Overview of Uplift Modeling in SAS

This is an overview of the process of uplift modeling and a list of good sources on how to approach it in SAS.

Before we get into it, I need to note that there exist two similar terms: model lift and lift/uplift/net lift/incremental lift model. The first is a predictive model accuracy metric, and it shows how good the model is at classifying subjects into groups based on their predictive response.

The second is the uplift modeling I am talking about in this post, which is modeling of how different the subject response is when exposed to marketing communication as opposed to not being exposed.

Therefore, uplift modeling is a technique that allows us to determine which targets are more likely produce incremental response when exposed to marketing material.

What do you need to create an uplift model? To create uplift model, we need to conduct an experiment. An experiment is a setup where we measure outcome from at least two groups, with at least one being exposed to our measured treatment and one not. Sometimes a natural experiment will work, but in a classic case, we designate test and control targets intentionally, with randomized assignment.

What does the uplift model do? It models what kind of lift the presence of driver can make on the behavior of the targets. It compares the response of similar targets with and without presence of the driver, and uplift or incremental lift simply means difference in that response.

Is uplift modeling the same as direct response modeling? No. There is a very important difference: uplift model adjusts for direct response activity that can be traced to direct response, but would happen anyway. It also adjusts for any activity driven my marketing but not captured in direct response. In other words, while direct response is a surrogate, uplift is the real deal.

Do we need to use advanced methods to build an uplift model? You would need to use advanced methods if you wish to build a true model. However, you can optimize your targeting strategy using uplift approach by measuring difference in test vs control outcome and then splitting it by customer segments.

SAS has created a special feature in Enterprise Miner called Incremental Response modeling that helps you find relevant input attributes differentiating between customers have higher incremental response rates and those that don’t. A modeling procedure can predict uplift for every customer and then put customers into deciles based on their potential incremental response due to a marketing campaign.

SAS Demo

This 12 minute video provides an overview of uplift modeling and SAS procedures that can help accomplish it. It walks through the basic modeling approach and shows how to use Net Weight of Evidence (NWOE) and Net Information Value (NIV) for input variable selection.