Our customer workloads are bursty, so being able to scale down to 0 (or close to it) saves us a lot of CPU and memory that would otherwise do nothing for most of the day.