Most ML jobs are sitting around scrubbing data. The fun math part is like 5% of the job, and usually there's some guy (or a small handful of guys) that's a PhD who is hired to do all of that because he was in the research area before he left academia.
Agreed. We call that team "operations research" and they are all PhDs. The ML devs build the pipelines and deal with engineering concerns around having/moving/securing lots of data.
I could be wrong, but this is my experience.