Once the linear regression and/or multinomial logit models have been developed
and a set of coefficients needed to build a predictive equation has been
derived, a hold-out sample of 2,300 observations would be used to verify the
models and to generate a predicted response rate to the marketing campaign.
The response rate represents the percentage of customers within the top nth
percentile of the holdout sample that actually made a positive purchase
decision. These response rates were then used to generate financial figures,
such as revenues, marginal costs, and return on investment (ROI).
Additionally, a sensitivity analysis of net revenue to percentage of
customers targeted can be conducted to optimize the model. The graph above
shows that net revenue can be optimized at $23,473.91 if 49.2% of the customer
base is targeted for the mail campaign and given 15.72% positive response rate.
Hence, it may not be optimal to implement a full-scale direct marketing campaign
or even a predefined decile/percentage of customers. |
Predictive Modeling
.