[thomasrossi]

I think the interesting observation is: if you are using ML chances are 80% of your algorithms are already published somewhere and available to all, so one should tune in that 20% of his/her own field, some field which is not a common research topic. I also think that a huge difference, once you have the same algorithmic machine, is the quality of data you collect. That is also quite differentiating, I mean, respect to your own business problem, gathering data a way or the other may really destroy or boost your ML performances.