When does the Actor Model work well for making backend services that can grow?

I've been checking out the Actor Model, especially with stuff like Akka, for making really concurrent and distributed systems. It looks good on paper, offering a neat way to handle state and concurrency without locks, which is great for scaling. But I'm wondering about real life. Besides just 'high concurrency,' when does the Actor Model actually beat out older methods like thread pools and shared memory, or even basic message queues? Are there specific kinds of problems or areas where it's a clear winner, or where it just makes things harder for no good reason? I'm mostly curious about how it handles state and keeps working even when things go wrong in actual production systems.