The thundering herd problem in distributed systems is an often unexpected failure mode. This is great, and easy to plan for. You'd save buckets of money because most capacity is used for spikes, and you now have a good idea of when spikes occur.

Every Monday, an hour after school, scale up 5x, because the timer has reset for students. Scale down from Wednesday to Friday. Scale up 2x over the weekends.

Or whatever your data tells you.