Why Segmentation Is Sometimes Useless
Dividing a market into unique segments makes sense. But the statistical methods we rely on for segmentation often result in segments that are strongly differentiated in useless and misleading ways.
A recent analysis of Facebook data by The New York Times illustrates this perfectly. The NYT analysis claimed to reveal “the hidden patchwork of global travel” in the United States by highlighting the “unusually popular” foreign destinations of Americans state-by-state. It revealed hidden insights like this: People in South Dakota travel to Liberia. People in Utah travel to Tonga. People in Virginia travel to Bolivia.
Then as you scratch your head wondering if this can possibly be true, the authors explain that these are not necessarily popular destinations for these states but rather “unique” destinations “compared with where most Americans travel during the summer” (italics added).
As it turns out, there are about 15,000 people of Tonga descent living in Utah. That’s one half of one percent of the state population. Tiny, yes, but unusually large compared to the number living in Illinois. Even just a handful of them paying a visit to Tonga will make Utah look like a hotbed of travel to exotic Pacific islands, compared to Illinois.
That is precisely the problem with segmentation algorithms. They find things that strongly differentiate, even if those differentiators are trivial. Unfortunately, we then leap to definitions and characterizations of segments based on those differences. Even worse, our business colleagues then commit to marketing strategies and campaigns based on our newly revealed “hidden insights.”
If you are doing segmentation, please remember this: the segments that will make most sense you probably already know about. Good data analysis can help you understand, elaborate, and develop excellent strategies for those segments. But don’t count on fancy segmentation algorithms to discover them for you.