In one of the lectures of the intro to machine learning course, Jeremy mentions his article Designing great data products and I found the ideas fascinating. I am trying to find other resources about this (maybe books or other moocs) that do deeper in this idea.
Using predictive models as inputs in a simulation of the business allows business people to run what-ifs scenarios in the future but also find optimized versions of the levers to pull to maximize the desired outcome.
After some Googling, the only things I can find is about Response Surface Methodology (RSM), but I would really like to read more about the subject.
Most articles on the web focus on algorithms, getting better accuracy on Kaggle competitions, but I actually think this Drive Train approach would be much more helpful in a business setting.