Market Segmentation: Build Groups That Change the Strategy

“Mid-market healthcare companies” sounds like a segment. It may be only a row the CRM can filter.

The companies inside it can have different buyers, adoption risks, workflows, alternatives, and definitions of value.

If those differences demand different product or go-to-market choices, the label hides the strategy instead of clarifying it.

A useful market segment is not merely a group with shared attributes. It is a group whose members are similar in a way that should change what the organisation offers, how it reaches them, or whether it should serve them at all.

Segmentation earns its cost through differentiated action.

State the decision before finding the groups

There is no universally correct segmentation of a market. The same population can be grouped differently for product design, market entry, pricing, sales coverage, onboarding, or service operations.

Begin with one decision:

• Which customers should the product target first?
• Does one workflow require a materially different offer?
• Which group can the current channel reach responsibly?
• Should an enterprise requirement enter the core product or a separate service model?
• Is poor retention concentrated around one value mechanism or adoption constraint?

Then write what would differ between segments. Possible actions include the product promise, capability set, evidence required, buying route, onboarding, pricing model, service level, or deliberate exclusion.

If every candidate group receives the same action, the model may be descriptive but it is not helping this decision.

Smith’s 1956 article framed market segmentation as responding to heterogeneous demand with different offerings, in contrast with product differentiation aimed more broadly across demand.

It is a foundational conceptual argument, not a modern empirical test or a complete method.

Its durable boundary is strategic: segmentation is connected to variation in demand and a differentiated response, not classification for its own sake.

Fix the market boundary and unit

Segmentation cannot repair an incoherent market definition.

Name the population being partitioned and the unit assigned to a segment. Is it a person, buying committee, account, organisation, location, transaction, or use situation?

A single organisation may contain several workflows and buying centres. One person may have different needs in different situations. Forcing either into one permanent segment can delete the variation that matters.

Write inclusion and exclusion rules:

Population and geography
Unit of segmentation
Problem or category boundary
Relevant role in the decision
Time horizon
Known exclusions

The market-analysis guide covers how customers, alternatives, channels, dependencies, and rules define the wider system.

Segmentation begins after that boundary is credible. It asks where variation inside the system warrants a different strategy.

Write the heterogeneity hypothesis

Do not collect dimensions because a survey or database makes them available. State the difference you expect and why it changes behaviour.

For example:

Organisations with a dedicated compliance owner may adopt the workflow differently because evidence review is a recurring responsibility with formal accountability, not an occasional task added to another role.

This is a hypothesis, not a segment conclusion. It identifies a possible mechanism and suggests what to investigate: role structure, workflow frequency, evidence requirements, authority, alternatives, and adoption behaviour.

Useful segmentation bases can include:

• problem urgency and frequency;
• desired outcome and acceptable trade-off;
• existing workflow and alternative;
• capability, readiness, or switching burden;
• buyer, user, and risk-owner arrangement;
• regulation, channel, integration, or service constraint;
• observed behaviour when its cause is understood.

Firmographics and demographics can be valuable for identification, reach, regulation, or economics. They become weak when used as unexplained proxies for need.

Company size may correlate with procurement complexity in one market and say little in another. Test the mechanism rather than turning the correlation into a persona.

Discover patterns without inventing certainty

Qualitative research can reveal differences in context, language, workarounds, authority, and meaning. It cannot estimate segment size from a small purposive sample.

Product and commercial data can reveal behavioural or economic variation. They cannot explain the mechanism by themselves, and current-customer data omits people who never adopted or never entered the funnel.

Use the methods together:

1. inspect existing customer, loss, support, usage, and market evidence;
2. conduct research across deliberately varied situations;
3. propose candidate differences and mechanisms;
4. define an assignment rule that could be tested;
5. examine prevalence, response, reach, and economics with suitable data;
6. return to contradictory and ambiguous cases.

A clustering algorithm does not, by itself, prove that its groups represent the true market structure. It partitions the variables, sample, distance choices, and parameters supplied to it.

Dolnicar and Leisch’s 2010 paper shows that cluster solutions depend on both data structure and algorithm parameters. It proposes bootstrap benchmarking for sample and algorithm randomness.

That is methodological work about data-driven clustering, not proof that every segmentation requires its exact procedure.

The relevant warning is broader: a visually neat cluster can be unstable. Test whether the groups survive reasonable changes in sample, variables, and method before building strategy around them.

Make assignment possible—and uncertainty visible

A segment that nobody can identify outside the research dataset cannot guide operations.

Define the minimum evidence needed to assign the unit:

• observable eligibility attributes;
• questions or behaviours that distinguish the mechanism;
• data source and owner;
• confidence or unresolved state;
• rule for overlapping or changing membership.

Avoid a long scoring formula that creates precision without meaning. Another person should be able to understand why an account belongs and what action follows.

Do not force exclusivity when the market is genuinely mixed. One organisation may have several use situations; one account may move as capability develops.

Use a primary situation for one decision, multiple membership, or an explicit “unresolved” state when that is more honest than a hard label.

Market segments are not user archetypes. An archetype can expose a behavioural or design pattern within a product decision.

A market segment must also support a market-facing choice, assignment, reach, and economic analysis for a defined population.

Test the segment before selecting it

For each candidate, build an evidence card.

Distinct response

Would this group require a different product, message, channel, price, adoption path, or service model? Which mechanism explains the difference?

Identifiability

Can the organisation recognise members with evidence available before the action is taken? A post-hoc label that appears only after churn may diagnose history but cannot guide acquisition.

Reachability

Can the team reach the relevant buyer and users through a credible channel? A large group outside the organisation’s access may not be an actionable target.

Substantiality and economics

Is the group large or valuable enough relative to acquisition, implementation, support, risk, and product costs? Use ranges and assumptions rather than a heroic single forecast.

Capability fit

Can the organisation deliver the promise with its current or deliberately acquired capabilities? Strategic fit is not the same as feature similarity.

Stability and change

Will the distinction persist long enough to act, or is it a temporary state? Which trigger would move an account between groups or invalidate the model?

Ethical and legal consequence

Could the assignment create unfair exclusion, sensitive inference, discriminatory treatment, or a claim the data cannot support? Commercial usefulness does not override rights or obligations.

The 2018 open-access book by Dolnicar, Grün, and Leisch treats market segmentation as an end-to-end process that includes deciding whether to segment, collecting and analysing data, selecting targets, customising action, and monitoring.

It is a methodological book, not a guarantee that its process produces superior commercial outcomes.

Its scope reinforces a useful standard: extracting groups is one step, not the finished strategy.

Select a target by comparing whole bets

Do not choose a primary segment from market size alone.

Compare the whole bet:

Important unmet or underserved need
Evidence that the segment is distinct
Reach and buying path
Product and service change required
Capability advantage or deficit
Economics and downside
Learning value
Conditions to enter, expand, pause, or stop

A smaller segment may be a stronger first choice because the team can reach it, learn quickly, and deliver a credible advantage. A larger one may justify investment if the capability gap is strategic and survivable.

Make the exclusion visible. State which segments will receive maintenance, opportunistic service, research, or no deliberate investment.

The product-growth strategy can help distinguish deeper penetration from market development, product development, and diversification once the target is chosen.

Turn the model into operating choices

A segmentation deck should change real artefacts.

For the selected segment, align:

• product promise and positioning;
• problems and use situations in discovery;
• roadmap trade-offs and explicit non-goals;
• acquisition and qualification rules;
• buyer, user, and implementation journey;
• pricing and service assumptions;
• success measures and counter-signals;
• sales, support, and product feedback loops.

Do not create a separate roadmap for every segment by default. Some differences need a distinct offer; others need communication, configuration, service, or no product variation at all.

Measure within segments only after the assignment and metric contracts are sound. Aggregate movement can hide opposing outcomes, but small or changing groups can also produce noisy comparisons.

Use the cohort guide when the decision requires mature comparisons across groups entering at different times.

A fictional segmentation decision

Consider a fictional document-approval product evaluating expansion beyond small operations teams.

The initial dataset groups accounts by employee count. Research suggests a more consequential difference: who owns evidence review and how failures are audited.

The team proposes three hypotheses: occasional review added to another role, recurring review owned by a specialist, and distributed review with formal local approvers.

It defines assignment evidence from role structure, workflow frequency, approval authority, and audit obligations. Company size remains descriptive but does not determine membership.

Before selecting a target, the team tests whether the groups differ in required controls, buying process, implementation burden, reach, and service economics.

No segment size, conversion rate, or winning choice is invented here. The analysis may show one target, several offers, or that the proposed partition is too unstable to use.

The segmentation is valuable because each possible result can change the strategy.

Treat the model as a versioned hypothesis

Markets, products, and customer capabilities change. The model should have an owner, evidence date, version, and triggers for review.

Review when:

• assignment produces frequent ambiguous cases;
• members of one group no longer respond similarly;
• the product or buying system changes;
• a new channel makes a group reachable;
• economics or obligations change materially;
• teams cannot name a different action by segment;
• an excluded population reveals a material harm or opportunity.

Track exceptions and movement. They may show a faulty rule, a transitional state, or a strategic shift.

Retire a segment when it no longer changes action. Split one when a label hides materially different mechanisms. Merge groups when their distinction has become descriptive rather than consequential.

A strong segmentation does not make customers tidy. It makes the organisation’s choices more explicit while keeping the uncertainty in the model visible.

Sources

• Smith, W. R. (1956). Product Differentiation and Market Segmentation as Alternative Marketing Strategies. Journal of Marketing, 21(1), 3–8. Foundational conceptual article, not a contemporary empirical validation or implementation method.
• Wind, Y. (1978). Issues and Advances in Segmentation Research. Journal of Marketing Research, 15(3), 317–337. Review of problem definition, research design, data, analysis, interpretation, and implementation in the segmentation literature of its period.
• Dolnicar, S., & Leisch, F. (2010). Evaluation of Structure and Reproducibility of Cluster Solutions Using the Bootstrap. Marketing Letters, 21, 83–101. Methodological paper on sample and algorithm sensitivity in exploratory cluster solutions.
• Dolnicar, S., Grün, B., & Leisch, F. (2018). Market Segmentation Analysis: Understanding It, Doing It, and Making It Useful. Springer. Open-access methodological book spanning decision, data, extraction, targeting, implementation, and monitoring; not evidence of guaranteed commercial impact.

Read next

• Market Analysis: Read the System Before You Choose
• User Archetypes: Build Models That Change Product Decisions
• Product Growth Strategies: Choose the Path Before the Tactic
• Cohort Tracking: Build Comparisons You Can Defend
• Go-to-Market Strategy: Design the Path to Adoption