By Kevin Gray
Marketers need to segment. We have to find the most profitable consumer groups for our products and concentrate our marketing efforts on these segments. Everyone knows that.
In fact, this general idea, while intuitively sensible, is not without its sceptics. But first let’s turn back the clock and take a quick look at the history of segmentation. The basic idea most likely dates back to antiquity, as it’s hard to imagine that successful Greek or Roman merchants did not size up patrons at least in some informal way. Considerably more recently, in 1924, Alfred Sloan announced that GM would build a car “for every purse and purpose”, with a different brand of car for each segment of the market. By 1931, the company had overtaken Ford to become the largest car manufacturer in the world and soon afterwards Ford also gave up on its one-car strategy.
Even earlier, in 1903, Walter Dill Scott published The Theory and Practice of Advertising, a book drawing upon the work of psychologists like Freud for inspirations about better ways to communicate with consumers. “Man has been called the reasoning animal,” he wrote, “but he could with greater truthfulness be called the creature of suggestion. He is reasonable, but he is to a greater extent suggestible.” Five decades later, in 1956, Wendell R. Smith wrote in the Journal of Marketing that “market segmentation involves viewing a heterogeneous market as a number of smaller homogeneous markets in response to differing preferences, attributable to the desires of consumers for more precise satisfaction of their varying wants”.1 With this article, modern consumer segmentation was born.
Segmentation is a balance of science and art and comes in many flavours, for example geo-demographic, behavioural, psychographic and occasion-based.2 However, two broader kinds are most relevant to our discussion – pre-determined (a priori) and discovered (post hoc). In the former type, the segments have been chosen in advance based on purchase history, claimed purchase interest, demographics or other criteria. An example would be when customers in a data base have been grouped into heavy, medium and light purchasers of a particular product category. Demographics, past purchase history and an assortment of other data are then employed in a predictive model for classifying new consumers into one of these a priori segments. Marketing is then tailored to these consumers, accordingly. Individually-targeted ads and recommender systems are new wrinkles on this basic notion. Predictive analytics is not required, however, and profiling pre-determined segments with multivariate analysis can be extremely informative. More on that a bit later.
With the second type of segmentation, post hoc, segments are uncovered, usually through cluster analysis of one sort or another. A variety of data can be used in the clustering but a popular way centres on attitudinal data. The attitudinal data typically are ratings of psychographic measures or importance attributes in a consumer survey. The resultant clusters (segments) are then cross tabulated against demographics, claimed purchase behaviour or “hard” data when available. Alternatively, these key marketing variables can be included in the clustering along with attitudinal data in what I call full-profile segmentation and some others multi-domain segmentation. There are many variations of this approach, including those utilizing customer data bases or web browsing history.
Both kinds of segmentation are frequently combined – post hoc segmentation is conducted first followed by predictive modelling, with the discovered segments used in place of a priori segments. These segment classifiers are often dubbed “typing tools” in marketing research. Segmentation of either kind is now almost considered common sense to many marketing researchers, as I’ve mentioned. That said, how competently segmentation is conducted is absolutely critical and many segmentations turn out to have no real business value. It’s not a simple data processing task and the data and statistical methods used, how they are used and how the results are interpreted can make or break a segmentation.3
Another reason for scepticism stems from how segmentations are actioned. Think about it for a moment; ideally, when designing and marketing a product, one would normally hope it has broad appeal, not just to small consumer niches. It is true that some products may be intentionally designed for specific demographic groups, for example, fashion items aimed at young women with high disposable incomes. Medical equipment obviously is not mass-marketed. Other products may accidentally prove more appealing to some consumer groups and we only discover why later. However, we need to be wary of generalizing from exceptions such as these. The reality is that the extent to which consumers are brand loyal has probably been exaggerated and the majority of brands are not primarily bought by small groups of consumers, irrespective of our targeting. (See How Brands Grow Part 1 and 2 by Byron Sharp and his colleagues for a detailed if controversial overview of this topic.)
Does this then mean that segmentation is actually a waste of time and money? No, because small differences in purchase propensity and frequency will often have a substantial impact on the bottom line. Problems arise, though, when a product that would naturally appeal to general consumers is erroneously marketed to a small group. It’s important to understand that demographics and other variables typically used to access target consumers are really proxies for attitudes and behaviours. A brand may draw disproportionately from men aged 25-34, for instance, but seldom exclusively. Other consumers with similar wants, needs and habits – but different demographics – may account for substantial sales volume and perhaps would account for more were it not for narrow (and wrong-headed) targeting. This would be a costly misuse of segmentation, and we should bear in mind that even a segmentation which is very good statistically can backfire if used inappropriately.
There is another often-forgotten but extremely valuable reason for conducting segmentation – to learn about consumers. Segmentation need not identify sharply-defined, easily-accessible target groups to prove its worth. The reality is that most consumer ‘segments’ are conceptual, not objective entities. However we conduct it, segmentation usually boils down to partitioning a multi-dimensional data file in a way we believe has business value. Moreover, remember that segments are not entirely homogeneous, only more homogeneous than general consumers.
Understanding how consumer behaviours, demographics and attitudes interrelate, however, by itself can speak volumes about how to communicate with consumers, about customer needs and gaps in the market, and about opportunities for new products and services. Segmentation, either a priori or post hoc, is an excellent way to develop knowledge and insights from data and carries with it much less risk than data dredging by running hundreds of ad hoc cross tabulations.
Very broad definitions of product categories are usually not actionable but we should also take care not to define our category too narrowly – carbonated beverages, sports drinks, mineral water and fruit juice, for example, often compete with one another even if we marketers lump them into separate product categories. We must think like consumers, not like marketing researchers. Furthermore, be especially careful to avoid targeting consumer groups simply because they seem like logical targets – these consumers may actually be lucrative targets but it’s risky to merely assume they are without having evidence that they really are.
In summary, there is a multitude of ways to undertake segmentation and it’s wise to explore approaches other than what you are accustomed to. Do not ignore segmentation as a way to learn about consumers. It also can provide important insights for branding, creative and execution even if we decide not to target particular consumer segments.
By Kevin Gray
Kevin Gray is president of Cannon Gray, a marketing science and analytics consultancy. Kevin wishes to thank Peter Fader, Professor of Marketing at the Wharton School of the University of Pennsylvania, for his helpful comments on a draft of this article.
1 This brief history of segmentation has been adapted from History of Segmentation.
2 See Wikipedia for concise overview of segmentation. A classic reference is Segmentation & Positioning for Strategic Marketing Decisions by James Myers.
3 One example of how method can affect results is clustering on principal component (“factor”) scores. Though common practice, this approach is biased towards finding segments of roughly equal size. Another example is the use of attitudinal items or demographics that are empirically unrelated to consumer behaviour in the category being researched.