A good pilot is safe enough to try and real enough to matter. Choose a service with predictable clinics and a lead who can give honest feedback. Define success before you start: a target time‑to‑note, a maximum acceptable regenerations per session, and a push success threshold. Limit the scope deliberately—two clinics per week and one referral template will teach you more than a large, fuzzy programme because signals are easier to read.

Run the pilot like a clinical quality cycle. Hold short huddles in week one, then twice weekly. Keep a visible board of issues and experiments: “We added a one‑line instruction to the GP letter; result: fewer edits.” Change only one or two things at a time so you can link cause and effect. Resist the temptation to add more training when a template fix would remove the confusion entirely.

Document decisions in plain language. “We are keeping Template v3 because it reduced average editing time by 90 seconds. We are retiring Template v1 because it duplicates content and causes confusion.” This narrative matters: it is what leaders read to decide whether to fund expansion and what clinicians read to decide whether to try.

Scale when the evidence is strong and the support model is ready. If the metrics have stabilised and champions have capacity, move to a second group with the same discipline: clear scope, quick feedback loops, and public learning. If not, extend the pilot without apology—doing a little more work now is cheaper than fixing an avoidable loss of confidence later.