Role of Modeling in Database Marketing Role of Modeling in Database Marketing • Forecasts (aggregate level) vs Predictions (individual level) vs Segmentation (no dep var) • Forecasts obtained thru Time Series etc • Applying Scoring Models to forecasts - Obtain average response in current period - Score current customers to get individual response rates - Obtain average forecast for next period - Proportionately adjust to get individual response rates in next period • Applying Scoring Models at individual level Example: Applying Scoring Models to forecasts • Database mktr. has 2 million names on file • Using RFM it decides that 1 million are worth mailing • Most recent summer mailing to the 1 million selected people pulled 2% • Next, logistic regression is used (0/1 response var) to score the 1 million persons mailed to • Avg response is 2% and response by deciles is obtained from model Example: Applying Scoring Models to forecasts • Analyst estimates (using forecasting techniques) that Fall mailing will pull 2.5% on average • Now, the 1 million individuals can be individually (& decile-wise) scored for a fall mailing by proportionately adjusting the average • Finally, what about the other 1 million people in dbase? • Again, analyst needs to estimate their overall response to a fall mailing, and adjust it to get individual response scores Role of Modeling in Database Marketing • RFM versus regression - RFM is arbitrary in nature - Regression can do more • CHAID versus RFM and regression - Can handle interactions - Can guide analyst about which interactions to include in a regression - Provides benchmark against regression results - A set of univariate CHAIDs can act as a quality control tool Role of Modeling in Database Marketing • Using Principal Components to model buying patterns - Factor Analysis to reduce data • What’s the right number of variables to use - Examine statistics for significance of var - Use t-stat/chi-square stat in reg/logistic reg • Typical model results - How response rates vary among deciles Role of Modeling in Database Marketing • Zip Code models - Work with outside mailing lists of zip code-based census data - Each zip code is associated with string of demographic variables - Zip code models not as good as models based on internal performance data Role of Modeling in Database Marketing • Zip code models (contd) • Some issues - Impact of demographic var in a zip code are assumed to work across all list categories and all lists in a category - Selection of independent var - Adjusting for zip code size (weighted least squares regression) Role of Modeling in Database Marketing • Lead conversion models - Similar to response models if we consider Conversion Rates to be like response rates - Divide all leads into deciles and assign a probability of conversion to each decile - Uses Falloff Rates between efforts to estimate conversion rates of subsequent efforts