Discussions about AI in healthcare tend to start with models: how accurate, how fast, how new. Inside a healthcare organisation, a different question decides the outcome — can the model reach the data it needs, in structure, in time, and can its output reach the professional who has to act on it?
That is an interoperability question. A decision-support function is only as good as the context assembled for it; an image-analysis result is only useful if it lands inside the reading workflow, linked to the right study, in a form the record can hold.
The pathway is the product
Organisations evaluating AI capabilities can borrow a habit from integration teams: trace the pathway before judging the model. Where does the input context live today, and in what structure? Which system presents the output, at which moment of the workflow? How is the output stored, and how is it distinguishable as AI-assisted later?
If those answers require new manual steps, the capability will cost attention rather than save it — regardless of model quality. If the answers run through governed, structured interfaces, even a modest model can deliver dependable value.
What to standardise first
The pragmatic order is: structure the data people already record, define the exchange for the pathways that matter most, and only then place intelligence on top. AI built on this foundation inherits its qualities — traceability, consistency, governance. AI built beside it inherits its gaps.
The organisations getting durable value from healthcare AI are rarely the ones with the most pilots. They are the ones whose information moves well.