Predictive Modeling

Ozgur Dogan wrote a nice, if short, article on Advanced Predictive Modeling Made Simple (Sort of) for Chief Marketer that came my way today. He makes three good, clear points in the article:

• Segment your customers
  Absolutely, Good segmentation can really improve results (like this for instance). Making your segments more and more fine-grained - heading towards "segments of 1" - is essential when dealing with the "long tail" and very beneficial just in general. When you have data that might repay many segments, something like genetic algorithms might be worthwhile.
• Focus on constant improvement
  This is also true but not just about analytic improvement. You must have the operational infrastructure to try and evaluate multiple approaches in parallel to see what works best and to allow you to put the best approach quickly into production. We call this adaptive control.
• Maximize the data you can use
  Analyzing transactional data to find the best predictors, and integrating external data effectively, are great ways to add value to the data you have. So is customer data integration. Just don't feel you have to wait for these things before you can do predictive modeling. They enhance the quality of models, for sure, they are not required.

I would add one more.

• Find and improve the hidden decisions
  I find that many companies make far more marketing decisions than they think they do. For instance, in a marketing campaign you actually make one decision per customer sent a mailer, for instance, on top of the segmentation, prediction, selection and creative decisions you make. EDM is about focusing on these high-volume, often hidden decisions are automating and improving them. This came up before in some posts on micro and macro decisions and customers who have done this have seen tremendous results - 2000% improvements for instance

If you are interested in predictive analytics, check out my FAQ

Technorati Tags: analytics, Chief Marketer, decisioning, EDM, Enterprise Decision Management, genetic algorithm, long tail, Marketing, predictive analytics, predictive model, segmentation, adaptive control