Response Modeling

The objective of this analysis is to identify customers likely to respond to an offer. The offer could be a promotion, or a communication that asks the customer to make a purchase or to give a response that could lead to a purchase. There are two types of models that are generally used for this purpose; regression scoring models and Logit models.

To estimate a response model, we first select a sample of customers who receive a specific offer. These customers respond either by making a purchase (responding favorably to the offer) or not making a purchase (responding negatively to the offer). The dependent variable is coded 0 (no buy) and 1 (buy).

We then select a set of explanatory variables. These can include the customer's historical buying behavior (such as, recency, frequency, and monetary); types of other products or services purchased; and customer characteristics (such as, demographic variables).

Once these independent variables are chosen, a model is estimated. SPSS and SAS are the more sophisticated packages that can be used to estimate a response model. These packages are particularly suited for estimating a Logit model. However if your knowledge of these packages is limited, you can use the regression tool in EXCEL to estimate a regression scoring model and get descent results.

With the results from the model, we score all other customers in the database who were not in the test sample. Those customers with scores above a specific cutoff level receive offers, and those below the cutoff do not. The cutoff is determined economically, based on sales, margins and cost of providing an offer.

Response Modeling: Regression Models
Response Modeling: Logit Models