Database read replicas, when do they actually make sense?

ive been seeing a lot of advice about implementing read replicas for databases, espescially in scaling discussions. it seems like a go-to solution. but im curious about the real-world scenarios where this pattern truly provides benefits versus just adding complexity. are there specific query patterns, data access frequencies, or application types where read replicas are a clear win? or are there cases where the overhead of managing them outweighs the read performance gains, especially for smaller applications? lets discuss the practical tipping points.