Early on, some large companies that were interested in Watson as portrayed by IBM, privately noted later that while they saw great promise, Watson in reality, was a set of technologies that needed to be stitched together at their site. That meant buying Watson—and then preparing data for Watson—was more a big integration project than a product purchase.
This was basically the gist of what I was saying in the roundup\review, the solutions being offered currently all require the individual organization to be fluent in taking advantage of training models. This is an extraordinary amount of work, and makes the gains from these efforts inaccessible to the masses. In theory, this is what Apple appears to be addressing with their Core ML framework. It will be very interesting to see who can deliver on this ideal first.