How to Perform a Migration Analysis

Oct 06, 2008 11:28 PM  By

What purchase pattern do your customers follow? When you profile your customers’ purchase patterns do most of them shop from only one merchandise category? How many would you like them to shop from?

If you would like them to migrate into other categories, one way to help them do so is to make cross-category offers to the customers most likely to accept them. You probably already do acquisition modeling to score demographically enhanced rental lists and identify prospects similar to your current customers. But it is possible to spend so much time modeling to acquire new customers that you overlook the potential to model and identify good migration customers. Here is how to do it.

To perform a migration analysis, first label your customers by their categories. Look at their last 12 months of activity and flag them as being buyers of one or of two-plus categories. Second, enhance your customer file with demographics at the household, block group or ZIP code level. Third, have your in-house or external analytics resource extract your current single and multiple category buyers as separate groups. Have your analysts split the groups in half and explore one half of each. You’ll understand why in a minute.

Start by profiling each group. You might find that multiple category shoppers have a larger household size or more income than single category shoppers. But, the differences between the one and two-plus category shoppers can be more subtle. In that case you will have to create a regression model for the two-plus category shoppers. Your analysts can use simple linear regression and go on to more complex versions if needed to find the relationship between the customers’ demographics and their buying behavior. After the model is created, use it to score the other half of your two-plus category buyers that you held back. A good model will correctly line up the hold-out group as well as it does the group you used to create it. This process validates the model.

Once you have a model that fits the characteristics of the two-plus category buyers you will be able to use it to score all of the customers in your file. Scoring the file means using the model to assign each single category customer a numeric score that indicates their likelihood of responding to an offer to purchase in one or more additional product categories.

The potential in sales response and revenue if you make an offer that does migrate one-category buyers into two and three categories can be dramatic. For example, you could try an offer to your one-category apparel shoppers of a discount on shoes, handbags or jewelry with every apparel purchase they make out of your catalog. You could see a 20 or 30 percent jump in this groups’ number of orders and sales. You can find this out with a model test on a small scale mailing. Take a sample of scored customers across the whole range of model scores and mail them your test offer. Then compare their purchase amounts with their previous buying results and with what the model projected they should do. You should clearly see whether to roll out your offer to all of your customers or try another one. Once you have tested and expanded your mailings to your modeled customers you might find that you yourself want to migrate from using just one model on your customer file to two or more!

Bill Singleton is a manager of analytics and consulting services with The Allant Group in Naperville, IL.