When is service mesh actually a good idea for microservices?

We've been using microservices for a bit, and a service mesh always comes up. It offers some great things, like managing traffic, seeing what's happening, and making things secure, all without changing our app code. That sounds pretty good, especially for bigger systems. But then you have to deal with more complexity, extra work to keep it running, and it might even slow things down a little. For smaller teams or stuff that isn't super important, it honestly feels like too much. So, when does a service mesh actually make sense? What are those times it's really worth all the trouble and cost, instead of just doing things the simpler way?