Just flip that around. First consider whether you can get data to solve a problem, and if you can get sufficient data, do machine learning. What you basically g

Just flip that around. First consider whether you can get data to solve a problem, and if you can get sufficient data, do machine learning. What you basically get is test-driven development with the development part being done automatically. If you can't get data cheaply enough, only then consider your normal choices. It's easier and less creative to just use the data.