Can you oversegment audience data in research?

I'll admit it. I love slicing and dicing data. There's something enormously satisfying about finding an unexpected correlation, discovering that one audience behaves differently from another, or spotting a pattern that wasn't apparent in the topline results. It's part of what makes market research so interesting. And sometimes, those discoveries lead to genuinely valuable insights that might otherwise have gone unnoticed.

But I've also seen what happens when we get a little too enthusiastic.

A marketing team spots an unusual response among younger consumers, and suddenly a presentation declares that "Gen Z doesn't like X." Except when you look more closely, the finding comes from just 15 respondents. Those 15 people may have expressed a perfectly legitimate opinion, but they're hardly representative of an entire generation. And there's no guarantee they're even the consumers the business should be targeting.

The temptation to keep breaking down the data is understandable. We have increasingly sophisticated tools, more variables to explore, and the ability to identify differences that were once difficult to detect. But every additional cut can reduce the number of respondents who support a finding while increasing the number of patterns we can discover.

So where do we draw the line? At what point does the pursuit of greater audience understanding start producing distinctions that are statistically fragile, commercially irrelevant, or simply too numerous to be useful?

Not every segment we discover is real

One of the less obvious problems with segmentation is that statistical software will happily divide an audience into groups, even when the differences between them are relatively weak. Ask an algorithm to identify six clusters, and it can produce six clusters. That doesn't necessarily mean six distinct types of consumers exist in the market.

This is a problem researchers Sara Dolnicar and Dominik Ernst explored in their work on the stability of market segmentation. Their research examined how data-driven segmentation solutions can change depending on the sample being analysed. In a review of 53 published segmentation studies, they found that 49 had not accounted for sampling variability when assessing the stability of their results.

Let's imagine a business commissions a segmentation study and receives five neatly defined audience groups, each with its own characteristics, motivations, and potential commercial value. The profiles appear convincing, and the differences between them seem clear. But would those same five groups emerge if the research were repeated with another sample of consumers? Would the boundaries between them remain broadly consistent, or would respondents move between segments depending on the data used?

This matters because statistical clusters aren't necessarily natural divisions within a population. Consumer attitudes and behaviours often exist along a continuum, with people sharing characteristics across several groups. A segmentation model imposes boundaries on that complexity, which can be useful for analysis but shouldn't be mistaken for evidence that consumers fit neatly into distinct categories.

None of this makes segmentation unreliable. It means the quality of a segmentation study depends on more than how convincing the final profiles look. Testing whether those groups remain stable when the analysis is repeated is essential, particularly when the findings will influence substantial marketing investment.

When small numbers become sweeping conclusions

There's another problem that emerges when we keep dividing research data into smaller groups. With every additional cut, the number of respondents behind an individual finding can shrink, sometimes dramatically. Yet the percentages in the final presentation can look every bit as convincing as those based on the full sample.

Eleven out of 15 respondents becomes 73%. Presented without the sample size, that figure sounds substantial. But change the opinions of just three people and the result falls to 53%. It's a useful reminder of how unstable percentages can become when we're working with very small groups.

The problem isn't limited to sample size. The more comparisons we make, the greater the likelihood of finding something that appears unusual simply by chance. Statisticians Andrew Gelman and Eric Loken explored this issue through what they called the garden of forking paths, describing how different analytical choices can produce apparently meaningful findings even without deliberate manipulation of the data.

This is particularly relevant when researchers explore dozens of audience combinations, looking for differences by age, income, geography, purchasing behaviour, or brand preference. Eventually, something is likely to stand out. And once it does, it's remarkably easy to build a compelling narrative around it.

I'm not suggesting researchers should stop exploring unexpected patterns. A small subgroup finding can raise an excellent question. But just know it doesn't always provide a reliable answer.

Are we identifying different customers or inventing differences?

Even when the statistical evidence is sound, there's another question worth asking. Do the differences we've identified actually matter to the business?

In their Harvard Business Review article Rediscovering Market Segmentation, Daniel Yankelovich and David Meer argued that segmentation had become too focused on describing consumers rather than understanding the purchasing behaviours that brands could meaningfully influence. Knowing who customers are, they suggested, isn't necessarily the same as knowing why they buy.

It's something that can get lost when research produces increasingly detailed audience profiles. Two groups might differ significantly in age, income, lifestyle or attitudes, yet have remarkably similar expectations when purchasing a particular product. Conversely, consumers who look almost identical demographically may have entirely different needs depending on the circumstances in which they're buying.

Researchers at the Ehrenberg-Bass Institute have also challenged the assumption that increasingly narrow targeting is always beneficial. Their work on buying behaviour and brand growth emphasises the importance of reaching a broad population of potential category buyers, rather than concentrating exclusively on narrowly defined audiences.

That doesn't mean segmentation has no place in marketing strategy. A specialist healthcare product, for example, may serve an audience with very specific needs, while different purchasing occasions can justify distinct propositions even within the same demographic group.

On the other hand, if two audience segments ultimately want the same thing, respond to the same proposition, and can be reached through the same channels, perhaps they aren’t as different as our segmentation suggests.

There's a danger in assuming that every measurable difference deserves its own segment, strategy or campaign. At the end of the day, the more finely we divide consumers, the easier it becomes to overlook what they have in common.

The hidden cost of too much precision

The challenges don't end when the research is complete either. In many ways, that's when the consequences of too much segmentation become most apparent.

A business might identify eight distinct audience groups, each with its own motivations, preferences, and purchasing behaviours. On paper, the segmentation looks sophisticated. But translating those findings into marketing activity is another matter. Can the business reliably identify these consumers in its customer database? Can it reach them through available media channels? Does it have the resources to develop different propositions, communications and experiences for each group? And would those efforts generate sufficient returns to justify the investment?

Research published by the American Marketing Association and Kantar in 2025 highlights the disconnect between developing segmentation strategies and putting them to work. While marketers widely recognise the importance of segmentation, only 36% of respondents reported using their segmentation for all its intended purposes.

That raises an uncomfortable question about how much of the complexity businesses commission is complexity they're actually equipped to use.

Every additional segment potentially creates another set of decisions about targeting, messaging, investment, and customer experience. Multiply those requirements across several markets, product categories, and distribution channels, and the operational burden can become considerable. Marketing teams may find themselves spending more time managing distinctions between audiences than developing propositions that appeal to them.

Of course, highly specialised segments can justify substantial investment. In B2B markets, for example, a relatively small number of customers may represent significant commercial value. What really matters here is whether a brand has the means to reach those audiences and the potential returns justify the resources required.

A segmentation study should help businesses allocate resources more intelligently. If implementing its recommendations requires an increasingly elaborate marketing operation without a convincing commercial benefit, greater precision may have become an expensive distraction.

Knowing when you've gone far enough

I'll admit that my enthusiasm for slicing and dicing data probably comes from an unhealthy interest in mysteries. Give me a dataset, and I'll happily spend hours looking for the clue nobody else has spotted, convinced there's some previously undiscovered truth hiding in the numbers.

And sometimes there is. Other times, I've uncovered something fascinating that has absolutely no commercial value whatsoever.

Imagine discovering that redheads blink twice as often as people with green eyes. Extraordinary. Stop the presses. But once the excitement has worn off, someone inevitably asks the question every researcher needs to hear: "Fascinating, Jodie. And what exactly are we supposed to do with that?"

And it is a completely valid and fair point. Because the value of research isn't measured by how many interesting observations we can extract from a dataset. It's measured by whether those observations improve our understanding of consumers and help businesses make better decisions.

That doesn't mean every finding needs an immediate commercial application. Exploratory research has enormous value, and sometimes the most unexpected discoveries lead to opportunities nobody knew to look for. But there's a difference between a pattern worth investigating and an insight worth building a strategy around.

The real discipline is knowing when to stop digging. Can we demonstrate that a difference is reliable? Does it reveal something meaningful about consumer needs or behaviour? Could acting on it improve how a brand competes, connects with customers, or grows? And importantly, is there a way to measure whether the strategy actually worked?

I'll probably never lose my enthusiasm for finding another interesting correlation. But I've learned that uncovering a mystery and solving a business problem aren't necessarily the same thing.

And if the most impressive thing we can say about our segmentation is how many segments we've identified, perhaps it's the research itself that needs a little more scrutiny.

Are you getting meaningful insights from your audience segmentation?

At Kadence International, we help businesses find the right balance between analytical depth and commercial relevance. Our research teams combine methodological rigour with experience across industries and global markets to identify the audience differences that genuinely matter, challenge assumptions, and turn complex data into insights that inform strategy.

Talk to Kadence about making your audience segmentation work harder for your business.