The hard part is the constraints, not the calendar

Reading free/busy from a calendar is trivial. What makes scheduling automation difficult is everything around it: which staff member is qualified for which appointment type, how much buffer a particular service needs, which rooms or equipment are shared, what happens across timezones, and how far ahead the diary may be filled. Those rules usually exist only in the head of whoever currently does the booking. Getting them written down is the bulk of the work, and it is worth doing even if the automation is never built, because it is also what makes the role handover-able.

Cancellations, changes and the things it must not decide

A booking system is judged on how it handles change, not on how it handles the happy path. Rescheduling, cancellation windows, no-shows and waiting lists each need a stated policy before an automation can apply one. Some decisions stay with people permanently: an assistant does not triage clinical urgency, does not decide whether an emergency slot is warranted, and does not overrule a practitioner's own diary block. Those boundaries are configured as hard stops rather than as instructions to a model, because an instruction is a preference and a stop is a guarantee.

Run a scheduling pilot against the live booking rules

A request should carry the appointment type, required duration, timezone, permitted date window and a request ID. Check staff and room constraints against calendar free/busy, then re-read availability before confirming so two requests cannot claim the same slot. A reschedule or cancellation follows the written policy and updates the source booking; an uncertain calendar response is reconciled by request ID before retrying. Send reminders only through the approved channel and preference on file. Before launch, the booking owner records completion rate, double-booking incidents and manual intervention on a representative sample; accept the pilot against those same measures and route unresolved conflicts to a named person.

How to measure the pilot

A pilot is accepted against the current process, not an industry promise. Agree the sample, baseline and owner before launch, then review the same measures after the workflow has handled real cases.