Domain knowledge is really important, you can't overstate that. But you've also succinctly described why a huge percentage of ML as practiced in industry under

Domain knowledge is really important, you can't overstate that. But you've also succinctly described why a huge percentage of ML as practiced in industry under performs or flat out fails. To a first approximation the person who reads a few tutorials and plays with example data sets, then sets out to apply it to their own domain, has no real idea what they are doing - and it shows.