The same audience on paper. Why does one message resonate and another repel?

Wendell Smith's foundational work described segmentation as a response to heterogeneous demand rather than an arbitrary exercise in dividing ages and income bands [S01]. Demographics can be useful for estimating reach, planning distribution and making sure a service is accessible. They do not automatically reveal which anxieties, expectations or memories become relevant to a particular proposition. A demographic variable can be an effective media-planning input without being the cause of purchasing behaviour.

Dolnicar, Grün and Leisch emphasise that segments do not always exist as neat natural groups waiting to be discovered [S02]. The variables, sample, method and business objective all influence the solution. If your questionnaire collects only age and income, the resulting groups cannot be presented as though their mental associations have been measured. Understanding those associations requires asking what people connect to the category and brand, examining context and testing whether different patterns meaningfully affect decisions. More complex targeting is not automatically more insightful research.

The right segmentation method depends on the question you need to answer. These lenses often complement one another, but they are not interchangeable.

A genuine investigation can combine these layers. An association pattern might help explain why people read the same product page differently. But if some cannot afford the product or cannot get it delivered, a mental explanation alone will be incomplete. The goal is not to replace demographics with psychology. It is to avoid confusing reach, context, meaning and observed action [S02][S05].

Kevin Lane Keller's customer-based brand equity model describes brand knowledge in terms of awareness and associations in memory [S03]. A new campaign does not arrive on an empty mental page. People bring product experiences, comparisons, conversations and their own categories of meaning. A word such as 'premium' might suggest dependable construction to one person and an unnecessarily high price to another. The text on the advertisement is the same. What gets activated by it is not necessarily the same.

Research on the automatic activation of attitudes demonstrates that sufficiently strong evaluative associations can become accessible relatively automatically under specific experimental conditions [S06]. That finding does not prove that every subconscious response causes a purchase. It does justify looking beyond a single survey question such as 'do you like this brand?'. It matters what a name, claim or visual brings to mind and how that relationship changes with the situation. Neurofactor's knowledge base explains the basic distinction in association.

When people disagree, there are at least three candidate explanations. Real heterogeneity means that their associations or priorities differ. Context dependence means the same person responds differently when buying a gift, replacing a broken product or making a business purchase. Measurement noise means reported preferences shift with question wording, sequence, fatigue or random variation. These situations call for different research and different marketing decisions.

Bettman, Luce and Payne describe consumer choices as partly constructed during the act of choosing rather than merely retrieved from permanently fixed preferences [S05]. Schwarz shows that seemingly small features of a survey can influence self-reports [S11]. It is therefore a serious mistake to discover four colourful archetypes in one survey and treat them as permanent truths about the market. Ask whether the findings repeat, whether they still appear when you change the measurement method and whether they explain responses to concrete claims or evidence outside the original questionnaire.

Start with the moment you want to improve. Is the problem an ignored advertisement, a confused first visit to a product page, hesitation after seeing the price, a comparison between brands, or a failure to return after trial? A segment model that helps at awareness may be less useful when someone needs delivery certainty. 'Our audience is people like this' is not a sufficient research question.

Specify the population able to make the choice, the visible alternatives and practical restrictions. Write down a small number of competing explanations. Perhaps one subgroup associates an adjustable desk with independence while another associates it with the risk of wasting money. Then define what would challenge the hypotheses. If both groups respond equally to different claims and choose the same evidence, separate messaging may add cost without value. Segmentation is useful only when it changes a real decision for the better; creating additional persona slides is not an outcome.

Sheth, Newman and Gross proposed five consumption values: functional, social, emotional, epistemic and conditional [S04]. These are useful analytical perspectives, not boxes in which people permanently live. An adjustable desk may provide function through its height range, identity through the appearance of a professional workspace, reassurance through reliable construction and emotional value through a sense of control. The buyer may consider several of these at once.

Which value becomes most relevant can change with circumstances. A freelancer replacing a broken desk may prioritise availability and service. The same freelancer setting up a new studio later may care more about aesthetics and flexibility. Even when a particular combination of values appears frequently in a dataset, you should describe it as a potential decision pattern rather than a type of person. The research challenge is to show when those values matter and whether they actually explain different reactions to a particular offer [S04][S05].

The following profiles are illustrative hypotheses only. They are not Neurofactor client results and are not validated psychological classifications.

Nothing here follows automatically from being thirty-five, self-employed or living in a city. Even a recurrent pattern has overlap and uncertainty. You can use these hypotheses to design four experimental message routes; you cannot use them as proof that the population consists of exactly four types of minds.

Entirely fictional case - not a client project, dataset or measured result. A manufacturer of modular home-office desks targets urban professionals aged thirty to forty-five. Its campaign headline is 'The premium desk for your best working day'. The design looks consistent, the product specifications appear sensible and the demographic targeting reaches the planned audience. Yet four imagined visitors might interpret the proposition quite differently.

One thinks premium means stability and asks about load capacity. Another expects an inflated price. A third wonders whether the desk will fit into a compact apartment. A fourth recalls buying a short-lived product and wants to know about spare parts. These are deliberately fictional reactions that turn an abstract problem into a testable one. They should never be described as interview quotations. The next research step is to discover whether such patterns actually occur, how frequently, and whether they affect meaningful choices beyond the interview.

These are not survey results. Each row describes one possible buyer concern among people who look identical on the selected demographic variables.

Notice that the last column contains research questions rather than persona names. Labels such as 'the perfectionist' can make a presentation memorable while pretending to explain behaviour. One person can fit several rows depending on context. Treat segment membership as uncertain and allow for overlap instead of assigning permanent mental labels to customers.

Start with open interviews and accounts of genuine choices. Ask what people expect, distrust, compare and need to see before committing. Do not lead with 'how important is sustainability to you?'. Let people describe the product and category in their own language before introducing prepared attributes. Vary the order of stimuli and keep the brand, category, price and use situation analytically distinct.

Use qualitative findings to form a transparent measurement model rather than to announce your final segments. Build a broader sample for structured questions, test whether terms mean the same thing to different respondents and use independent coding of open answers. Record whether an expression came spontaneously from a participant or was introduced by the interviewer. A useful finding could be that 'durable' connects with repairability rather than prestige for a group, but that conclusion requires actual evidence. Keller's brand-association framework provides theoretical grounding, not a licence to invent unobserved clusters [S03].

An open association task discovers language. An attribute grid compares associations with different brands. Under suitable design and validation, an indirect task can explore relative link strengths between carefully selected concepts. Reaction times depend on task design, comprehension, practice and analysis; they are not direct readings of hidden motives. The original IAT research and later predictive-validity meta-analysis support limited, context-dependent uses of indirect measures, not individual forecasts of who will buy an adjustable desk [S07][S08].

Add observed choice when the question requires it: which proof people request, which product configuration they explore and whether a test message changes a meaningful decision. EEG and eye tracking can, in appropriate research designs, provide additional observations of processing or visual attention. Neither method can independently tell you that a participant secretly associates a product with freedom. Multiple methods are useful precisely because each leaves uncertainty that another may help address.

Choose a method based on the decision at stake. More expensive technology is not automatically better evidence.

The aim is a defensible chain from question to measurement to decision. If an interview suggests that repairability matters, the next step might simply be to compare a repairability proof asset against a generic design claim. You do not need to label a consumer as a 'sustainability persona' in order to test that message.

A neutral average response to 'premium' could mean almost everyone feels neutral. It could also mean one group is attracted and another feels excluded. Looking only at the mean can obscure that distinction. But a wide distribution does not automatically justify drawing two stable segments. Sampling error, response wording and different choice situations may be enough to generate an apparent contrast.

A credible study therefore examines distributions, prespecified subgroup contrasts and the reliability of any grouping algorithm. Research on heterogeneous treatment effects demonstrates why estimating differences in response to an intervention is more difficult than simply splitting respondents into categories after seeing the outcome [S12]. Use held-out data, sensible primary outcomes and interaction tests where appropriate. The important question is not merely whether respondents have different attitudes, but whether different proof or communication causes meaningfully different responses under comparable conditions.

Return to the fictional desk company. Version A of its product page opens with 'Premium design for professionals'. Version B says 'Designed to adapt, be repaired and support years of use', supported by verifiable specifications. Which version wins? The honest answer is that we do not know. Their purpose is to represent two competing hypotheses that could be tested.

For a buyer who connects premium with sound engineering, version A might be reassuring if proper evidence follows. For someone who hears 'unnecessary luxury', the same phrase may create friction. Version B could appeal to a person prioritising long-term responsibility, yet do little for a visitor concerned mainly with delivery this week. There is no universal optimum headline independent of the audience and moment. Research on constructive choice and framing supports taking that context seriously, without promising a guaranteed uplift from any particular wording [S05][S10].

An association finding should lead to a practical change. If people connect 'premium' with 'overpriced', replacing the word with 'exclusive' may simply preserve the problem. Perhaps price is felt as an unexplained risk. Then product durability, repair options, warranty terms and an honest comparison may matter more than copywriting style. The appropriate proof must actually exist and be easy to inspect.

Create a four-column working sheet for each hypothesis: possible association, suspected barrier, verifiable evidence, next useful action. Ensure the core brand promise remains consistent across every version. A marketer can adapt the route to understanding without inventing four incompatible company identities. If no meaningful response difference appears in testing, simplify rather than multiplying campaigns. Association-based segmentation earns its place through evidence and decisions, not through elaborate labels [S03][S05].

This matrix is a fictional design aid, not a validated result. It shows how research findings could eventually inform message testing.

If you cannot verify a promise, it should not enter the message matrix. And if the response to these variants is similar, that is a finding too. A good test gives the marketing team permission to abandon unnecessary complexity.

Imagine someone replacing a broken desk today. Availability, delivery and clear return terms become pressing. Six months later the same person may be designing a larger office and exploring modular flexibility or aesthetics. Their age and income may not have changed. Their decision frame has. The distinction matters because marketing platforms can make a fixed demographic profile look like the whole explanation.

Tversky and Kahneman demonstrated that how decision problems are framed can alter preferences in experimental settings [S10]. Bettman and colleagues describe how the decision process itself influences which preferences are used [S05]. Neither paper proves that a particular desk advertisement will win. Together they justify measuring the occasion and decision context rather than treating a reported association as an unchanging character trait. Context-aware content can also be more privacy-preserving than building a detailed persistent record about each individual.

Use eight practical steps. 1. Define the decision, audience and feasible alternatives. 2. Collect spontaneous associations before supplying answer words. 3. Specify the proposed clusters and competing explanations. 4. Check sample quality, overlap, measurement reliability and repeatability. 5. Form one credible claim and proof asset per hypothesis. 6. Randomise suitable participants to message variants within defined decision contexts. 7. Measure a prespecified behavioural outcome such as a qualified request, informed product comparison or purchase. 8. Replicate the most important interactions before expanding the campaign.

Document the minimum meaningful effect, sample-size requirements and planned comparisons. Repeatedly trying different segment definitions until one appears to work is a route to false positives. Athey and Imbens provide statistical methods for studying heterogeneity in causal effects; that methodological contribution is not a guarantee of campaign success without valid data and randomisation [S12]. The objective is to separate an attractive story about buyer types from an observed, reproducible difference in communication impact.

A message can attract more curiosity clicks while losing people when the product details appear. Another message may draw fewer visitors but better match a genuine decision need. Both are plausible possibilities, not claims about a real campaign. Decide in advance which outcomes matter. Ajzen's theory of planned behaviour also distinguishes intentions and perceived control from behaviour actually performed [S09]. For an inexpensive consumer product it may be an informed checkout; for a complex B2B proposition it could be a qualified technical conversation. Track unintended consequences as well as immediate engagement.

Build a measurement ladder: exposure and attention, understanding, association response, meaningful engagement and where suitable purchase or use. Record cost and operational complexity. A segment that looks statistically distinct yet is impossible to reach without intrusive tracking or creates costly, contradictory copy may have little practical value. Dolnicar and colleagues emphasise that market segmentation is both a technical and managerial choice [S02]. A successful model must make resource allocation better, not merely make dashboards more colourful.

A striking difference in associations is a starting hypothesis. This matrix helps translate evidence into a restrained decision.

This is a decision aid, not a scientifically validated scoring system. The acceptable thresholds for effect, cost and uncertainty should be set for the real decision. One robust and useful contrast is preferable to seven decorative clusters.

A term that feels natural in Dutch may carry different connotations in English or German. 'Premium', 'sustainable' and 'flexible' are not universal mental codes. Experience with the product category, national context and local usage can change which idea is salient. Even people who speak the same language may connect a word to different choices. A translated cluster label does not validate a translated segment.

Test whether the terms are understood and measured comparably in each market. Conduct local qualitative work, check translation equivalence and compare measurement structure before drawing conclusions about country-level differences. If the questions work differently, the apparent gap could come from the instrument rather than the audience. Revisit results as competition, products and economic conditions change. The aim is living, testable market knowledge rather than a permanent portrait built from a single survey [S02][S11].

Start with one decision that needs improvement, such as the first product promise on a landing page. Write down the current demographic assumptions and the meanings you think lie behind response differences. Collect open and structured association data. Check distributions, uncertainty and the stability of any clusters. Translate promising patterns into evidence that your organisation can actually supply. Test variant messages in a real decision context using one primary outcome. Finally, decide whether the gain is robust, affordable and achievable without disproportionate personal profiling.

This question connects to several existing Neurofactor articles. For the difference between awareness and choice, read why people know your brand but choose another. For competitive similarity, explore why competing brands feel the same. For opposite reactions to the same need, see why the same need attracts and repels. The overall foundation is associations in marketing. This article adds the missing step: finding, validating and acting on heterogeneity within one nominal audience.

Demographic characteristics can help you find an audience. Associations may help you understand what a choice means to different people within it. Between those two lies the essential research work: identifying patterns that recur, testing whether context explains them and showing whether different evidence or messages make decisions meaningfully easier. Without that work, segmentation becomes an attractive collection of speculative personas.

The lesson is not that every individual needs a unique message. It is that people should not be assumed to share a decision frame simply because they share a profile on paper. Measure the differences, keep the underlying brand promise consistent and vary communication only where the evidence supports it. That is how segmentation becomes a tool for better choices rather than another layer of targeting jargon.