I feel the trouble is the hype around what model.fit() can and cannot do is where you need that real data science expertise. Training a poor model is easy - convincing executive types that it is a poor model and derimental to the target process takes real skill