Why so many customer segmentations die in a slide deck

You may have seen this happen, or paid for it. The segmentation presentation goes well: seven segments, each with a name, a portrait and a page of needs. The deck goes round the company the same afternoon. A few weeks later, a sales manager asks which of their accounts belong to which segment, and nobody can answer, because the segments were built on attitudes measured in a long survey and none of those answers live in the CRM. By the next planning cycle, sales has gone back to sorting customers by size and region.

Good statistics don’t save a segmentation on their own. Research on segmentation has pointed at the same weak spot for about thirty years: much of the trouble shows up when the segments have to leave the deck, and some of it starts earlier, in the data and in the way the project is set up. Sally Dibb and Lyndon Simkin documented the practical problems of implementing segmentation in industrial markets in 1994.¹ Yoram Wind and David Bell, then both at Wharton, write that the most difficult part of any segmentation project is translating the results into strategy and programmes.² And when Sara Dolnicar and Katie Lazarevski surveyed 167 Australian marketing managers who’d worked with segmentation, 68% agreed that the segmentation solutions presented to them were like a black box.³

Why this is a challenge

That research describes many barriers. Three come up often, and they’re the ones a buyer can do something about.

The segments are built on what people think and handed to teams who see what customers do. Attitudes and needs are useful raw material for a segmentation, because they help explain why people choose. But a sales team sees company size, sector, order history and the last email. In a study of the European car-parts aftermarket, Dibb and Robin Wensley found that even a fairly wide set of structural variables explained only a very small part of how customer requirements varied, and that segments built on those requirements didn’t map easily onto dimensions the company could act on.⁴ If nothing connects the attitudes to the data people can see, the segments stay inside the survey.

They’re tested on the model, and often not on the people who’ll use them. A cluster analysis can find seven groups that are statistically distinct. Whether a salesperson can tell them apart on a Tuesday afternoon is a different question, and it’s often left until after delivery. Dibb and Simkin noted in 2010 that the literature describes criteria for judging the quality of segments, but there’s little evidence that companies use them.⁵ When Lee Quinn studied ten UK fashion retailers, the managers preferred simple, descriptive segmentations they could use to make sense of the market and explain it to others.⁶

The project ends at the presentation. The research firm hands over the deck, the client team moves on to the next priority, and nobody owns the work of turning segments into scripts, offers and targeting rules. Wind and Bell, drawing on their own experience, list among the conditions for success involving everyone who’ll make decisions with the segmentation, sales management included, from the moment the problem is defined.² Dibb and Simkin’s 2001 work groups the barriers into three kinds: the infrastructure in place before the project starts, the segmentation process itself and implementation.⁷ Dibb’s later work splits them into harder ones, like data, money and people, many of which respond to early treatment, and softer ones, like culture, leadership style and coordination between departments, which are more difficult to handle in the same way.⁸

The one-minute test

Here’s a simple rule of thumb. Pick a real customer and ask someone in sales or product to place them in a segment, using the CRM and a couple of questions they could ask in a first call. If it takes much more than a minute, or needs a survey, the segmentation probably isn’t finished.

That sounds strict, but it’s the standard you’d apply to any other tool. A pricing model that only the analyst can run is a pricing model the company rarely uses.

You don’t have to give up depth to pass the test. The segments can still be built on needs and motivations. What changes is that each one also comes with a short list of signals people can actually see: a handful of CRM fields, a few questions a salesperson can ask in a first call, the channels where that kind of customer turns up. In quantitative research this is called a segmentation typing tool, and the questions are sometimes called golden questions. Expect some errors. Wind and Bell point out that placing a customer in a segment is a matter of probability, a fact many applications seem to gloss over,² and in Dibb and Wensley’s study the structural data explained very little of what customers needed, so the short questions often do most of the work. A good typing tool comes with a figure for how often it places accounts correctly. Whatever you call it, the typing tool is the bridge between the study and the people who are meant to use it.

“We already have years of sales data”

It’s a fair objection. Sales data tells you what customers bought, when and how much, and it does that better than any survey. Behavioural segments built from that data pass the one-minute test by design. In a study of five UK service companies, Christine Bailey and colleagues found that segments were still used to choose customers, design offers and plan mass communications, while one-to-one communication relied more and more on propensity models and triggers.⁹

What sales data can’t tell you is why the customers who left, left, why the ones who stayed nearly didn’t, or what the people who never became customers bought instead. Years of sales data are a detailed record of what your customers did. The reasons are rarely in the CRM in a form you can analyse, and the people who never got as far as a sales conversation aren’t in it at all.

The research points the same way from the other side. Dibb and Wensley’s finding that structural data explains little of what customers need is one half of it. The other half comes from Daniel Yankelovich and David Meer, who argued in Harvard Business Review in 2006 that attitude-based segments built for advertising were no better than demographics at predicting what people buy, and that a segmentation should earn its keep by guiding real decisions: which products to develop, which channels to use and what to charge.¹⁰ A useful segmentation needs both sides, the motivations that only research uncovers and a link back to the data the company already holds.

Three checks before you accept a segmentation

None of these is complicated, and they sit alongside the statistical checks, such as whether the same segments come back when the analysis is run again.³ They’re often skipped because they happen after the part that looks like work.

1. Twenty real accounts. Take 20 accounts from your CRM, ideally ones that took part in the survey, and place each one with the typing tool: the CRM fields plus the short questions. If you need the full survey, or the tool gets many of them wrong, the segmentation goes back to the analysis.
2. Two salespeople who weren’t in the project. Give them the segments and ask each one to name real customers for every group. If both can, and they put different customers in different segments, the segments make sense to the people who’ll use them. If they struggle, or the same customers fit several segments, the segments need more work.
3. The way each segment talks online. Compare the segments with social listening data. If the groups use the same words, worry about the same things and gather in the same places, the differences may exist in the questionnaire and nowhere else. Find each segment online through what you know about its members beyond their survey answers, such as their roles, interests or communities, and then compare what they say. In B2B markets where buyers say little in public, this check tells you less.

The checks follow the logic of our H3™ method: one source gives you a hypothesis, and each independent source that agrees makes it stronger. When the survey, the people who’ll use the segments and social listening agree, you can trust the segmentation a lot more. When they disagree, the disagreement can be the most useful finding of the whole project.

What to ask before you buy one

If you’re about to commission a segmentation, six questions will tell you a lot about what you’ll receive:

• How will my team assign a new customer to a segment, and how long will it take them?
• Will you test the segments on our own accounts before the final presentation?
• If you run the analysis again, do the same segments come back?
• Who from sales and product will be involved, and at which stage?
• What does each segment need from each team: marketing, sales, product and service?
• How will we know, six months from now, whether the segmentation is being used?

A firm that has thought about these will have the answers ready. A firm that hasn’t will only talk about the statistics.

A segmentation earns its budget on the day a salesperson uses it without opening the deck.