A small number of AI-assisted diagnostic services have established payment pathways, mostly in imaging and mostly where the tool substitutes for an existing billable interpretation. Everything else is funded from operational budgets, which means it must justify itself through cost avoidance rather than revenue.

That distinction has consequences for what gets developed. Tools that reduce staffing needs, shorten length of stay or prevent readmission penalties can be sold on an internal business case. Tools that improve a clinical outcome without touching one of those levers are substantially harder to fund, regardless of how well they work.

Several policy groups have proposed a general category for algorithmic clinical services with outcome-linked payment, which would in principle correct the distortion. The technical obstacle is attribution: establishing that the model, rather than everything else in the care pathway, produced the outcome.

In the meantime the market is doing what markets do. Investment is concentrated in the operationally justifiable, and the clinical categories with the clearest patient benefit and the weakest financial hook remain conspicuously underserved.