The Data Challenge
When designing the custom segmentation system, levels needed to be built in that could be reclustered or collapsed into a small number of segments. The Union-Tribune needed a system that would “roll up and down” to these different levels to provide consistency across the organization.
Integras worked with the Union-Tribune to understand its goals and objectives; the data inputs required by these goals; and its budget, resources and implementation requirements. It was important that the segmentation system be able to serve the needs of multiple departments that handled various applications: acquisition, retention, cross-sell, churn, response, etc. The segmentation system, thus, would need to predict multiple behaviors. Typically, the more behaviors a company incorporates into a custom segmentation system, the more the resulting system gives up in terms of predicting any one behavior. It’s always a trade-off.
On the upside, when multiple behaviors are considered together, the model creates segments that have greater meaning across a wider range of departments and applications. On the downside, the system creates segments with less differentiation for any one given behavior.
Careful consideration was given to these pros and cons, and it was determined that the corporate-wide integration advantages would outweigh the loss of discrimination that might come with building one-off models for each application/department.
Building the System
The Union-Tribune provided Integras with data from three sources: an internal household-level customer file; a proprietary readership study; and the local do-not-call list. The customer file was instrumental in providing key information to create the custom segments.
By extrapolating data from the customer file, Integras identified a number of key consumer behaviors including how much a consumer paid for a subscription, length of the subscription, length of time on the file, etc. It was this type of behavioral data that was used to drive and create the custom segments. The attitudinal and needs-based data—derived from readership surveys—were used to tailor the custom segment descriptions. The goal was to define segments that behave similarly and then understand messaging insights that would evoke those behaviors.