ABM will continue to learn. Usually on cold start situations or post significant changes (adding new queues; making significant alterations of route paths; a drastic step up/down in volume/AHT) it takes a couple of weeks. If ABM reacted more immediately it would overcorrect/oversteer and be chaotically stochastic.
Worst-case post significant change you could produce a forecast using each method and make a schedule-time decision on which to use and then at some point 1, 2, or 3 weeks later stop dual producing schedules and use ABM again.
We are looking at ways to better support cold-start and large changes more immediately (e.g., for someone to denote change is expected and you want to have it be not considered an outlier).
------------------------------
Jay Langsford
Senior Director, Workforce Optimization Engineering
------------------------------