The nature of ML is to reach for easily measured proxy problems which can then be optimized. Why worry about a backoffice integrated micro-service you have no d

The nature of ML is to reach for easily measured proxy problems which can then be optimized. Why worry about a backoffice integrated micro-service you have no data on when you aren't sure that you can design and implement a rewrite of an OSS project?