In my experience it really depends on the vertical. For some things (ford cars) demographics don't work well, but for others (mercedes) demographics like househ

In my experience it really depends on the vertical. For some things (ford cars) demographics don't work well, but for others (mercedes) demographics like household income and age perform really well. I think their is so much input for rtb that it is prime for ML. Most exchanges have some basic ml auto-optimization, but eventually someone will automate almost the whole thing.