Good question Abdul. I look for real pain, not a feeling: slow queries you can't fix with an index, background jobs piling up, or one part of the app scaling very differently from the rest. Until one of those shows up, the boring stack usually has plenty of room left.

That makes sense real performance bottlenecks are a much better signal than anticipating problems too early. Do you usually start by optimizing the existing architecture, or split out specific services once the bottleneck becomes clear?

Optimize first, almost always. Most bottlenecks go away with better queries, caching or moving slow work to a background job. I only split out a service when one piece clearly needs to scale or deploy on its own, and even then I pull out just that piece.