i wouldn't underestimate the power of good tools here. All the software libraries for ML are very easy to get started with, and make it very easy to prototype cool things. It seems like in other applied areas it's a lot more work to get less results.
I think that's exactly the problem with ML. You can get interesting looking results fast with very little effort. But then, getting from an impressive demo to something actually useful in the real world is much harder, and will lead you down endless rabbit-holes as you try to improve your results, but things only get worse, not better...
After 5 years of growth in data mining at my giant pharma, this is exactly what I'm seeing. Most ML projects remain toys while the number of them that advance into something useful can be counted on one hand.
(Of course it's a bit hard to assess the impact of a revolution like ML (esp DL) when your company already has hundreds of statisticians who have been employing similar data/experiment analysis techniques for decades, thereby diluting the signal of how disruptive novel forms of ML are within the enterprise.)
