when do you decide to implement caching

hey everyone, i've been thinking a lot about performance lately. we've got a few services that are starting to see some serious load, and i'm wondering when you all typically decide it's time to introduce caching. is it purely based on request latency, or are there other metrics you look at? do you usually go for an in-memory cache first, or jump straight to something like Redis or Memcached? curious to hear your thought processes