Pretty much every large firm has multiple problems ML can solve better than linear/logistic regression. Smaller firms may still have one or two. In some industr

Pretty much every large firm has multiple problems ML can solve better than linear/logistic regression. Smaller firms may still have one or two. In some industries the core competency will be how good your ML model is as everything else becomes a commodity. There are new advents that make ML better for small data-sets as well as opportunities for data-brokerage to increase access to data. And these are just current applications, new applications are still nascent (i.e self driving cars). Treating ML as a software problem instead of a science project - with a pipeline of adding/creating data, cleaning, modeling, analyzing, learning, and iterating is also incredibly important but it's not like most companies are doing this particularly well either.