Debugging Performance bottlenecks

Hi
Trying to debug possible performance bottleneck in instancing static meshes being driven by vertex animation.

Clearly I am GPU bound based on a stat unit call.
Where should I focus my attention to optimize performance?

ran stat unit

Camera facing Open Sky
24 fps and 2-8 mesh draw calls

Game 10 ms
Draw .1 ms
GPU 47 ms

Camera facing crowd static meshes
5 fps and 6-10,000 draw calls
Game 10 ms
Draw .1 ms
GPU 240 ms