I build response models for database marketing using SAS. Each model predicts likelihood to respond to a marketing campaign. The exploratory data analysis done during the modeling process can also yield business insight into purchasing behavior.
Building the model requires data from around 1,000 current customers who have responded to a similar campaign or have purchased the product in the past. Customer data is required in order to build the model. Generally speaking the more customer data that is available the better the model. The modeling process analyzes all available data, often hundreds of customer variables, and selects the ones which have the most predictive power.
The starting point for most models is past purchase history. Recency, frequency and monetary value (RFM) of past purchases are almost always important components of a model. Past product mix can also be helpful in predicting purchase ie cross sell. Appended demographic data such as age and education level can also be beneficial. In B2B applications firmographics such as NAICS code can be predictive.
When completed a model can be evaluated and decisions made as to how many customers should be promoted based on return on investment (ROI) estimates. These are examples of charts used to evaluate models. (Click on each chart for more information.)
|Response Chart: Evaluating the Model||Lift Chart: Model vs. No Model||Lift Table: Estimating ROI