Segmentation, Targeting and Positioning Analytics

Introduction and Context

Segmentation, targeting and positioning (STP) analytics converts heterogeneous market data into decisions about whom an organization should serve, what value proposition it should offer, and how that proposition should be communicated. Although STP is traditionally presented as a strategic marketing framework, analytics makes it more precise by using evidence about customer needs, behaviors, value, and competitive response.

A market is rarely homogeneous. Customers differ in demographic characteristics, motivations, usage patterns, price sensitivity, channel preferences, profitability, and future potential. Treating all customers as if they were identical can produce inefficient media spending, irrelevant offers, excessive discounting, and weak customer experiences. Conversely, highly detailed segmentation can create operational complexity and segments too small to serve economically.

The analytical task is therefore not simply to discover groups. It is to create segments that are meaningful, measurable, reachable, sufficiently substantial, stable enough for decision-making, and compatible with organizational capabilities. Targeting then involves selecting segments and allocating resources among them. Positioning translates the selected strategy into a distinctive, credible, and measurable place in the customer’s mental and behavioral marketplace.

Learning Objectives

By the end of this chapter, you should be able to:

1. Explain demographic, geographic, psychographic, behavioral, needs-based, and value-based segmentation.
2. Prepare and evaluate data for segmentation analysis.
3. Apply clustering and classification concepts to marketing problems.
4. Evaluate segment attractiveness, accessibility, profitability, and stability.
5. Select target segments using strategic and quantitative criteria.
6. Quantify positioning decisions using perceptual maps, preference data, and competitive analysis.
7. Recognize ethical, privacy, algorithmic, and managerial risks in STP analytics.

1. The STP Logic

STP consists of three connected decisions:

• Segmentation: dividing a broad market into groups with relatively similar characteristics or responses.
• Targeting: evaluating segments and deciding which ones to serve.
• Positioning: designing the offering and communication so that the brand occupies a clear, valued, and differentiated place relative to alternatives.

The sequence is analytical but not strictly one-directional. A firm’s positioning capabilities may influence which segments it can serve, while observed response data may lead to revised segments. STP should therefore be treated as an iterative decision system rather than a one-time report.

A useful distinction is between a descriptive segment and an actionable segment. A descriptive segment may show that customers differ on age or purchase frequency. An actionable segment additionally indicates what the organization should do differently—for example, offer premium support, use a particular channel, or adjust product features.

Market heterogeneity and response heterogeneity

Market heterogeneity refers to differences in customer characteristics or needs. Response heterogeneity refers to differences in how customers respond to marketing actions. Response heterogeneity is often more strategically useful. Two customers may have different ages but respond similarly to price, while two customers of the same age may have very different willingness to pay.

Where possible, segmentation should connect customer descriptors to outcomes such as conversion, retention, margin, satisfaction, or product adoption. This reduces the risk of creating attractive-sounding but commercially irrelevant profiles.

2. Segmentation Bases and Data

Demographic and geographic variables

Demographic variables include age, income, occupation, education, household structure, and life stage. Geographic variables include country, region, climate, urbanization, distance to store, and local economic conditions. These variables are often inexpensive to obtain and useful for media planning, distribution, and regulatory analysis.

Their limitation is that they are usually indirect indicators of needs. Income may correlate with purchasing power, but it does not by itself reveal preferences or price sensitivity. Demographic variables should therefore be used as explanatory or activation variables rather than assumed to be the true cause of behavior.

Psychographic and attitudinal variables

Psychographic segmentation uses values, lifestyles, personality, interests, opinions, and attitudes. Examples include preference for convenience, environmental concern, innovation orientation, risk avoidance, or desire for social recognition.

Psychographic data commonly come from surveys, interviews, social listening, and inferred digital behavior. Survey items should be clearly worded and measured consistently. Reliability and validity matter: a group of questions intended to measure “brand involvement” should actually represent that construct rather than unrelated opinions.

Behavioral variables

Behavioral segmentation uses observed actions, including:

• purchase frequency and recency;
• monetary value and margin contribution;
• product or category usage;
• browsing, search, and content engagement;
• response to promotions;
• channel and device preference;
• service contacts and complaints;
• churn, renewal, and referral behavior.

Behavior is often more directly actionable than demographics. However, observed behavior is influenced by previous marketing activity, availability, prices, and measurement quality. A customer’s low purchase frequency may reflect low need, poor distribution, or a failed campaign.

Needs-based segmentation

Needs-based segmentation groups customers according to the problem they are trying to solve. For example, users of a banking service may seek low fees, financial coaching, speed, security, or premium status. Needs-based segmentation is especially useful for product design and positioning because it links segments to desired benefits.

Value-based segmentation

Value-based segmentation considers current and expected economic contribution. A simple historical customer value measure is contribution margin:

where CM_i is customer i’s contribution margin, R_i is revenue from that customer, and C_i is variable cost attributable to serving the customer. Both quantities should use the same time period and currency. Fixed costs should not be subtracted when the purpose is to compare incremental customer contribution.

For longer-term decisions, customer lifetime value (CLV) is more appropriate. A simplified discounted model is:

where [[M_{i,t}]] is expected contribution margin before service cost in period t, p_i,t is the probability that the customer remains active in period t, S_i,t is expected service or retention spending, d is the discount rate per period, T is the planning horizon, and AC_i is acquisition cost. The model assumes that expected margins, retention probabilities, costs, and discounting are estimated consistently. It is not a universal truth about customer value; it is a decision model dependent on forecasts.

CLV should not be used as a justification for treating low-value customers unfairly. It is one input into resource allocation, alongside fairness, legal obligations, strategic learning, and service commitments.

Data preparation

Segmentation quality depends heavily on data preparation. Analysts should:

1. define the unit of analysis, such as person, household, account, store, or transaction;
2. remove duplicate records and investigate missing values;
3. align observation periods and avoid using future information;
4. distinguish structural zeros from missing observations;
5. transform strongly skewed variables where appropriate;
6. standardize variables when distance-based methods are used;
7. document data provenance, permissions, and limitations.

Standardization is important because variables measured in large units can dominate distance calculations. A common z-score is:

where x_i is an observation, μ is the sample mean, and σ is the sample standard deviation. This transformation produces unitless values and is suitable when relative deviations from the mean are meaningful. It can be distorted by extreme outliers, so robust alternatives may be considered.

3. Clustering for Segment Discovery

Clustering is an unsupervised learning approach: the analyst does not initially know the segment labels. The algorithm groups observations according to a similarity or dissimilarity measure.

k-means clustering

In k-means, the analyst specifies the number of clusters, k. The algorithm assigns each observation to the nearest cluster centroid and updates centroids repeatedly. Its objective is to minimize within-cluster squared Euclidean distance:

where C_j is cluster j, x_i is the feature vector for observation i, and μ_j is the centroid of cluster j. The method is most appropriate for numeric variables where Euclidean distance is meaningful. It is sensitive to scaling, outliers, initial seeds, and the chosen k. It also tends to produce roughly compact, spherical clusters.

The algorithm generally follows these steps:

1. choose k initial centroids;
2. assign each observation to its nearest centroid;
3. recompute each centroid as the mean of its assigned observations;
4. repeat assignment and recomputation until convergence or a stopping condition is reached.

Because different initializations can produce different solutions, analysts should use multiple starts and examine whether the results are reproducible.

Hierarchical and alternative methods

Hierarchical clustering builds a nested sequence of groups and can be displayed using a dendrogram. It is useful for exploring possible numbers of clusters but may be computationally demanding for large datasets.

For mixed data, alternatives include k-prototypes, latent class analysis, model-based clustering, and methods using appropriate categorical or mixed-type distances. The method must match the measurement scale. Treating nominal categories as if the numerical difference between category codes were meaningful is incorrect.

Choosing the number of clusters

No single statistic determines the correct number of segments. Analysts should combine:

• internal fit measures, such as within-cluster variation and silhouette analysis;
• stability across samples, time periods, and random initializations;
• substantive interpretability;
• segment size and reachability;
• expected managerial value;
• operational cost of serving each segment.

The silhouette coefficient for observation i is commonly expressed as:

where a(i) is the average distance from i to observations in its own cluster, and b(i) is the smallest average distance from i to observations in another cluster. Values range from −1 to 1. Higher values generally indicate better separation, but the measure does not establish that a segment is strategically useful.

Worked example: interpreting clusters

Suppose a retailer standardizes three variables—annual visits, promotion response, and average contribution—and applies k-means with k = 3. The resulting groups contain 40%, 35%, and 25% of customers. Their profiles are:

The labels are interpretations, not outputs of the algorithm. The analyst must validate them using independent periods and test whether different actions produce different outcomes. For example, Segment A may respond to exclusive service rather than discounts, while Segment B may respond to targeted promotions. Segment C may not justify expensive individualized treatment, but a low-cost reactivation campaign could still be tested.

4. Classification and Predictive Segment Assignment

Clustering discovers groups; classification assigns new or existing customers to known groups. A business may first identify segments through clustering and then train a classification model using easily available variables so that new customers can be assigned quickly.

Methods include logistic regression, decision trees, random forests, support vector machines, and neural networks. The choice depends on sample size, prediction objective, interpretability, and governance requirements.

For binary segment membership or response, logistic regression models the log-odds of an outcome:

where p(x) is the predicted probability of the outcome, β₀ is the intercept, β_j are coefficients, and x_j are predictor variables. The model assumes a linear relationship between predictors and log-odds unless transformations or interactions are included. A classification threshold should be selected according to the cost of false positives and false negatives, not automatically set at 0.5.

Classification performance should be evaluated on data not used for training. Relevant metrics include accuracy, precision, recall, F1 score, area under the ROC curve, calibration, and subgroup fairness. A model with high accuracy may still be commercially poor if it fails to identify a small but valuable segment.

5. Evaluating Segment Attractiveness

A segment is not automatically attractive because it is statistically distinct. A practical evaluation considers the following criteria:

A simple weighted scoring model can support discussion:

where Score_j is the score for segment j, w_m is the weight assigned to criterion m, and r_jm is segment j’s rating on criterion m. Usually, weights sum to 1 and ratings use a documented scale. This is a decision aid, not an objective measurement. Sensitivity analysis should examine whether changing plausible weights changes the ranking.

Segment stability

Stability can be assessed by comparing assignments across time, samples, or model specifications. The adjusted Rand index is one possible agreement measure when comparing two partitions, although the interpretation depends on the data and clustering method. More practically, analysts should track:

• membership migration;
• profile changes;
• value and response changes;
• predictive assignment accuracy;
• sensitivity to sampling and variable selection.

A rapidly changing segment may still be useful for real-time personalization, but it is less suitable for a long-term proposition requiring major investment.

6. Targeting and Resource Allocation

Targeting involves selecting one or more segments and determining the level of investment. Common approaches include:

• Undifferentiated targeting: one broad offer for the market.
• Differentiated targeting: distinct offers or programs for several segments.
• Concentrated targeting: focus on one specialized segment or niche.
• Customized or one-to-one targeting: individualized treatment using customer-level data.

The appropriate approach depends on demand heterogeneity, economies of scale, brand architecture, operational capability, and privacy constraints.

Targeting should consider incremental profit rather than response rate alone. A campaign’s incremental contribution can be represented as:

where ΔΠ_j is incremental profit for segment j, N_j is the number of customers exposed, p_1j and p_0j are expected purchase probabilities with and without the intervention, m_1j and m_0j are contribution margins under those conditions, and C_j is incremental campaign cost. The model assumes that the difference between treatment and control outcomes represents the intervention’s causal effect. Without a control group or credible causal design, observed differences may reflect pre-existing customer differences.

Worked targeting example

Assume a segment contains 10,000 customers. A randomized test estimates a purchase probability of 18% with a campaign and 15% without it. Contribution margin per purchase is £20 in both conditions, and campaign cost is £8,000.

Incremental purchases are:

10,000 × (0.18 − 0.15) = 300.

Incremental contribution is:

300 × £20 = £6,000.

Thus, incremental profit is:

£6,000 − £8,000 = −£2,000.

Although the campaign increased response, it destroyed value under these assumptions. The firm might reduce cost, improve the offer, target a more responsive subgroup, or discontinue the intervention. This example demonstrates why targeting decisions should use incremental economics rather than descriptive conversion rates.

7. Positioning Analytics

Positioning concerns the associations customers hold about a brand or offering relative to competing alternatives. A positioning statement typically identifies a target audience, frame of reference, point of difference, and reason to believe. Analytics helps test whether these elements are understood, valued, distinctive, and credible.

Perceptual maps

A perceptual map represents brands or products in a reduced-dimensional space based on similarity judgments, attribute ratings, or observed choice. Multidimensional scaling can map perceived similarity; factor analysis or principal components can summarize correlated attributes. A map is an analytical representation, not a literal picture of the customer’s mind.

Managers should distinguish:

• current perception: where customers believe the brand is;
• ideal point: what customers prefer;
• competitive position: where alternatives are perceived;
• feasible position: where the firm can credibly compete.

A gap between current and ideal position is not automatically an opportunity. The gap may reflect low demand, high cost, weak credibility, or an attractive position already occupied by a strong competitor.

Choice and preference analysis

Conjoint analysis estimates how attributes and levels contribute to preference or choice. A basic choice model may express the utility of option j for person i as:

where U_ij is latent utility, x_j is the vector of product attributes, β_i is the person’s preference-weight vector, and ε_ij is an unobserved component. The model assumes that utility is represented adequately by the included attributes and that the error structure matches the estimation method. Utilities are relative; they are not measured in currency unless the model includes a valid price attribute and appropriate assumptions.

If a logit choice model is used, the probability of choosing option j among a choice set J is often written:

This expression assumes the relevant multinomial logit structure, including the independence of irrelevant alternatives property. More flexible models may be required when that assumption is unrealistic.

Positioning decisions should combine preference with economic feasibility, brand meaning, competitor response, and delivery capability. A highly preferred claim that cannot be reliably delivered is not a sustainable position.

8. Illustrative Case: FreshRoute Grocery

Background

FreshRoute is a fictional regional grocery-delivery company serving urban and suburban households. Its customer database includes orders, basket value, delivery location, app activity, promotion use, customer service contacts, and a short annual survey about shopping motivations.

Problem

FreshRoute had rising acquisition costs and used the same 15% discount for almost all new and existing customers. Management wanted to identify profitable segments, improve retention, and clarify the brand’s position against low-price supermarkets and premium organic specialists.

Analysis

The analytics team defined the household account as the unit of analysis and used six months of behavior plus survey responses. It excluded post-campaign variables from the initial targeting model to reduce leakage. Numeric variables were standardized, and extreme basket values were reviewed rather than automatically deleted.

Clustering identified four interpretable groups:

1. Convenience-led professionals: frequent evening orders, high willingness to pay for delivery reliability, and low coupon dependence.
2. Value-seeking families: large baskets, high promotion response, and strong sensitivity to delivery fees.
3. Health-oriented planners: high interest in fresh and organic products, moderate frequency, and strong content engagement.
4. Occasional shoppers: low frequency, low contribution, and inconsistent app activity.

The team validated the clusters in a later two-month period. The groups remained recognizable, although some households migrated between the first three groups. This indicated useful behavioral tendencies rather than permanent identities.

Approach and solution

FreshRoute selected the first three segments for differentiated programs. It tested:

• priority delivery windows and subscription benefits for convenience-led professionals;
• threshold-based basket incentives rather than blanket discounts for value-seeking families;
• curated organic bundles, meal planning content, and supplier stories for health-oriented planners.

A randomized holdout group was maintained for each program. The company also built a transparent classification model using order frequency, basket composition, and channel behavior to assign new customers provisionally. Customers could move between programs as behavior changed.

For positioning, survey-based perceptual analysis showed that FreshRoute was perceived as more convenient than traditional supermarkets but less credible than specialist organic services on product expertise. The firm therefore adopted the focused position “reliable fresh shopping for busy households,” rather than claiming to be the cheapest or most specialized organic provider.

Outcome

The company observed improved contribution in the first and third segments and reduced unnecessary discounting in the second. The results were interpreted cautiously because the test lasted only one quarter and seasonal effects remained possible. Management decided to continue measurement through retention, incremental margin, complaint rates, and customer fairness indicators.

Learning points

The case illustrates five principles. First, segments should be connected to distinct needs and actions. Second, value-based targeting should use incremental economics rather than historical revenue alone. Third, segment membership can change. Fourth, positioning must be credible and competitively differentiated. Fifth, experimentation and holdout groups are necessary for evaluating causal impact.

9. Ethics, Governance, and Contemporary Developments

STP analytics can affect prices, access, service quality, and representation. Important governance practices include:

• collect only data needed for a legitimate purpose;
• obtain appropriate consent and communicate material uses clearly;
• protect personal data through access controls, minimization, retention limits, and security;
• assess whether proxy variables reproduce prohibited discrimination;
• test model performance and calibration across relevant groups;
• provide human review and appeal mechanisms for consequential decisions;
• avoid targeting vulnerable customers with exploitative offers;
• document assumptions, versions, data sources, and decision rights.

Privacy regulations vary by jurisdiction, but responsible practice extends beyond legal compliance. A segmentation model can be lawful yet harmful if it systematically excludes customers from beneficial offers or infers sensitive attributes without a defensible purpose.

Artificial intelligence can support natural-language analysis of survey comments, look-alike modeling, recommendation, and automated campaign selection. However, AI does not remove the need for valid measurement, causal testing, explainability, and governance. Automated personalization can create filter bubbles, inconsistent prices, or discriminatory outcomes if objectives and constraints are poorly designed.

Contemporary practice increasingly combines first-party data, privacy-preserving measurement, experimentation, real-time decisioning, and customer data platforms. The strategic trend is toward dynamic segments, but dynamic does not mean uncontrolled. Firms should define update frequency, minimum sample sizes, model drift thresholds, and procedures for suppressing unreliable or harmful decisions.

Best-Practice Workflow

A defensible STP analytics project can follow this workflow:

1. Define the strategic decision and success metric.
2. Specify the customer unit and observation period.
3. Audit data quality, consent, representativeness, and leakage risks.
4. Select segmentation variables linked to needs, behavior, value, or response.
5. Prepare and scale data appropriately.
6. Compare plausible clustering or latent-structure methods.
7. Profile and name segments using evidence, not stereotypes.
8. Test segment stability and response differentiation.
9. Evaluate attractiveness, risk, fit, and incremental economics.
10. Select targets and design differentiated actions.
11. Quantify and test positioning against customer preferences and competitors.
12. Use experiments, monitoring, and feedback to revise the system.

Challenges and Limitations

Segmentation is subject to several limitations. Data may be incomplete, biased toward digitally active customers, or generated by earlier marketing decisions. Clustering can produce groups even when no meaningful natural groups exist. Analysts may overfit a large dataset and mistake random patterns for strategy. Segment labels can become stereotypes, and organizational systems may be unable to execute the recommended differentiation.

Value models depend on uncertain forecasts of retention, margin, and cost. Perceptual maps simplify complex meanings and may change with context. Classification models can remain accurate while becoming unfair or commercially obsolete. Finally, targeting based only on short-term profit can damage trust, brand equity, or long-term customer relationships.

Sound practice therefore combines statistical evidence with customer research, managerial judgment, ethical review, experimentation, and ongoing validation.

Chapter Summary

Segmentation, targeting, and positioning analytics links customer heterogeneity to marketing decisions. Demographic, geographic, psychographic, behavioral, needs-based, and value-based data each provide different insights, but actionable segmentation requires a connection between a segment’s characteristics and a feasible marketing action. Data preparation, appropriate measurement, and careful definition of the unit of analysis are foundational.

Clustering can discover candidate segments, while classification can assign new customers to established segments. Neither method guarantees strategic usefulness. Analysts must evaluate interpretability, stability, accessibility, response differentiation, economic value, organizational fit, and ethical risk. Targeting should emphasize incremental contribution and causal evidence rather than response rates alone.

Positioning analytics uses perceptual mapping, preference measurement, and choice models to identify valued and differentiated positions. A successful position must also be credible and operationally deliverable. Modern AI and dynamic personalization expand analytical capability but increase the importance of privacy, fairness, transparency, monitoring, and governance.

Key Terms

Actionable segment: A customer group for which a distinct and feasible marketing action can be designed.

Classification: Supervised learning that assigns observations to predefined categories.

CLV: The discounted expected contribution from a customer over a specified future horizon, net of relevant costs.

Clustering: Unsupervised grouping of observations according to similarity.

Contribution margin: Revenue minus variable costs attributable to a customer, product, or activity.

Differentiability: The extent to which segments respond differently to marketing actions.

Dynamic segment: A segment whose membership or profile is updated as new data become available.

k-means: A clustering method that minimizes within-cluster squared Euclidean distance for a specified number of clusters.

Needs-based segmentation: Grouping customers according to the problems or benefits they seek.

Perceptual map: A reduced-dimensional representation of perceived similarities or attribute positions.

Positioning: The intended and perceived place of an offering relative to alternatives.

Psychographic segmentation: Grouping based on values, lifestyles, attitudes, interests, or personality.

Segment stability: The persistence and reproducibility of segment membership and profiles.

Substantiality: The economic or strategic significance of a segment.

Targeting: Selecting segments and allocating resources to serve them.

Value-based segmentation: Grouping customers according to current or expected economic contribution.

Within-cluster variation: The degree of dispersion among observations assigned to the same cluster.

Response heterogeneity: Differences among customers in their responses to marketing actions.

Review Questions

Short Answer Questions

1. Distinguish between descriptive segmentation and actionable segmentation.
2. Why should variables be standardized before many distance-based clustering methods?
3. State two limitations of demographic segmentation.
4. What is the difference between clustering and classification?
5. Why is incremental profit more useful than campaign response rate alone for targeting decisions?

Long Answer Questions

1. Explain how demographic, psychographic, behavioral, needs-based, and value-based segmentation can be combined in a marketing analytics project.
2. Describe the k-means algorithm, including its objective, assumptions, and major limitations.
3. Develop a framework for evaluating segment attractiveness and stability.
4. Explain how perceptual maps and conjoint or choice analysis can support positioning decisions.
5. Discuss the ethical and governance challenges of AI-enabled segmentation and personalization.

Analytical / Application Questions

1. A retailer obtains four clusters but finds that one cluster is small, difficult to reach, and no more responsive than the broad market. Should it be retained? Justify your decision using segment attractiveness criteria.
2. A campaign increases conversion from 10% to 14% in a target segment, but the incremental cost per customer is high. Design an analysis that determines whether the campaign creates incremental profit.
3. A bank’s classification model assigns customers to premium-service segments using income, location, digital activity, and service history. Propose a validation and fairness assessment before deployment.

Multiple Choice Questions with Answers

Question 1

Which characteristic most directly indicates that a segment is actionable?

A. It has a distinctive name

B. It can be linked to a feasible marketing intervention

C. It contains customers from different regions

D. It is the largest segment

Correct Answer: B. It can be linked to a feasible marketing intervention

Question 2

What is the primary purpose of standardizing variables before distance-based clustering?

A. To remove all missing values

B. To make categorical variables numerical

C. To prevent variables with large units from dominating distance

D. To guarantee statistically significant clusters

Correct Answer: C. To prevent variables with large units from dominating distance

Question 3

Which method is unsupervised?

A. Logistic regression classification

B. Decision-tree prediction

C. k-means clustering

D. Campaign response modeling with labeled outcomes

Correct Answer: C. k-means clustering

Question 4

In the k-means objective, the algorithm seeks to minimize what quantity?

A. Between-period revenue growth

B. Within-cluster squared distances

C. The number of customer records

D. The marketing budget

Correct Answer: B. Within-cluster squared distances

Question 5

A campaign reaches 10,000 customers, raises purchase probability from 15% to 18%, and earns £20 contribution per incremental purchase. Campaign cost is £8,000. What is incremental profit?

A. £2,000

B. £6,000

C. −£2,000

D. −£8,000

Correct Answer: C. −£2,000

Question 6

Which criterion asks whether a firm can effectively reach and serve a segment?

A. Accessibility

B. Substantiality

C. Differentiability

D. Stability

Correct Answer: A. Accessibility

Question 7

What does classification do after segments have been defined?

A. It assigns observations to known categories

B. It always creates new clusters

C. It eliminates the need for validation

D. It measures only brand awareness

Correct Answer: A. It assigns observations to known categories

Question 8

Which statement about a perceptual map is most accurate?

A. It is a literal geographic map of customers

B. It proves that an unoccupied position is profitable

C. It represents perceived relationships in a reduced analytical space

D. It measures only historical revenue

Correct Answer: C. It represents perceived relationships in a reduced analytical space

Question 9

Why should a holdout or control group be used in targeting evaluation?

A. To increase the apparent campaign response

B. To estimate what would have happened without the intervention

C. To guarantee every customer receives the offer

D. To eliminate the need for cost data

Correct Answer: B. To estimate what would have happened without the intervention

Question 10

Which is an important governance concern in automated segmentation?

A. Whether segment names are alphabetized

B. Whether proxy variables create discriminatory outcomes

C. Whether every segment has exactly equal size

D. Whether clustering always uses k = 3

Correct Answer: B. Whether proxy variables create discriminatory outcomes