I used to think this. Then I learned/realised how absolutely crappy the tooling available to those marketing departments is. The solution is for push to not be

I used to think this. Then I learned/realised how absolutely crappy the tooling available to those marketing departments is. The solution is for push to not be a manually run operation, it has to be run entirely by a kind of hybrid reinforcement learning system that can automatically manage the explore/exploit tradeoff in order to learn a person's preferences and then update its assessment as those preferences change over time.