When the Segments Disagree

Segmentation finds the forks, not the answer

Most audience segmentation gets treated as a sorting problem. Define the personas, pick the primary one, design for them.

Running it as a perception test changes what you get. You are not asking each segment what they want. You are showing every segment the same range of elements and watching where their responses land. Sometimes the clusters sit on top of each other. Sometimes they don't. Both results tell you something you can build on, and the second one tells you something you would otherwise learn after launch.

Overlap is the more valuable finding, and it is the one teams skip past on the way to the differences. If executives, employees, customers and prospects all cluster on the same forms and textures, those elements can carry the brand. They are the load-bearing part. Nothing downstream has to hedge them.

Figure 1 · Two segments, one set of elements

What a split tells you, and what to do about it

A divergence between segments is a fork in the decision, not a verdict.

On a rebrand, the move is usually to take what everyone aligns on and make it core and central. Where segmentation shows a difference, mark it. That is a place where you can shift, or lean slightly, without breaking the whole. If a particular segment is heavy with your key buyer persona, you can weight the core toward what they responded to and let the other segments sit a little further from center.

On a campaign you have more room. The brand holds, and the campaign leans hard into whatever the target segment showed you. That is a targeting decision made against evidence instead of a persona document written eighteen months ago.

Figure 2 · What to do with a fork

Either way, the split gives you something to be deliberate about. Without the segmentation the two groups' responses average together into a middle that represents nobody, and nobody in the room knows the average is hiding a fork.

The split that settled a rebrand

The clearest version of this in Constellations' own work came out of a capital construction software rebrand with 22 stakeholders on it.

Sales and marketing were describing the same company in incompatible terms. Sales talked about the brand like Captain America. Marketing sounded like Mr. Spock. Both groups were confident. Both had reasons. In an ordinary process that argument gets settled by whoever outranks the other, or by a compromise that splits the difference and satisfies no one.

They tested it against 500 employees instead. The data showed that disruption was the wrong idea for a company selling into infrastructure, which is a conclusion neither camp had arrived at on their own. One direction came out of it unanimously, with no revision rounds, and sales rose 33% in the twelve months after launch.

The disagreement between two internal segments was the most useful thing in the project, because it identified exactly where the brand's meaning was unsettled and pointed the test at it.

What segmentation looks like in practice so far

Most segmentation in Constellations tests runs by channel. You generate a link for a group, send it to them, and their responses stay identifiable inside the larger result set. Different audiences show minor leans one way or another, which is what you would expect.

One test went out to industry professionals in different specializations: marketers, brand agency people, UI and UX designers, an academic group of faculty, and a more technical group. These were small individual data sets, not large samples, so the read is directional at best.

What showed up anyway is worth noting. Every group aligned on visual puns, the kind of clever construction you see in marketing work. That question was straightforward: which of these looks interesting to you, and which looks least interesting. The professional and academic groups landed close to each other, favoring more illustrative and traditionally rich brand work. The technical group leaned noticeably more toward a tech aesthetic, which is the one real outlier in the set.

There isn't a case to build on that yet, and it would be dishonest to present a handful of small groups as a finding. It is a hypothesis worth testing at real scale, which is the only honest thing to say about a sample that size.

The rule of thumb for what to segment

If you are even slightly interested in whether a group sees things differently, split them out.

The reason is asymmetric cost. Segmenting is cheap. You generate a channel link, send it to that group, and you can toggle their results on and off, viewing them alone or blended with everyone else. If the split turns out to be boring, you have lost nothing and you can look at the combined result exactly as if you had never segmented.

Figure 3 · The asymmetry

Not segmenting is expensive in a way you cannot see. Once responses come in through a single undifferentiated pool the groups are gone. If executives and customers pulled in opposite directions, that fact is now averaged into a middle position that reads as consensus, and you will never know it wasn't.

Reasonable segments to separate:

• Internal versus external. The two groups have different exposure to the brand and different reasons to be loyal to it.
• Executives versus staff. The gap here is often the one quietly driving the revision cycle.
• Existing customers versus prospects. Heritage to one group reads as dated to the other, and that fork matters most in a logo redesign.
• Buyers versus users, when those are different people.
• Any channel with its own audience composition. The list you email and the audience you advertise to are rarely the same crowd.
• Any group somebody in the room has a strong assumption about, because assumptions about groups are exactly the thing worth checking.

You can always blend them back together. That is the whole reason to do it.

Frequently asked questions