Understanding how the techniques work and how they relate to outcomes on a particular kind of problem. That would help me understood where a particular technique can be applied for what particular problems. With that knowledge, I could hire someone to do the machine learning coding if and only if I could expect it to be worth our while. Actually doing the coding could be automated soon enough. There is already visual programming for machine learning. And I reckon the cohort of people doing machine learning are more likely to be skilled in automating programming procedures than computer scientists on average. So, I do not expect it worthwhile to learn the nitty gritty of data-science in depth, but rather the concepts in breadth.