The naive interpretation of no-show data is that some patients are flaky. The data does not support this. The same patient who misses a 9am Tuesday will reliably show up for a 4pm Thursday. Patterns are stable, knowable, and ignored by most scheduling systems.
What changes the rate
Three things, in order of impact: matching the slot to the patient's pattern, sending the right reminder at the right channel, and offering a one-tap reschedule before the slot is at risk. The model just runs the matching.
