Part of what I do professionally involves attempting to identify groups of people who act a certain way, so that businesses can meet their specific needs. The marketing term for this is “customer segmentation,” and it’s a powerful set of tools for setting marketing strategies.
One problem with customer segmentation, however, is that when we carve out a particular segment of customers because they exhibit Behavior X on average, we accidentally include a bunch of customers who do not exhibit Behavior X, because they diverge from the average. Sometimes they even exhibit precisely the opposite behaviors. Just because you own a smart phone and an Xbox and happen to be a male aged 27 years doesn’t mean you’re going to watch the Super Bowl this year, even though most people who are like you on paper probably will.
Or, to put it pictorially:
If you run a business, then you don’t actually care who gives you money (for the most part), you just want them to give it to you. Finding out what your key customer segments are is a means by which you attempt to identify people who have a high probability of giving you money, subject to the assumption that demographic data is predictive of a person’s giving you money. It also helps give you a language with which to communicate to them.
But that’s all it is. Your prediction may very well be wrong.
In business, this means that a lot of the people who get your direct-mail-or-whatever will completely ignore it and not give you any money. That’s because, despite your smart phone and Xbox ownership, your sex, and your age, you might not want to watch the Super Bowl; or, despite your looking a lot like the average childless male gunfighter, you might actually be a female customer service rep who is raising her grandchild; or, etc., etc.
Notice: It is quite often the case that many individual members of the nebulous group of people we believe to have a certain set of characteristics do not actually have any of those characteristics at all. Our customer segment includes both groups of people.
The error a lot of marketeers make occurs when they observe the multitude of behaviors exhibited by a particular segment that do not seem to cohere in a way that can be exploited through marketing strategies. They first think that they didn’t get the marketing message right, so they invest a lot more money into getting it right. When that doesn’t work, they decide that they got their segmentation wrong, and then go looking for “the real segment.” (When they find it, the process starts over again.)
Either way, for a long time, they’re stuck in their segmentation paradigm. Whatever else they choose to change, they don’t change the paradigm. The problem here is that, at a certain point, the marketeers have conflated their customer segment with what they really want to know. They don’t really want to know “Who is in Segment A?” Instead, they want to know, “Who is going to give me more money?” Segmentation is just a means to an end. It’s just a paradigm.
So it goes with all forms of segmentation. It’s tempting to describe the world – and especially human interaction – in terms of “people segments,” subsets of the population who we believe, on average, to behave a certain way. As I stated before, though – and I don’t want to belabor the point, but it bears repeating just one more time – any time you choose to create a segment, you include large swaths of people who behave very differently than the segment itself is said to behave. What this means is that, when you aim to talk about a segment – even your own – you wind up being terribly wrong about a good number of people in that segment.
So, what good is a segment?
Segmentation can be useful as a sort of mental model: If all individual behavior more or less corresponded to the behavior we wish to assign to a group, then what does the world look like? Once we’ve answered that question and come up with a rudimentary view of the world, we can then think about relaxing some of our assumptions, making the “segments” a little fuzzier, and assessing what that does to our basic world view.
This only works, however, if we agree to at least two rules.
First, we shouldn’t make the mistake of conflating our segment-oriented mental model with the actual state of the real, physical world. It might be useful to pare things down a little bit to gain some grasp over the broadest strokes, but the simple fact of the matter is that the real world is more complex than our mental models.
Second, as we relax our assumptions – and we must relax them – we must also agree to test how relaxing them impacts not only the behavior of our model, but also its credibility. If your model of how Purple People related to Blue People only explains, say, 30% of the interactions between Purples and Blues, and only does so if we agree to use the loosest language possible, then perhaps its time to reconsider whether Purple-versus-Blue is the right cut of the data. Maybe it’s Purple-versus-Green. Maybe it’s Triangles and Circles instead. Purple-versus-Blue might be the story with the most emotional power, it might be the story that wins you the highest number of friends or blog followers, it might be the story that wins the largest quantity of grant money, and it might be the one that everybody wants to hear about.
But it might still be wrong nonetheless. As one who puts people into boxes, it’s your responsibility to ensure that your boxes are worth anything.