One thing that has helped me a lot with monitoring is custom application-level metrics.

If you have a good idea of the usage patterns of your service, create metrics backed by the patterns. This can help you find things that CPU/Memory will hide.