Predictive Models I’ve Built

While at Epsilon, from 2000-2016, and at Time Inc. prior to that, my primary use of data was to create analyses and predictive models for client applications.  I also created predictive models that became data products in their own right, many of which are still sold to Epsilon clients today.

Generally working with customer file data using SAS and R, I created the following types of modeling solutions:

  1. Likelihood to Pay on a Marketing Offer – Predict the likelihood of payment and non-payment using ensemble modeling techniques
  2. Response Optimization – Enable optimal business results in client marketing campaigns through response models
  3. Segmentation – Used nearest neighbor and hierarchical segmentation techniques to segment the population into categories, or personas
  4. Prospect Models – Predict prospects that look most likely to exhibit a specific behavior, such as, response, pay, or product purchase
  5. Cross-Sell Models – Response models using customer transactional data to predict likelihood to respond to a new product offer
  6. Product Optimization – Given the likelihood of response, predict which product will be most profitable to offer to an existing customer.
  7. Area-Level Data Imputation – Impute missing data values using geographic roll-ups to enable complete data set coverage
  8. Propensity Models – Use census-balanced weighting to model look-alikes to self-reported consumer behaviors
  9. Econometric Models – Predict estimates for household characteristics such as income, net worth, and home value, applied as census-balanced scores across the US market
  10. Area-level Models – Predict geographies with populations most likeyl to exhibit an interest or purchase behavior