Most organizations know their market isn't monolithic. Different groups of customers want different things, respond to different messages, and make decisions in fundamentally different ways. The problem isn't a lack of awareness — it's a lack of precision. Demographic segmentation tells you who people are but not how they behave. Intuition-based segments feel right but fracture under scrutiny. And when your segmentation doesn't hold up, every downstream decision built on it — targeting, messaging, product development, pricing — is working from a flawed foundation.
Cluster analysis is the statistical approach we use at Meraki Research to build segmentation schemes that reflect how your market actually works. It's a family of clustering algorithms that examine data points across multiple variables and group respondents based on measured similarity — not assumptions. The result is clusters of people who genuinely resemble each other in attitudes, behaviours, needs, or preferences, and who are meaningfully different from people in other groups.
We've built segmentation solutions for organizations ranging from Fortune 500 consumer brands to B2B technology companies to healthcare systems. Our approach goes beyond running a clustering algorithm and reporting the output. We design segmentations that your teams can understand, identify, and activate in the real world.
Our Approach: The Meraki Segmentation Architecture Framework
Segmentation only works if the groups it produces are distinct, stable, identifiable, and actionable. Our Segmentation Architecture Framework is built around those four requirements, moving through five phases that connect statistical rigor to market reality.
Phase 1: Segmentation Basis Selection
The most critical decision in any cluster analysis is choosing what to segment on. We work with your team to define the segmentation basis — the set of variables that will determine how groups form. Common bases include:
- Needs and motivations: What outcomes or benefits do people seek?
- Attitudes and values: How do people think about the category, brand, or problem?
- Behaviors: What do people actually do — purchase patterns, usage frequency, channel preferences?
- Occasion or context: When and why do people engage with your category?
We deliberately separate the segmentation basis from profiling variables (demographics, firmographics, media consumption). Profiling comes after — it describes the clusters once formed, but shouldn't drive their formation. This distinction is essential to producing segments that capture genuine differences in how people relate to your market.
Phase 2: Data Preparation and Dimensionality Reduction
Raw survey data typically contains redundancy. If 25 attitudinal items really reflect five underlying dimensions, running a clustering algorithm on all 25 introduces noise and can obscure the real structure. We use factor analysis or principal components to reduce the data set to its essential dimensions before clustering.
This step also addresses the problem of variable scaling. Clustering algorithms are sensitive to the range and variance of input variables. We standardize all inputs to ensure that no single variable dominates the algorithm simply because it happens to have a wider numeric range.
Phase 3: Cluster Extraction
We apply multiple clustering algorithms to your data and compare results, rather than relying on any single method. Our standard toolkit includes:
- Hierarchical clustering to explore the natural groupings in the data and identify a plausible range for the number of clusters. We use agglomeration methods (Ward's, average linkage, complete linkage) and examine the distances at which groups merge.
- K-means clustering to assign every data point to its nearest cluster centre, producing clean, non-overlapping segments. K-means is the workhorse of market segmentation for good reason: it scales well and produces interpretable groups.
- Latent class analysis when your input variables are categorical or mixed, or when you need a probability-based clustering approach that assigns each individual a likelihood of belonging to each cluster.
We run each algorithm for multiple cluster counts (typically 3 through 8) and evaluate solutions on statistical fit, segment separation, and — critically — interpretability. A six-cluster solution that produces two nearly identical groups is worse than a five-cluster solution with clear differentiation.
Phase 4: Validation and Stability Testing
A segmentation that falls apart under scrutiny has no value. We validate every cluster solution through:
- Split-half replication: We divide the data set in half and run the clustering independently on each half. If the same groups emerge, the structure is real.
- Discriminant analysis: We test whether the cluster assignments can be accurately predicted from the input variables, confirming that the groups are genuinely distinct.
- Profile differentiation: We examine whether the clusters differ meaningfully on external variables (demographics, behaviours, sales metrics) that weren't used to form them. Strong external differentiation is a hallmark of a useful segmentation.
If a solution doesn't validate, we iterate. We adjust the variables, try different algorithms, or reconsider the number of groups. We never deliver a segmentation we can't defend.
Phase 5: Profiling, Naming, and Activation
Once the cluster structure is confirmed, we build rich profiles of each segment. This is where the data becomes a story. We analyse how each group differs on demographics, psychographics, media behaviour, brand perceptions, purchase behaviour, and any other profiling variables in the data set.
Each segment receives a name and a one-paragraph description that captures its essence — written in language your marketing, sales, and product teams will immediately understand. We create visual segment profiles, persona cards, and a typing tool (a short set of questions that can classify new individuals into their segment) for ongoing activation.
This typing tool is often the most valuable deliverable. It lets you apply the segmentation to CRM data, future surveys, and customer interactions without rerunning the full analysis.
When Cluster Analysis Is the Right Approach
Cluster analysis is the right tool when you need to:
- Identify distinct customer groups in your market based on measured similarity rather than assumed categories.
- Move beyond demographics to understand segments defined by needs, attitudes, or behaviours.
- Create actionable targeting frameworks that your sales, marketing, and product teams can use to allocate resources.
- Discover groups you didn't know existed. Because clustering is a form of unsupervised learning, it finds structure in the data without requiring you to specify the groups in advance. You may discover segments that challenge your current assumptions.
When Cluster Analysis Is NOT the Right Fit
- You need to identify which factors drive a specific outcome. That's regression analysis.
- You want to reduce a large number of variables to underlying dimensions. That's factor analysis.
- You already have well-defined segments and need to understand their price sensitivity. Conjoint analysis or Van Westendorp may be more appropriate.
- Your data has pre-assigned labels (e.g., loyal vs. lapsed) and you want to predict group membership. That's a classification problem, not an unsupervised learning task.
What We Deliver
Every cluster analysis engagement includes:
- Segmentation report detailing the number of clusters, the basis variables, the algorithm used, and validation results. Delivered as an executive presentation and a technical appendix.
- Segment profiles — rich descriptions of each group covering demographics, attitudes, behaviours, and key differentiators. Formatted as visual persona cards and detailed profile tables.
- Typing tool — a short questionnaire (typically 6-10 items) with a scoring algorithm that classifies new individuals into their segment with high accuracy.
- Segment assignments appended to your data file, so every respondent carries their cluster membership for downstream analysis and activation.
- Strategic recommendations on how to approach each segment — messaging themes, channel priorities, product positioning, and resource allocation guidance.
- Sizing and value estimation — how large each segment is in the population and what it represents in terms of sales potential or lifetime value.
Typical turnaround is 3-5 weeks from clean data, depending on the complexity of the segmentation basis and the number of markets or categories involved.
Industry Applications
Consumer Packaged Goods: Building needs-based segmentations that reveal which groups of shoppers drive category growth, enabling targeted innovation, messaging, and shelf strategy. Our CPG segmentations routinely identify two to three clusters that account for the majority of category sales.
Financial Services: Segmenting customers by financial attitudes, risk tolerance, and service expectations — not just assets under management. Organizations in banking and insurance use our segmentations to personalize offers and improve retention across distinct groups.
Healthcare: Creating patient segments based on health attitudes, treatment preferences, and engagement levels. Health systems use these clusters to design targeted outreach that improves adherence and outcomes.
Technology & B2B: Identifying buyer segments in complex sales cycles where decision-making units involve multiple individuals with different priorities. Our B2B segmentations account for both organizational and individual factors.
Retail: Segmenting shoppers by occasion, channel preference, and value orientation — revealing which groups are most responsive to promotion, loyalty programs, or premium positioning.
Media & Entertainment: Building audience clusters based on content preferences, consumption patterns, and platform behaviour to optimise programming and ad targeting.