I think that's the key here. It's not ML magic that's driving success, it data literacy. Steve Levitt (of Freakanomics fame) said his advice to students would be to ensure they have base knowledge in understanding data analytics regardless of their domain. ML is just the sexy subset that gets all the attention

And ML is a tiny part of the whole data science toolchain and process. Getting the data to the point where you can use it for something interesting is probably the hardest part.

My point was that cutting edge ML can add a bit more on top of what data literacy can achieve. At scale that’s worth a lot.

What I meant was for eg applying deep neural networks on large data sets for eg can give you a few percentage points improvements on systems implemented without them.

At scale it makes a big difference.