Andrew Wiseman explores why even statistically robust customer segmentations can fail to create impact, and why successful segmentation depends on purpose, stakeholder engagement, activation and turning insight into action.
There are few sentences in research quite as dispiriting as “We did a segmentation a couple of years ago…”. You can almost hear what comes next. Someone vaguely remembers the segment names. Somebody else thinks there were five, although it might have been six. Eventually, someone finds the presentation on the shared drive and opens it for the first time since the launch workshop.
The customer segmentation itself might have been excellent. The analysis was robust. The groups were genuinely differentiated. The profiles were rich, distinctive and statistically sound. Yet nobody is using it. So what exactly was the point?
Increasingly, I think most segmentation failures aren’t analytical failures. They’re failures of purpose, engagement and activation. Customer segmentation only creates value when organisations use it to make better decisions, prioritise audiences and deliver more relevant experiences.
Start with “why”. Not because everyone else has one.
One of the biggest warning signs is a segmentation brief that starts with “We think we need a segmentation.” Why?
Sometimes there’s a very good answer. The business needs to prioritise audiences, shape propositions, develop more relevant communications, identify growth opportunities or personalise CRM. Sometimes it’s considerably fuzzier. The old segmentation feels a bit dated. A competitor has one. Someone senior has decided it’s the sort of thing a proper marketing organisation ought to have. None of those are especially good reasons to spend a lot of money building one.
Before worrying about algorithms, sample sizes, the number of segments or what we might eventually call them, there’s a much more important question: What decisions do we need this segmentation to help us make? If we can’t answer that, there’s a decent chance we’ll end up with something interesting rather than useful.
Is the clustering actually the least important part of segmentation?
Researchers can get terribly excited about the method. K-means. Hierarchical clustering. Two-Step. Latent Class. Distance measures. Fit statistics. Cluster stability. All important. We absolutely need confidence that the solution is robust and that the groups genuinely differ in meaningful ways. There is also a danger of disappearing down an analytical rabbit hole. Great. You’ve done some beautiful work with Euclidean distance. So what?
A marginally more elegant statistical solution isn’t necessarily a better segmentation if nobody understands it, believes in it or knows what to do with it. I’m not arguing that the clustering doesn’t matter. Of course it does. I am arguing that it’s only part of the job. Once we’re comfortable that the underlying analysis is sound, the really interesting questions are much more human.
- Can people recognise the groups?
- Can they remember them?
- Can they identify them?
- Do they know what each group wants?
- Does knowing someone’s segment actually change a decision?
If not, we may simply have produced some extremely expensive PowerPoint.
Don’t just quantify what you already believe.
Another common trap is the client who says “We already have a pretty good idea what our segments are.” Great. That gives us hypotheses. It doesn’t necessarily give us the answer. There’s a danger that those existing beliefs simply become the quantitative survey. The business thinks customers differ on A, B and C, so we write lots of questions about A, B and C, run the clustering and triumphantly discover groups that differ on A, B and C. Sometimes that isn’t discovery. It’s confirmation bias with a dendrogram attached.
This is where exploratory qual can be invaluable. Not because we’re going to ask 30 people which segment they think they belong to. The role of qual is to uncover the motivations, tensions, language, needs and distinctions the organisation may not already have on its radar. It gives us a much better chance of putting the right stuff into the quant.
Otherwise we can end up with a very sophisticated answer to the question we’d already decided to ask.
Bring people in before you have something to show them.
Stakeholders are too often treated as the audience for a segmentation rather than part of its development.
- Do the research.
- Run the analysis.
- Create the profiles.
- Then unveil it.
The problem is that the people expected to use the segmentation may have had almost no involvement in getting there.
We’ve seen a richer, attitudinal segmentation struggle to gain traction because the business already had deeply embedded customer personas based on much simpler distinctions. The new segmentation was richer. It was more nuanced. That didn’t magically make the old way of thinking disappear.
Marketing may need one thing from the segmentation. CRM may need another. Product teams another. People working directly with customers may see them through a completely different lens.
Bringing those people into the process early doesn’t mean designing the answer by committee. It means understanding how the segmentation will actually need to work in the organisation that has to use it. Without that, the segmentation risks becoming something research has done to the business, rather than something the business owns.
Strategy is not the same as activation.
We’ve also seen segmentations work very well at a strategic level and still struggle to influence day-to-day marketing. Senior teams understand the groups. They use them to think about propositions, priorities and future growth. The segments appear in strategy presentations.
Then someone asks: “Great. Which of these people are actually in our database?”. Silence. That gap matters.
Not every segmentation needs to be CRM-typed. If its job is to guide overall strategy, proposition development or innovation, that may be absolutely fine. If the ambition is personalised campaigning, CRM or different customer journeys, however, we need to think about identification from the start.
Typing questions, database matching and behavioural proxies shouldn’t suddenly become an activation problem six months after the research has finished. Otherwise you can end up with five beautifully differentiated groups and absolutely no idea which one just opened your email.
We’ve seen this evolve in practice with one client. Their segmentation initially worked well for strategy but was harder to translate into campaigning. Once the segments were fully matched back to the customer database, they became much more powerful.
- Campaigns could be targeted.
- Behaviour could be analysed by segment.
- Performance could be tracked.
- Different teams had a common language.
The segmentation stopped being a research output and started becoming part of the way the organisation worked.
Your final debrief is not activation.
This is one of my particular bugbears. We present the final segmentation. Everyone nods. There are some good questions. Someone says how interesting Segment Three is. We send the deck around. Activation complete.
Nope. That’s a debrief. One of the most useful things we’ve found is getting people involved before the final reporting stage. Call it a discovery session. A working session. A chance to kick the tyres on the emerging segments.
The label isn’t important. Put the solution in front of the people who will actually use it.
- Do they recognise the distinctions?
- What surprises them?
- Where do they challenge it?
- What language works?
- What would they genuinely do differently for each group?
- What would make this easier to use?
This isn’t about allowing the loudest person in the room to overturn a statistically robust solution because they don’t like Segment Four. It’s about seeing whether the solution survives contact with the real business before we polish it to within an inch of its life. That can make the final naming, profiling, implications and activation considerably stronger.
Don’t turn a segment into one person.
Speaking of names, I have a bit of a thing about segments built around a single fictional person. Boring Brenda. Savvy Steve. Digital Dave.
Leaving aside the fact that these often sound like the output of a fairly grim Friday afternoon brainstorm, they can actively limit how people understand the segment.
A segment is rarely one demographic type. It might include men and women. Younger and older customers. Different incomes, life stages and backgrounds. What brings them together is usually something more interesting: a shared set of motivations, needs, attitudes or behaviours that actually matter to the category.
Call the segment Brenda and, consciously or otherwise, you’ve immediately given everybody a mental image of Brenda. The 28-year-old man with exactly the same motivations suddenly feels like he shouldn’t be there.
Good segment names should help people remember what distinguishes the group. They shouldn’t accidentally redefine who is allowed to belong to it.
Make the segments feel real.
Avoiding fictional personas doesn’t mean keeping the segments abstract. Quite the opposite. Segmentation outputs can become very abstract very quickly.
Heatmaps. Index charts. Spider diagrams. Pen portraits. All useful. Not always memorable. Activation sometimes means finding much more human ways to make the groups tangible.
For a fitness brand, we’ve explored something as simple as:
“What’s in each segment’s gym bag?“
What are they wearing? What kit do they carry? What role does exercise play in their life?
For a music audience, you can create a playlist for each segment. For another category, build out social profiles. What would people in this group follow? What appears in their feed? Which brands show up? What would they post?
Top Trumps-style cards can make differences between groups instantly understandable. For a nighttime economy or entertainment brand, you can get more playful still. What would each segment wear on a night out? What music gets them going? Which prizes appeal? What does their ideal experience look like?
That may sound gimmicky. It shouldn’t be. The purpose is to translate a multidimensional statistical construct into something people can actually remember.
If a creative team can immediately understand what makes a group tick, that may do more for activation than another 20 charts ever will.
AI makes stress-testing easier. That doesn’t mean it makes it better.
There’s a newer challenge in all of this. It’s now incredibly easy to take some segment profiles, a few tables or bits of statistical output, drop them into an AI tool and ask “Is this a good segmentation?”.
That isn’t necessarily a bad thing. I’m all for challenge. A second pair of eyes can be useful. The problem comes when the answer sounds authoritative despite not knowing what actually happened. The tool may have no idea why particular variables were selected, what was standardised, how outliers were handled, what other solutions were tested and rejected, what the stability checks showed, how the data were weighted or what came out of the qualitative work.
It may not know that a statistically marginally weaker solution was selected because it made far more sense strategically. In other words, it may be stress-testing a simplified reconstruction of the segmentation rather than the segmentation itself. That puts more responsibility on researchers to explain the journey, not less.
The final answer matters. So does explaining why it’s the final answer.
Somebody has to own it, love it and nurture it.
Even a well-activated segmentation can slowly disappear if nobody is responsible for keeping it alive. Someone needs to own it. That might mean making sure the segments appear in campaign briefs. Keeping CRM typing up to date. Helping new starters understand them. Bringing them into proposition and product discussions. Refreshing the materials when the business changes.
Without ownership, segmentation can quickly become everybody’s tool and therefore nobody’s job. It also shouldn’t be frozen in amber. Customers change. Markets change. Categories change.
That doesn’t mean rebuilding the segmentation every 18 months because someone has got bored of the colours.
- Sometimes the typing needs recalibrating.
- Sometimes the profiling needs updating.
- Sometimes new behavioural data can make it more useful.
- Occasionally, the underlying structure genuinely does need another look.
The point is to treat segmentation as an ongoing business tool rather than a research event with a completion date.
So what?
A successful segmentation isn’t defined by whether the clustering algorithm worked. It isn’t defined by the quality of the final personas. It definitely isn’t defined by whether everybody enjoyed the debrief.
A good segmentation activation gives an organisation a genuinely useful way to understand customers differently and make better decisions as a result. That means asking some harder questions right at the start.
- Why do we need it?
- Who needs to use it?
- What decisions will it change?
- How will people recognise or identify the segments?
- What will make them memorable?
- Who owns it afterwards?
Perhaps the first question on the next segmentation brief shouldn’t be “How many segments do we need?” It should be “What are we actually going to do differently once we’ve got them?” Because if nobody has a decent answer, we can probably save everyone some time and create a folder called Segmentation 2026 on the shared drive now.
At Mustard, we think segmentation starts with the business problem, not the clustering algorithm.
We involve the people who need to use it, think about activation while we’re designing it and, where needed, make sure it can actually live in CRM rather than merely in PowerPoint. We’re also quite partial to a playlist, a Top Trumps card or a well-packed gym bag if that’s what helps make the segments stick.
If your current segmentation is spending a bit too much time gathering dust on the shared drive, we know an agency that might be able to help.