The Diagnostic Debate Is Interesting. The Stock-Out Is Killing People This Week.

Health leaders in this region spend a remarkable amount of attention on diagnostic AI. Whether a model can read a scan. Whether it can be trusted. What it means for the profession. Those are real questions and somebody should be asking them. But while that conversation runs, medicine is expiring in one store while a facility two hours away has none of it. That is happening this week, it is entirely solvable with tools that have existed for years, and it does not require anybody to trust a model with a clinical judgement. The pattern is not unique to healthcare. Every sector over indexes on the sophisticated application and under indexes on the boring one. In healthcare the cost of that preference is unusually high, because the boring problem is a supply chain problem and supply chain problems in medicine are measured in outcomes rather than margin. There is a practical argument too. An organisation that has never deployed anything should not start with the application where being wrong is most consequential. Get the method right on stock and scheduling, build the institutional habit of measuring whether it worked, then move closer to care with some evidence that you can do this. Start where being wrong costs you an afternoon. Not where being wrong costs somebody their treatment. *Stanislaus Martins advises enterprise teams across Sub Saharan Africa on practical AI adoption. This is operational guidance, not clinical advice. [Speaking and training](/speaking).*