Dear sir or madam,
I am an experienced Machine Learning Engineer and I have also worked as a Data Scientist for 2 years.
1. It all depends. It is necessary to test different classification models to choose the one, which does the best job. But according to my experience, Boosted Trees work the best in such a case. Regarding the second part, in fact, if you have the value of the purchase, we may go for regression trees, for example. In this case, after prediction, you will allocate the budget to customers with the highest predicted value of purchase. But as I said, it all depends and requires experiments.
2. I have done models of tariff upsell and customer churn (I worked in telecom industry).
3. Did not get the question.
4. Did not get it either, but if you mean the quality of my response, I try to do my best.
Yours sincerely,
Vlad