KwantumLabs Research Reference KwantumLabs Research Reference A practitioner's guide to segmentation: what it is, when each approach works, and how to apply it to B2B interview research.
Before choosing a segmentation method, ask: What decision will this segmentation inform? The method follows the question, not the other way around.
Wendell Smith introduced market segmentation in 1956, distinguishing it from product differentiation. Product differentiation bends demand to fit the product (one offering, many customers). Market segmentation bends the product to fit demand (different approaches matched to different buyer groups). Segmentation is a strategic commitment to treat different buyers differently because doing so is more profitable.
What should we build or improve?
Segment on needs and jobs-to-be-done.
How much will different buyers pay?
Segment on willingness to pay and price sensitivity.
How should we position and communicate?
Segment on pain points and evaluation criteria.
Who should we prioritize?
Segment on firmographics and buying behavior.
Three frameworks shape how researchers classify segmentation approaches. Understanding these helps you choose the right method for your situation.
A priori segmentation defines segments before analysis. The researcher selects the variables ("segment by company size" or "segment by seniority"). The number and type of segments are predetermined.
Post hoc segmentation discovers segments from data through clustering or classification. Patterns reveal natural groupings that the researcher did not predetermine.
Wind (1978) established this distinction and argued that the choice of segmentation basis (what you segment on) is more consequential than the choice of statistical method.
Wedel and Kamakura (2000) added a second dimension, creating the standard 2x2 framework:
Sausen et al. (2005) proposed a third distinction based on what the segmentation tells you:
Descriptive answers "who is different?" (demographics, firmographics). Predictive answers "what will customers do?" (purchase likelihood, churn risk). Prescriptive answers "how should we treat each group?" (optimal offer, channel, message). Prescriptive is the most actionable but hardest to implement.
Each segmentation method has specific strengths, limitations, and data requirements. Here are the major ones, organized by their underlying logic.
Russell Haley (1968) argued that the benefits customers seek are the most meaningful basis for segmentation. Demographics describe who people are; benefits describe why they buy.
His classic toothpaste example: the market segmented by benefit sought (decay prevention, brightness, flavor, price) rather than demographics. Each benefit segment had distinct brand preferences, but the benefit was the causal driver.
Jobs-to-be-done (JTBD) is the modern evolution. Christensen et al. (2005) argued that segmentation should be based on the "job" the customer is trying to accomplish, not customer attributes.
When to use: Product development, messaging, or value proposition design. Produces the most strategically useful segments but requires effort to collect and link to observable identifiers.
"The benefits which people are seeking in consuming a given product are the basic reasons for the existence of true market segments." (Haley, 1968)
The B2B equivalent of demographic segmentation. Segments by industry, company size, location, technology stack, or growth stage.
Advantages: Easy to implement. Data is publicly available or collected early in sales conversations. Every company can be assigned to a segment. Sales teams understand it intuitively.
Limitations: Describes who companies are, not what they need. Two 500-person tech companies may have completely different challenges. Competitors can copy firmographic segmentation because the data is public.
Best use: As a targeting overlay on top of needs-based segments. The question becomes: "Which firmographic traits predict membership in our most attractive needs-based segment?"
The most common statistical technique for post hoc segmentation.
K-means (MacQueen, 1967) partitions observations into k clusters where each observation belongs to the cluster with the nearest mean. Requires pre-specifying k. Sensitive to initial seed placement. Works best with continuous, roughly spherical data.
Hierarchical / Ward's method (Ward, 1963) builds a tree (dendrogram) of nested cluster solutions. Does not require pre-specifying k. Useful for exploring possible structures, but computationally expensive.
Two-step approach (Punj & Stewart, 1983): Use hierarchical clustering to determine the number of clusters, then apply k-means. This is the recommended best practice.
Punj & Stewart warned that clustering results are highly sensitive to the choice of distance metric, algorithm, and variable scaling. Always run multiple methods and compare. Never treat cluster output as "truth" without validation.
Magidson & Vermunt (2002) argued that LCA is superior to k-means for segmentation because it:
Nylund et al. (2007) found that BIC was the best criterion for selecting the correct number of latent classes. With small samples (under 500), LCA can overfit or underfit, so interpret cautiously.
When to use: Mixed variable types, formal model selection criteria needed, or when probabilistic membership is more realistic than hard assignment.
A common practice: collect survey data, run factor analysis to reduce dimensions, then cluster on factor scores.
Dolnicar & Grun (2009) challenged this approach through simulation and found that:
Takeaway: If you have a manageable number of variables (under 15-20), cluster directly on the raw data rather than reducing with factor analysis first.
RFM (Recency, Frequency, Monetary) segmentation (Hughes, 1994) scores customers on how recently they purchased, how often, and how much. Customers are scored in quintiles, creating segments from "5-5-5" (best) to "1-1-1" (least engaged).
Simple, actionable, and requires only transactional data. Recency is highly predictive of future behavior. Fader et al. (2005) extended RFM by connecting it to Customer Lifetime Value modeling.
Limitation: Purely behavioral. Does not capture needs, attitudes, or motivations. Most relevant for direct marketing and retention, less so for interview-based research.
| Method | Data type | Pre-specified? | Probabilistic? | Best for |
|---|---|---|---|---|
| A priori rules | Any | Yes | No | Simple, known segments |
| K-means | Continuous | k is | No | Large datasets, continuous variables |
| Hierarchical | Continuous | No | No | Exploratory, small-medium datasets |
| Latent class | Categorical / Mixed | No | Yes | Survey data, mixed variable types |
| Mixture models | Continuous | No | Yes | Segmentation + response modeling |
| RFM | Transactional | Yes | No | Direct marketing, retention |
B2B markets have unique characteristics that shape how segmentation works: fewer buyers, longer sales cycles, buying committees instead of individual consumers, and publicly available company data.
The most influential B2B segmentation framework, published in Harvard Business Review in 1984. It proposes five levels, moving from outer (easy to observe, cheap to collect) to inner (harder to assess, more differentiating).
The principle: start with the outer nest and only move inward when outer levels fail to differentiate. Each inner layer requires more data collection effort but provides more precise segmentation.
The central challenge: the most strategically useful segments (needs-based) are the hardest to identify in practice. Yankelovich & Meer (2006) argued that effective segmentation must pass three tests:
After identifying needs-based segments, find 1-3 simple, observable questions that predict segment membership. In B2B, these are often company size, hiring volume, technology maturity, or industry vertical. If a sales rep can determine segment membership in the first five minutes of a discovery call, the segmentation is operationally viable.
| Criterion | Firmographic | Needs-based |
|---|---|---|
| Data availability | High (public data) | Low (requires research) |
| Implementation ease | Easy | Hard |
| Strategic differentiation | Low (competitors can copy) | High (unique insight) |
| Actionability | High (directly targetable) | Medium (needs overlay) |
| Explanatory power | Low to medium | High |
Best practice: Combine both. Use needs-based segmentation as the foundation, then identify firmographic proxies that predict segment membership. The result is strategically meaningful and practically targetable.
In 2021, Roberto Cortez, Ann Clarke, and Per Vagn Freytag published the most comprehensive review of B2B segmentation research to date. Their systematic review of 88 papers spanning 34 years (1986-2019) reveals both how far the field has come and how much remains unresolved.
The authors screened 836 articles across 17 journals (including top-tier outlets like the Journal of Marketing, Journal of Marketing Research, and Marketing Science, plus B2B-specialized journals like Industrial Marketing Management). Two experienced researchers independently scored each article, achieving 0.85 inter-rater reliability. Three researchers then coded every paper in the final set of 88.
Their findings paint a sobering picture. After three decades of research, the field still lacks comprehensive guidelines for developing robust B2B market segments. Academic interest is declining (publications dropped from 30 papers in 1986-96 to 25 in 2008-19). Nearly half of all papers came from a single journal (Industrial Marketing Management). Only 17% appeared in top-tier marketing journals.
Most concerning: 67% of the papers reviewed had no theoretical foundation. The research is overwhelmingly quantitative (52%) and heavily fragmented. Researchers keep studying pieces of the segmentation puzzle without assembling the full picture.
The paper's central contribution is an integrative framework that treats segmentation as a continuous, four-phase process rather than a one-time analytical exercise.
Evaluation feeds back into the next cycle. Segmentation is never "done." Firms that have not resegmented in 10-20 years risk staying in unprofitable markets, losing differentiated positions, or targeting segments that lack growth potential.
This phase is almost entirely absent from the literature, yet the authors argue it is foundational. Two questions must be answered before any analysis begins:
This validates something we already do in client engagements: starting with "what decision will this segmentation inform?" before choosing methods. Pre-segmentation is where most internal segmentation efforts fail because teams jump straight to analysis without aligning on market definition or purpose.
Three activities within this phase:
Identifying variables. The review traces a clear evolution over decades:
A critical gap: while academic research has shifted toward micro-segmentation, there is no evidence that B2B firms have actually followed. Most companies still segment primarily on firmographics.
Choosing a model. Five model categories appear in the literature:
| Model type | Approach | Usage in B2B |
|---|---|---|
| Network analysis | Qualitative; maps value co-creation across network actors | Emerging |
| Matrix analysis | Predefined variables in 2x2 classifications | Common in practice |
| Cluster analysis | Hierarchical or K-means grouping | Most common in research |
| Latent class analysis | Regression-based; identifies unobservable clusters | Underused (requires econometric skill) |
| Optimization procedures | Includes profitability alongside needs/behavior | Growing but criticized |
The authors note that latent class analysis outperforms most cluster analysis approaches but remains underused in B2B because it requires advanced statistical expertise.
Selecting target markets. The authors stress the importance of a formal segment attractiveness metric to avoid the temptation of total market coverage. Common selection criteria include: ability to reach buyers, competitive positioning, market size, compatibility with firm objectives, profitability, and expected growth.
This is where segmentation most commonly fails. Three activities are required:
When we deliver segmentation in client reports, the segments are only as valuable as the client's ability to implement them. This research reinforces why our reports should include concrete implementation guidance: which teams need to change, what the marketing mix looks like per segment, and what "killer questions" let sales reps quickly identify segment membership in discovery calls.
Three outcome categories should be measured:
Evaluation feeds back into the next segmentation cycle, reinforcing the continuous nature of the process.
The framework identifies four contextual factors that influence the entire process:
The review's most important takeaways for anyone doing B2B segmentation in practice:
When working with interview samples of 15 to 100 participants, the approach shifts from statistical clustering to qualitative typology building. Here is what the literature says about validity at these sample sizes.
Malterud et al. (2016) introduced the concept of information power: the more relevant information each participant holds, the fewer participants you need. Five factors determine required sample size: study aim breadth, sample specificity, use of established theory, quality of dialogue, and analysis strategy.
| Sample size | Feasible | Positioning |
|---|---|---|
| 15-25 interviews | 2-3 qualitative typologies | Exploratory, hypothesis-generating |
| 25-50 interviews | 2-4 qualitative segments | Directional, sufficient for initial strategy |
| 50-100 interviews | 3-4 segments with reasonable confidence | Strong qualitative evidence |
| 100-300 surveys | Basic cluster analysis or LCA | Pilot quantitative segmentation |
| 300+ surveys | Full LCA, mixture models | Statistically robust |
With interview samples, you cannot claim statistical generalizability, precise segment sizes, or confidence intervals. You can claim: "Our interviews revealed distinct patterns of needs and behaviors suggesting X segments that warrant further quantitative validation."
Use thematic analysis (Braun & Clarke, 2006): familiarization, initial coding, generating themes, reviewing, defining, and naming themes.
Look for clusters of participants who share similar needs, behaviors, attitudes, and pain points across multiple dimensions.
Define 2-4 segments, each with a descriptive label, primary needs, tool preferences, spending patterns, and firmographic profile.
Map observable traits (company size, industry, role) that correlate with each segment. These become the "killer questions" for sales teams.
Check: do the segments respond differently to key questions (tool preference, willingness to pay, frustrations)? If two segments look the same on outcomes, they are not meaningfully different.
Not all segmentation solutions are useful. Here are the established frameworks for evaluating whether your segments are worth acting on.
The textbook standard from Marketing Management (Kotler & Keller, 2016):
Measurable: Segment size and characteristics can be measured. Substantial: Large and profitable enough to serve. Accessible: Can be reached through marketing channels. Differentiable: Segments respond differently to different marketing strategies. Actionable: Programs can be designed to serve each segment.
Wedel & Kamakura (2000) added three more:
The most rigorous modern approach (Dolnicar et al., 2018):
Do segments make structural sense? Separation between clusters, compactness within them.
Do segments replicate across bootstrap samples? If not, they may be statistical artifacts.
Do segments differ on variables NOT used to create them? If needs-based segments also show demographic differences, confidence increases.
Do segments predict meaningful outcomes (purchase behavior, brand choice, willingness to pay)?
Can the organization identify members, reach them, and design programs for them?
Dibb (1998) identified four common failure modes:
IT systems cannot support segment-level decisions or reporting.
Departments cannot agree on segment definitions.
Segments are created but never acted upon. The most common failure.
Segmentation is treated as a one-time project rather than an ongoing capability.
Use these guides to match your situation to the right segmentation approach.
| Question | Segment on | Method |
|---|---|---|
| "Who should we target?" | Firmographics + needs | Nested approach (Bonoma & Shapiro) |
| "What should we build?" | Needs / jobs-to-be-done | Benefit segmentation (Haley) |
| "How should we message?" | Pain points + evaluation criteria | Needs-based with persona overlay |
| "How should we price?" | Willingness to pay + value perception | WTP-based clustering |
| "Which customers are most valuable?" | Behavioral + spend | RFM or CLV-based segmentation |
| Data available | Recommended approach |
|---|---|
| Interview transcripts only | Thematic analysis into qualitative typologies |
| Survey data (Likert, ratings) | Latent class analysis or k-means clustering |
| Transactional / behavioral data | RFM analysis or behavioral clustering |
| Mixed (interviews + structured) | Qualitative typologies validated against structured variables |
| Large dataset (1,000+) | LCA, mixture models, or ML approaches |
For interview-based B2B research (the typical KwantumLabs engagement): Start by coding themes, look for natural groupings in needs and behaviors, define 2-4 segments, then map them to observable firmographic traits that sales and marketing teams can use. Position segments as directional buyer profiles backed by qualitative evidence, not statistically proven market structures.
The most useful segmentation is the one that gets implemented. A simple, identifiable 3-segment model that sales teams actually use beats a sophisticated 7-segment model that sits in a slide deck.
All sources cited on this page, organized by contribution area.