo appeared.
Where roadmaps break down
- Launching many pilots in parallel, which spreads scarce money thin and reproduces the pilotitis the roadmap exists to prevent.
- Scaling a tool before it clears its governance and cost-per-outcome gate, then discovering it is unsafe, unfair, or unaffordable at full volume.
- Treating sustainability as a final-quarter afterthought instead of a first-quarter design constraint baked into every decision.
- Letting an ambitious timeline override local readiness, so the plan collapses the moment data, connectivity, or workforce capacity is not there.
Track progression through the gates
- Number of use cases that clear each phase gate versus those parked or dropped.
- Time from pilot to integrated workflow for the tools that pass their gates.
- Share of scaled tools with a secured domestic or pooled funding path.
- Equity and performance metrics held within threshold across all live tools.
Frequently asked questions
Why phase AI adoption instead of moving fast?
Because the 2025 funding contraction makes wasted effort unaffordable. Phasing with gates concentrates scarce money on the few use cases that will actually survive, rather than spreading it across pilots that never integrate or never become affordable. It feels slower, but it is faster than repeated pilotitis, because each phase only advances what has already proven itself in the field.
What has to be true before scaling a tool?
It must clear its gates: local validation on the deployment population, a demonstrated cost per outcome below the relevant threshold, passing equity and data sovereignty checks, and sustained adoption in the real workflow. A tool that looks impressive in a demo but fails any one of these gates does not scale, however tempting it is to push it forward.
When should sustainability planning start?
In the first quarter, not the last. Name the domestic or pooled funding source that will carry recurring costs after the launch grant, and factor total cost of ownership into the design from the outset. Treating sustainability as a Q4 afterthought is how programs collapse at grant-end despite promising pilot results and real early impact.
Related reading
Go deeper on this sector and topic.