A medicine budget aimed where it could do more.

The human problem

Scarce essential medicines can be distributed unevenly, leaving some clinics without enough stock while needs differ across districts.

The prior constraint

Allocation decisions had limited timely demand data and had to balance many medicines, facilities, and constraints.

AI’s actual role

An optimization system used available health and supply information to recommend allocations under budget and logistics constraints.

The documented result

The Nature paper reports a five-district evaluation and an estimated 19% increase in medicine consumption; the authors report national-scale computation costing about US$30 per month. The national two-million-person figure describes intended reach, not a measured health outcome.

Why it may matter

Better allocation can make a fixed public budget more useful without pretending software creates medicines, transport, or clinical capacity.

Limitations

The field work covered five of Sierra Leone's 16 districts.

Medicine consumption is not the same as improved health, and the 19% figure is a Synthetic Difference-in-Differences estimate from the five-district pilot rather than a randomized health-outcome result.

National scaling still depends on procurement, transport, data quality, and local judgment.

Unresolved questions

Do stock availability and patient outcomes improve after national use?

How robust are recommendations when demand or inventory data are incomplete?

How can district staff override or contest an allocation?