MKTG 3596 Lecture Notes - Lecture 5: Simple Random Sample, Logistic Regression, Test Pilot
Document Summary
75% of consumer direct marketers use rfm. Easy to understand, no statistics required: recency = interpurchase time, frequency = purchase incidence, monetary = purchase amount. Attempts to predict additional outcomes from observed outcomes. Binary classification models (rfm, logit models, & decision trees) (only 2 outcomes, yes or no) Other models focus on linear outcomes, like purchase amount. We want to predict things like who is most likely to respond, so we only have to send catalogs to those, cut costs. Do not have laser point accuracy, no matter what model we use. Usually try to increase our accuracy from the mean response rate. First create a subsample (test/ pilot group) (ex. send coupons to a sample of your customer database) Then after you (ex. mail catalogs) you record the responses. Then you use these customers" behaviors (both responders and non- responders) to develop your predictive model. Rfm: past behavior is predictive of the future (just like clv)
