I think where we'll really start to see the needle move is when we'll have more automated feature engineering. The 'data wrangling' involved here is by far the most expensive part of the process but also best way to improve model performance.
At Airbnb, we use machine learning to build customer lifetime value models for guests and hosts. These models allow us to calibrate our marketing spend to expected return at very granular levels, down to the user, if we like.
That's scary.