Developing an Effective Enterprise-Wide Segmentation

Editor’s Note: Every year, participants in the GreenBook Research Industry Trends (GRIT) Report survey vote for the most innovative companies in market research. In Insights That Work, these top companies share the real-life challenges and solutions of their biggest clients.

---

Research Challenge

Rogers Communications Inc., one of Canada’s largest telecommunications providers, needed a robust, insightful and actionable segmentation that could be applied to their entire database of customers. The challenge was to build a segmentation with both behavior and attitudes but with the ability to predict based solely on customer information file data.

Customer clusters built only with available customer information file data enable companies to have a segmented view of their customer base. However, these clusters leave marketers scratching their heads as to how to understand them without rich attitudinal data. Conversely, traditional survey-based segmentations deliver rich personas but are difficult to action with current clients without extensive and continuous surveying.

Rogers came to Dig Insights with a unique request: A segmentation that was built with both attitudinal and behavioral data that allowed segment prediction to performed with only one or the other data sources. Rogers turned to Dig due to combine our strengths in survey-based research, analytics, and data science.

Research Solution

The joint Dig and Rogers project team conducted an online survey of 27,000 Rogers customers and 8,000 general population category users. Concurrent with the survey, we conducted an in-depth review of the behavioral data variables available at the customer level from Rogers databases. After an exhaustive audit, a selection of over 300 variables was chosen, transformed (when needed) and appended to the same dataset that contained the survey responses for Rogers customers.

Multiple cluster solutions were tested based on combinations of survey and behavioral data variables until we were able to find a solution that both met our criteria for an effective segmentation (substantial, differentiated, relevant, accessible & responsive, stable & replicable), and was also highly predictable with only either survey or behavioral data (as measured by cross-validated results).

After a detailed analysis of these segments, multiple prediction algorithms were created using Random Forest models. These models were then operationalized both directly on the Rogers database and in ongoing surveys.

Research Outcome

Rogers can leverage this segmented view across any line of business and product within the organization and is able to track progress against these segments in their continuous tracking studies (with other MR firms). A tangible example of using this data to drive strategy is a recent study to investigate how the segments behaved prior to upgrading their phone.

We were able to identify significant differences across each of the customer groups. Some of these differences were expected (e.g., type of phone upgraded to). However, we also identified several behaviors that were more surprising, including: a gap in the number of interactions with Rogers from one segment to another, the response rate for specific email communication and the number of channels (including their digital footprint) that customers researched prior to making an upgrade.

Future offers have been tailored with greater success for each segment.