I always do this, but how do when to re-architect or at what level of granularity to look for optimizations? Do we burn time looking at async processing, do we

I always do this, but how do when to re-architect or at what level of granularity to look for optimizations? Do we burn time looking at async processing, do we just scale the cluster, etc