Thanks for the reply @steffens!
So I'm probably better of just indexing my custom data (which also follows the ECS as much as possible) in the default filebeat indices?
That way the ILM is working as expected and the overhead of the filebeat mapping will be minimal, correct?