With donor budgets contracting sharply, AI in global health must prove value in the currency the sector uses: cost per outcome and cost per disability-adjusted life year (DALY) averted. This playbook builds the business case for ministries, NGOs and funders under real financial constraints. It covers cost per DALY thresholds, the 2025 collapse in donor funding, how AI affects coverage and efficiency, total cost of ownership beyond the license fee, and sustainability once grant money ends. It shows how to evaluate AI as a health investment, not a technology purchase, so that scarce dollars buy the most health possible.
Every dollar must buy more health
Global health economics turned harsh in 2025. The dismantling of USAID removed roughly $40 billion in annual funding, and both the Global Fund and Gavi faced replenishment shortfalls that forced hard prioritization. When a shrinking budget must cover the same or a growing disease burden, the only defensible question about any AI tool is whether it averts more DALYs per dollar than the next best use of that same money. The WHO-CHOICE tradition uses cost-effectiveness thresholds tied to income, and many programs treat an intervention costing well under a country's GDP per capita per DALY averted as attractive, though thresholds vary and should be set locally rather than borrowed from a richer setting.
AI does not automatically clear that bar. A tool can raise diagnostic accuracy yet still cost more per outcome than a cheaper existing approach once you count the hardware, connectivity, training, and support it requires to function. The winning cases are those where AI extends a scarce workforce or cuts commodity waste, converting a fixed budget into more coverage and more outcomes. The losing cases are technically impressive models with a total cost of ownership no ministry can sustain after the launch grant expires. Framing the decision as a health investment rather than a technology purchase forces the right comparison and keeps procurement honest. It also reframes who should be in the room: a decision judged on cost per outcome belongs to health economists and program leads as much as to technologists, and pulling those disciplines together early is what stops an organization from buying an impressive tool that fails the one test that matters when the money runs short.
A cost-per-outcome business case model
Evaluate AI as a health investment. Compare incremental cost against incremental outcomes versus the current standard of care, and stress-test whether the economics still hold after donor funding recedes and the tool must stand on domestic financing. Run the model under a pessimistic funding scenario as well as a central one, because a tool that is only cost-effective while a generous grant subsidizes it is not a sustainable investment, and pretending otherwise simply defers the reckoning to the moment the grant closes.
| Dimension | Question | Test to pass |
|---|---|---|
| Cost per DALY | What does the tool cost per DALY averted? | Below the locally agreed cost-effectiveness threshold |
| Total cost of ownership | What is the full cost beyond the license? | Hardware, connectivity, training, and support all counted |
| Coverage effect | Does it extend reach with the same budget? | More people screened or treated per dollar spent |
| Efficiency effect | Does it cut waste or redundant work? | Fewer stockouts, unnecessary referrals, or repeat tests |
| Sustainability | Can it run without the launch grant? | Domestic or pooled funding path clearly identified |
Build the case in health-economic terms
- Express value as cost per DALY averted or cost per outcome, benchmarked against the local cost-effectiveness threshold, rather than as accuracy or model performance alone.
- Model total cost of ownership across hardware, connectivity, training, maintenance, and support, not just the headline software license fee.
- Quantify the coverage and efficiency gains explicitly: extra people reached per dollar, stockouts avoided, and repeat tests or referrals eliminated.
- Require a sustainability plan from day one that names the domestic or pooled funding source that will carry the tool after the initial grant.
- Compare against the cheapest adequate alternative, so AI must beat the status quo on cost per outcome and not merely on capability or novelty.
How the ROI case goes wrong
- Selling accuracy improvements with no translation into cost per outcome or DALY averted, leaving funders unable to compare it to anything.
- Underestimating total cost of ownership by ignoring connectivity, devices, training, and long-term support that dominate cost in low-resource settings.
- Building a case that only works while donor money flows, with no credible path to domestic financing after the grant closes.
- Comparing AI to doing nothing rather than to the cheapest existing approach that already delivers an acceptable outcome.
Prove value where budgets are counted
- Cost per DALY averted or per confirmed outcome versus the standard of care.
- Total cost of ownership per facility per year, fully loaded across all inputs.
- Additional coverage delivered per dollar compared with the pre-AI baseline.
- Share of recurring cost covered by domestic or pooled funding by year two.
Frequently asked questions
How should AI value be measured in global health?
In the sector's own currency: cost per DALY averted or cost per confirmed outcome, benchmarked against the local cost-effectiveness threshold. Accuracy gains only matter if they translate into more health per dollar than the next best use of the same money. Evaluate AI as a health investment competing for scarce funds, not as a technology purchase judged on its features.
Why do funding cuts change the AI business case?
The 2025 loss of roughly $40 billion in USAID funding, plus Global Fund and Gavi shortfalls, means the same disease burden must be covered with far less money. That raises the bar: an AI tool now has to beat the cheapest adequate alternative on cost per outcome, and it must have a credible path to keep running without the launch grant that first paid for it.
What is often missing from AI cost estimates?
Total cost of ownership. Buyers focus on the license fee and overlook hardware, connectivity, training, maintenance, and long-term support, which often dwarf the software cost in low-resource settings. Include all of it, and require a sustainability plan that names who funds the recurring cost after donor money ends, or the tool will not survive.
Related reading
Go deeper on this sector and topic.