impute missing values Segment customers using clustering and possibly PCA. predict probability of response using neural nets, lasso logit, gbm, rfs. Cross validate to pick the best. If necessary, correct for negative bias in data. for those who did respond, regress expenditures on features. Try a few models. Same as above. Choose customers such that estimated probability×expenditures given response > 30. in presentation, show segments of customers (bar graphs and qualitative description) differences in features for successful targets (bar graphs), expected profit, other dashboards. Give intuition about whom your contacting.