Cross-sell Response Model

2,726 views

Published on

A direct mail campaign was administered to existing customers who had never purchased an IRA (Individual Retirement Annuity) as of November 2007 for an offer to purchase.

After receiving confirmation of receipt for the IRA offer, these customers were tracked until July 2008, in order to predict the likelihood of a customer purchasing an IRA.

Applied clustering techniques was utilized to distinguish different groups.

A cross-sell response model was built using logistic regression.

The data was split into a sixty-percent training dataset and a forty-percent scoring and validation set.

The rate of purchase was less than 1%; therefore, the data was oversampled by taking a random sample of the non-responders and applying an offset to the model.

Vist http://www.saraconsultingllc.com to learn more about the presenter.

Published in: Business

Cross-sell Response Model

  1. 1. Author: Melinda Richmond Date: May 7, 2009 CROSS-SELL RESPONSE MODEL Disclaimer: Dummy Data
  2. 2. <ul><li>A direct mail campaign was administered to existing customers who had never purchased an IRA (Individual Retirement Annuity) as of November 2007 for an offer to purchase. </li></ul><ul><li>After receiving confirmation of receipt for the IRA offer, these customers were tracked until July 2008, in order to predict the likelihood of a customer purchasing an IRA. </li></ul><ul><li>Applied clustering techniques was utilized to distinguish different groups. </li></ul><ul><li>A cross-sell response model was built using logistic regression. </li></ul><ul><li>The data was split into a sixty-percent training dataset and a forty-percent scoring and validation set. </li></ul><ul><li>The rate of purchase was less than 1%; therefore, the data was oversampled by taking a random sample of the non-responders and applying an offset to the model. </li></ul>Methodology
  3. 3. Parameter Estimates Consider Brandon, a customer who is fifty-one years old, single with no children, lives in the West Region, currently is in a Premium Paying status with low attrition risk, and tenure of ten years. Brandon’s total cash netflow is $7,689.92 based on the last twelve months and total assets is $84,910.85, he owns RA and SRA products. The likelihood that Brandon will purchase an IRA product is Y=0.76%.
  4. 4. Odds Ratio Y1 and Y2 helps to explain the likelihood of an IRA purchase while holding all other effects constant, and the odds ratio is the # times most likely that a customer will purchase an IRA. ODDS RATIO = (Y1 / Y2)
  5. 5. Validation
  6. 6. Profile Example
  7. 7. <ul><li>Overall Selection of Customers to Target: </li></ul><ul><li>Attrition Risk: High, Medium </li></ul><ul><li>Total Assets = $84,810 or More </li></ul><ul><li>Cash Flow = $7,890 or More </li></ul><ul><li>Tenure 10 years or More </li></ul><ul><li>TPA Account </li></ul><ul><li>Life Insurance Account </li></ul><ul><li>Age 60 or More </li></ul><ul><li>RA Account </li></ul><ul><li>SRA Account </li></ul><ul><li>Premium Paying Status </li></ul><ul><li>This selection should be coupled with Customer Lifetime Value or ROI in order to impress upon the marketing team how much money would be saved by targeting less customers. </li></ul>Conclusion

×