• Follow Curiously Persistent on WordPress.com
  • About the blog

    This is the personal blog of Simon Kendrick and covers my interests in media, technology and popular culture. All opinions expressed are my own and may not be representative of past or present employers
  • Subscribe

  • Meta

Segmentation is not a science

An Advertising Age blog post inaccurately entitled “The Death of Customer Segmentation” (those pesky subs) argues that traditional market segmentations should be combined with “self-segmentation” techniques such as user recommendations, networking groups, opt-in alerts and consumer generated content/feedback.

The author, Michael Fassnacht, finds traditional segmentation problematic because:

  • It is too static in a fast-paced society
  • People can belong to different segments at different points
  • Consumers want more control over their marketing activity

He believes that incorporating these self-selection elements will empower the consumer to pick the most relevant messages for themselves.

I can’t quite articulate my problems with this article but they broadly fall into three areas

1. Segmentation is not a science

Segmentation is an art. It is about interpretation. It is about tendencies. It delivers a framework; not a silver bullet of success.

Segmentation analysis is created through factor and cluster analysis. . There is no correct answer – the final solution is decided by weighing up the coherence of each cluster against the ability to manage the result (ie a twenty cluster solution will produce more coherence than a four cluster solution, but twenty targets are harder to manage than four).

Clusters are only indicative; they aren’t precise or mutually exclusive. To illustrate this, I will use my favourite segmentation example: Phones 4u.

Take “Flashing Blades”:

All these descriptions will be tendencies. Flashing Blades are more likely to be these demographics, and more likely to follow this lifestyle. Otherwise, it would be a segment of about six people, not 1.68m. So within this group there will be some women, some pensioners, some teetotallers and some pacifists.

So, those identified as “Flashing Blades” aren’t pure “Flashing Blades”. They are just closer to this target than the others. Being part of different segments isn’t an issue.

However, a lack of dynamism could be a more valid criticism. In a few situations. Some behaviour patterns change, but it is unlikely that the core of a personality will. So while I have recently changed some of my leisure habits (one reason why my link updates are on temporary hiatus) I am still fundamentally the same person and will likely approach most markets in the same way.

And a segmentation is only likely to focus in one market, despite the complex interactions between them. In that respect, a successful segmentation may be self-defeating; I may be identified as a potential but through adept targeting I become a consumer/user/fan. Does that then invalidate the model or does it mean I can move into a new segment?

In theory I can, but a segmentation model is based on interplay of variables among one group of respondents at one point in time. A template of a few “golden questions” can approximate assigning people into segments, but over time segments are the same in name only – different biases and different tendencies will become primary determinants in cluster behaviour and so the segmentation becomes gradually less insightful and the golden questions less relevant.

While a segmentation is far from perfect, introducing consumer input isn’t the answer.

2. Self-segmentation will be ignored by the majority of consumers

Fundamentally, most people just aren’t that into picking out relevant messages, despite the author’s assertion. We are going back to perspective biases.

Of Facebook’s 200m members, how many fully customise their experience and give certain activities or adverts a thumbs up? Not many, I would guess.

I don’t think I would be incorrect to posit that most see advertising as a necessary evil – we put up with it in exchange for free content. If we had the option to turn off advertising, we would, even if we view some adverts as entertaining or informative.

If this were to happen, opt-in mechanisms would then become useful – people can’t only rely on word of mouth for product information. But I can’t see this happening very soon and more importantly, does the consumer even know what they want?

3. Self-selecting tools are only of limited value

Unless you are an Amazon-esque retailer with massive size and scope, opt-in elements are going to be fairly limited. I can’t see how any meaningful segmentations can occur outside of one person choosing to positively rate one item and another person not rating it.

Even in cases where it may be appropriate, there is still going to be a disconnect between what the consumer wants and what the advertiser wants. For instance, a low income urban male may choose to find out more information about Burberry products. But Burberry want to target Kate Moss, not the Blackout Crew.

Perspective biases can also become an issue – something may improve our lives, but we cannot conceive of it and so no activity around it is registered. The consumer doesn’t always know best.

This is why observation may be the missing element. If traditional segmentation does need to be enhanced, surely behaviouaral targeting tools (I’m thinking more Audience Science than phorm) would be better suited?

Rather than as a supplement to a traditional segmentation, they could be used as a sense-check. Does real-life behaviour match our abstract targets? Again, this is only going to be relevant for certain sectors, but I see it as being far more powerful than opt-in models.

Ultimately,  segmentation models are frameworks. They are not a science and they are far from perfect, but they can be useful in the short-term. In the longer term, they become more difficult to maintain (particularly in fast-evolving industries) but perhaps a degree of behavioural targeting can be used as an indicator for the continuing relevance of the segmentation. It certainly shouldn’t be reliant on consumer opinion.

This post has been incoherent even by my lofty standards, so any thoughts on segmentation would be most welcome…


Image credit: http://www.flickr.com/photos/rogerss1/

Reblog this post [with Zemanta]

7 Responses

  1. (here from @tomewing)

    Most consumer segmentations I have seen turn out to be be somewhat weak as models, because the elements that go into the model are only modestly predictive of behaviour. They are then labelled and used by the marketing department as though they were cast in stone – as you say, far more an art than science.

    Mind you, I’m not sure you’re right about Burberry targeting – last I looked, low-income urban males were very, very interested in Burberry and indeed in Burberry knockoffs.

  2. Thanks for the comment Alison. As you allude to, what you get out will depend on what you put in.

    Re. Burberry, while those young men may have been buying Burberry/fake burberry – the company themselves weren’t too pleased with it. It cheapened their brand and made it less attractive to their core, affluent audience.

  3. Completely agree – I’d go so far as to say that in the majority of cases the segmentation itself isn’t the problem – it’s the expectations & preconceptions of audience. There was an article a few years back (I forget who the author was) showing that a major factor in the success of a segmentation was having an experienced & well-backed internal segmentation team.

    On self-segmentation, we know that people only classify themselves into the segment we put them in about half the time. Of course, we all think & hope that the differences are down to segmentation offering an alternative, indirect viewpoint, rather than any issues with the process…

    One other point – the same person can indeed belong to different segments, but segmentation is increasingly about grouping together moments in time, rather than (or as well as) individuals.

    p.s. say hi to Mr Hurry from me!

  4. I just began reading How Customers Think by Gerald Zaltman, which is helping me think about this. My problems with Fassnacht’s article are (1) the implication that because technology is changing, people’s values are changing as a result and (2) language like “moving through segmentation” as if purchase behavior/decisions were the result of a linear process.

    Those who have figured out that the fact that someone states a preference or interest in a product has little bearing on that person’s purchase behavior have come a long way. But we spend too much money and time looking at “what” and not enough time looking at “why.”

    And, as you assert, asking people why they do what they do won’t be of much help. Zaltman says that, for the most part, we make decisions before we are consciously aware of having made a decision. To the extent that consumers are able to articulate why they do what they do, it’s only an after-the-fact rationalization based on the rational/conscious part of the decision, which is only 5% of the decision. He says, “The areas of the human brain that involve choice are activated well before we come consciously aware that we’ve made a choice. That is, decisions ‘happen’ before they are seemingly ‘made.’ In fact, unconscious judgment not only happen before conscious judgment, but they guide them as well.”

    Zaltman says that we (both marketers and as consumers) make decisions as a result of an integration of brain, mind, body, and societal factors. Too often, marketers look at only one of those factors and forget how it is inextricably linked with the others. (This includes how we, as marketers, decide which factor to give credence to.)

    The reason approaches like Amazon’s work, I think, is because they give consumers the the ability to navigate the Amazon site in a way that feels somewhat intuitive to the user. I’m not convinced that Amazon “segments” its consumers any more than Google does. It sets up processes that make it easy for consumers to find what they’re purposely shopping for, and, based on the choices consumers make, offer up ideas for consumers to recognize additional items they might want to purchase. It’s essentially a directory of purchase options that enables consumers to shop every-which-way.

    I compare this to the original “consumer segmentation” tool: the Yellow Pages. The Yellow Pages presents purchase options, sorted according to how they think consumers ought to look for them. If some keys my car, I need to figure out that keying–>paint repair–>car paint repair–>auto body. The reason Google is kicking the Yellow Pages’ butt is that Google doesn’t ask me to do all that work to get from keying to auto body; I can just type “keying repair” and eventually get to a body shop that gets what my problem is.

    In other words, and I’m sorry this response is so long and convoluted, for too long, companies that have been trying to segment consumers have really been trying to get consumers to sort their shopping behavior into the categories that suit the companies’ own organizational system, much as the Yellow Pages has. The challenge now is how to segment consumer purchase behavior in a way that takes into account all the different behavioral choices that are becoming increasingly available — and all of the ways body-mind-brain-societal factors that influence those choices.

    I think that trying to segment consumers is like trying to segment an apple pie. You can cut it into pieces, but once you take a bite, how much of it is apple, how much cinnamon, how much flour? How much vitamin C, how much simple carbohydrate, how much yellow, how much sour, how many calories?

    Part art, part science, much understanding that the two are inextricably interwoven, in the same way that we cannot make good decisions without having fully engaged both the emotional and rational sides of our brains.

    I know this was long-winded. Hope there was something of value here. Thanks for your post.

  5. Dear Simon,

    By all means self segmentation cannot – and should not – replace traditional market segmentation as based on rigorous multivariate analyses, i.e. on cluster analysis principles. Traditional clusters are still very useful and in a way, this is exactly what Amazon does. Although it does not execute a traditional analysis, it entrusts this task to an algorithm which clusters its customers based on their closeness in a graph where links (edges) point from customers to products, the products (or combinations thereof) thus forming clusters that customers are assigned to. Nothing new there, this is still good old cluster segmentation.

    Which leads me to my second point. Although self-segmentation – construed as the process whereby a consumer self-consciously assigns themselves to a given segment – might be void, it does not entail that there are no new and relevant ways to segment consumers. Going back to the Amazon example, this approach still relies on common patterns of consumption within consumers (buying certain products, reading certain magazines, etc.) rather than on the actual and direct connections that exist between individuals.

    In other words, one should add subjective segmentation (individuals knowingly connected to other individuals) to objective segmentation (individuals connected to attributes that other individuals are also connected to). Rather than solely segmenting individuals based on what they buy, where they live, how much they make and so on and so forth, individuals can also be clustered based on who they interact with. This is data that has not really been available before, at least on a large scale, but now these patterns can be observed on the social web. And if you’re aiming for a campaign where you can leverage the power of social recommendations (word of mouth), well, this is a form of segmentation that should be seriously considered.

  6. Thanks all for the extremely insightful comments:

    Traditional segmentation is quite one-dimensional and behavioural targeting information does offer additional information that can both inform and add to clusters. In theory, the Amazon example is self-improving over time, though there is still the issue of static labels being applied to dynamic clusters.To use the Google example, BT in some ways is like moving from the Windows based folder structure to the tagged Google desktop search where pieces of information can exist in multiple places at once


  7. […] you’re interested in reading more analysis of this article, check out Simon Kendrick’s post on the Curiously Persistent blog for a very thorough breakdown of the […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: