While every customer has their own story, creating segments of size one is usually impractical. Instead, we want to create segments that are “similar within and different across” while at the same time “keeping them REAL” — Relevant, Efficient, Actionable and Lasting:
- Relevant: Segments should be defined on attributes that can explain differences in customer behavior. That is, there should be some plausible explanation for the uniqueness of the segment causing a behavior rather than just co-occurring with it. For example, a geographic segmentation that results in customers from Miami buying more beach towels than those from Montreal is better than a geographic segmentation that results in customers from Miami buying more yellow highlighters than those from Montreal.
- Efficient: As mentioned before, creating segments that are too granular is impractical and misses out on the opportunity to identify common behaviors and preferences. While there is no “correct” number of segments, you want to have few enough so that you aren't overwhelmed by the overhead associated with treating the segments differently.
- Actionable: Similarly, marketers should be able to measure and act upon the segmentation dimensions. For example, an online car rental site may have anecdotal evidence that indicates that taller-than-average customers are more likely to choose sedans with extra legroom over compact cars, but if the company doesn't know that up front and/or can't collect that information easily, then segmenting on customer height is not very useful.
- Lasting: Good segmentation relies on dimensions that will remain relatively stable over time. Segmenting customers based on how they responded to a one-time promotion is useless for categorizing customers who weren't exposed to that promotion. If you want to track the performance of these segments over time and the customers in that segment are changing very quickly, any comparisons you make will be meaningless.