Client, relies heavily on direct mail to acquire new customers

• They would like to evaluate the propensity of a customer to respond in light of particular events viz. Holiday Season / Back to School etc.
• To get the best return on their marketing investment they would like to mail only to those prospects who are highly likely to respond

• Develop separate direct mail response models to predict response at an individual prospect level for a particular event / season/time frame
• Use Consumer Bureau & Demographic attributes to model response
• Develop a combined scorecard using these individual models
• score "eligible" prospects using the model and select those with highest score for mailing given the event
  • PROCESS STEPS
    • Variables Considered
      • Consumer Bureau Attributes
      • Demographic Variables
      • Dummy Variable indicating Event / Season
    • Modeling Process
      • Data Manipulation (Merging, Cleaning, etc.)
      • Data Transformation (e.g., Logs, Inverses, etc.)
      • Eliminating Information Redundancy (i.e., multi-collinearity)
      • Reducing the # of Variables (i.e., logistic regression)
      • Tweaking the Model (i.e., checking for intuitiveness of coefficients and their sign, decile rank-ordering, etc.)
      • Validating the Model (using a holdout sample)
      • For 2 events A & B (mutually exclusive and exhaustive) two separate models f(x) and g(x) were developed
      • The two models are combined by use of dummy variables as follows:
        h(x) = t1(f(x)) + t2(g(x)) where
        t1 = 1 if Event A else 0
        t2 = 1 if Event B else