I tried to pick a very contrived example. The architecture rails forces you to have is great if all of your business logic is simply CRUD. Once your app starts doing other things though, that's where careful consideration is needed as to where that logic should live.
You could imagine callbacks or observers that do all sorts of things:
ping an external service
touch another model or models
send an internal email
or does a combination of a bunch of stuff.
Things get even tougher when you start having flags that turn logic on or off. It has the potential to really become a mess, very fast.