What they think, or what they do? Where do you start with a segmentation?

I should start today's post by declaring a bias. I think segmentations are among the most useful, but also among the most abused tools available to marketers.

Useful, because you've got far too many and far too wide a variety of customers to market to each of them individually - particularly above the line. A segmentation makes marketing manageable.

Abused, for a variety of reasons, including...

• Behaving as if all of the consumers in a segment are the same.
  
  
• Failing to understand that an index towards high income men in one segment doesn't mean that everyone in that segment is a high income man.
  
  
• Once you've got a segmentation, abandoning any insight beyond the 'pen portrait' one-page summary of each segment. Or worse, beyond the two-word descriptive segment name.
  
  
• Forcing every business issue through the segmentation, whether it helps or not.
  "My customers are defecting! Quick, find out which segments are defecting fastest!!"
  

With that out of the way, how should you build your segmentation? Let's look at two approaches - basing it attitudinally (on what the customer thinks) or behaviourally (on what the customer does.)

Sometimes, the decision is made for you. If your product hasn't launched yet, or if you're in a market without behavioural data, then it's probably going to have to be mostly attitudinal.

But if you have a customer behaviour database - if you're a bank, insurance company or other brand with a direct customer relationship, I really don't understand why you'd start with data on attitudes - and many companies do. Much better to profile behaviour, create segments, and then find out why your customers are behaving in a certain way.

By all means find the attitudinal reasons why customers are defecting, but doesn't it make more sense to start with the group of customers who are doing it?

There are five big advantages to a behavioural starting point, if you have the opportunity to take this approach.

• It's repeatable. If you have a customer database, you'll need to refresh the segments at some point. The definition of 'lapsed' is unambiguous. You can find out afterwards why they're lapsed.
  
  
• You get as much (if not more) attitudinal difference between your segments by basing them on behaviour as by basing them on attitudes. Seriously, if you get the chance, try it.
  
  
• Pick the right behaviours and your segmentation immediately reflects your business goals.
  
  
• You only get to survey some of your customers to create attitudinal segments, then you have to model where the rest of your customers go, based on probability. Every customer has a behavioural history, so if you say lapsed you mean it. If you say 'Technophile', then some of them won't be.
  
  
• It can be a hell of a lot cheaper, as your surveys can be smaller and you don't have to model the surveys back onto the customer database.
  
  

Not everyone will agree, which is great. But I feel better for getting that off my chest...