It works well with master data [1], i.e. semi-static data of limited volume that is widely needed throughout systems. Think physical store locations and such.
Services can either store it in their own database and still get an update stream, or services can read the complete set on boot and keep it in memory. It can avoid the need for RPC/REST calls so centralized services that keep track off this kind of data, preventing cascading failures if those fail.
