The strongest current case studies focus on defined work—chart review, documentation or drafting—not vague transformation. That distinction matters because a broad promise cannot be measured or governed.
What practices should take from this
Start with a workflow that occurs often, has a clear beginning and end, and is currently owned by identifiable staff. Measure active minutes, waiting time, rework and after-hours completion before changing anything.
Then run a limited pilot with explicit human review. If the tool saves drafting time but doubles correction time, the capacity was not recovered. If it shifts work from a physician to an already overloaded medical assistant, the practice has moved the burden rather than removed it.
The most credible AI plan is a capacity plan: a baseline, a safe intervention, a measured result and a decision about whether to scale.