When should you start thinking about a specific plan for cache invalidation?

we've been using Redis for caching API responses for a while now, and things are mostly smooth. But, as our data gets more complex and writes increase, we're starting to see stale data issues crop up more often. We currently have a pretty basic 'expire everything after an hour' approach, and it's just not cutting it anymore. Has anyone else hit this point? What strategies did you adopt for cache invalidation when simple TTL wasn't enough? I'm curious about approaches for time-sensitive data versus less critical data.