Instance-based metrics like memory usage become especially useful when collected from all the instances in an auto scaling group.

By collecting instance or even application-specific metrics (for example: number of threads used, internal request duration, and so on) you can make your auto scaling groups increase and decrease in size at the most appropriate times to your workload and performance profile.

To do this, make the auto scaling group name one of the dimensions (you can define multiple dimensions) sent along with your metric value.