Customer Segmentation, Targeting and Personas

Introduction and Context

Customer and digital analytics seek to convert customer data into better decisions about acquisition, service, retention and growth. Segmentation is central to this process. A firm rarely benefits from treating every customer as identical because customers differ in needs, purchasing power, channel preferences, responsiveness and likely future value. Segmentation identifies meaningful groups; targeting determines which groups should receive attention; personas translate analytical findings into an accessible representation of typical customers; and differentiated value propositions explain why an offering is relevant to each target group.

Segmentation is not simply the creation of attractive demographic profiles. A useful segment must be identifiable, sufficiently substantial, reachable, responsive to a distinct marketing action and economically or strategically worthwhile. Digital environments make segmentation more powerful because organizations can observe browsing, search, app, transaction, service and engagement behavior. They also create risks involving privacy, discrimination, excessive personalization and opaque automated decision-making.

Learning Objectives

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

1. Explain demographic, behavioral, needs-based, value-based and lifecycle segmentation.
2. Calculate and interpret RFM measures and customer scores.
3. Explain how clustering and supervised scoring support segmentation.
4. Assess segment validity, attractiveness and operational usability.
5. Translate analytical segments into targeting strategies, personas and differentiated value propositions.
6. Evaluate ethical, privacy and governance issues in customer segmentation.

1. The Logic of Segmentation, Targeting and Positioning

Segmentation divides a heterogeneous market into relatively homogeneous groups whose members are expected to respond similarly to a marketing intervention. Targeting is the selection of one or more segments for resource allocation. Positioning is the intended place of a brand or offering in the target customer’s mental model relative to alternatives. A value proposition states the benefits delivered, the customer problem addressed and the reason to believe the organization can deliver those benefits.

The process can be represented as:

Data and research → Segment formation → Segment evaluation → Target selection → Persona and value proposition → Activation → Measurement and refinement

The sequence is iterative rather than purely linear. Campaign results may reveal that a segment was poorly defined, that the proposed offer was not differentiated, or that the channel did not reach the intended customers.

Segmentation versus Classification

Segmentation is usually exploratory or strategic: it discovers groups without initially knowing the correct labels. Classification or scoring is predictive: it assigns an individual to a known category or estimates a future outcome, such as churn or response. For example, clustering may discover four customer groups, whereas a logistic regression model may estimate the probability that each customer will respond to an offer.

A segment should be more than a descriptive cluster. It should support a decision. If two groups have different statistical profiles but require exactly the same offer, channel and service model, separating them may add complexity without value.

2. Major Bases of Segmentation

2.1 Demographic and Firmographic Segmentation

Demographic variables include age, household composition, income, education and location. Firmographic variables perform a similar role in business-to-business markets and include industry, firm size, revenue, ownership and operating geography. These variables are often easy to obtain and explain, but they are imperfect proxies for motivation and behavior. Two customers of the same age and income may have entirely different needs and price sensitivity.

Demographic variables should therefore be used carefully. They may improve media planning or service design, but they should not be assumed to explain behavior without validation. Sensitive characteristics may also create legal and ethical risks if used to deny access, vary prices unfairly or infer protected attributes.

2.2 Behavioral Segmentation

Behavioral segmentation groups customers according to observed actions, such as:

• purchase frequency, category mix and basket size;
• product usage, feature adoption and tenure;
• website searches, pages viewed and app events;
• email opens, clicks and content engagement;
• customer-service contacts and complaint patterns; and
• promotion, channel and payment preferences.

Behavior is often more actionable than demographics. A streaming service can respond directly to viewing intensity or genre preference, while an online retailer can use purchase recency and category affinity. However, behavior is context-dependent. A low purchase frequency may indicate dissatisfaction, an infrequent-use product, seasonality or a customer who is still in the consideration stage.

2.3 Needs-Based and Attitudinal Segmentation

Needs-based segmentation identifies the problem customers are trying to solve and the benefits they value. Attitudinal research may measure motivations, perceived risk, convenience orientation, identity, trust and price consciousness. Methods include interviews, focus groups, surveys, conjoint analysis and qualitative analysis of reviews or service conversations.

Needs-based segments are strategically valuable because they connect customer insight to product and communication design. Their difficulty is measurement: needs are latent constructs and may not be reliably inferred from a single click or purchase. Combining qualitative insight with behavioral evidence is usually stronger than relying on either alone.

2.4 Value-Based Segmentation

Value-based segmentation distinguishes customers according to current profitability or expected customer lifetime value (CLV). Revenue alone is insufficient because customers generating high sales may also incur high discounts, returns, service costs or fulfillment costs.

A simple historical contribution measure is:

Here, all terms refer to customer i over a defined period and must use consistent currency and time boundaries. This is not a complete CLV model because it excludes future cash flows.

A simplified discounted CLV model is:

Where m is expected contribution margin per period, p is the probability that the customer remains active or purchases in period t, c is customer-specific service or retention cost, r is the discount rate per period, T is the forecast horizon, and AC is acquisition cost. The model assumes that the estimates are reasonably calibrated and that cash flows occur at comparable periodic intervals. More advanced models incorporate contractual renewal, transaction frequency, changing margins, referral value and uncertainty.

Value-based segmentation supports differentiated investment. High-value customers may justify premium service, while low-value customers may be served through automation. This should not become an excuse for neglecting customers or providing discriminatory access to essential services.

2.5 Lifecycle Segmentation

Lifecycle segments describe a customer’s relationship stage, such as prospect, new customer, active customer, developing customer, at-risk customer, lapsed customer or reactivated customer. Lifecycle logic is especially useful for journey design. A new customer may need onboarding, an active customer may need cross-selling, and an at-risk customer may need problem resolution before promotion.

Lifecycle is not identical to value. A new customer may have high future potential but little observed history; a long-tenured customer may be low-value but strategically important. Combining lifecycle, behavior and value produces more useful decisions than using a single dimension.

3. RFM Analysis

RFM is a practical descriptive method based on three customer-level variables:

• Recency (R): time since the most recent purchase. Lower elapsed time generally indicates greater recent activity.
• Frequency (F): number of qualifying purchases during a specified period.
• Monetary value (M): monetary contribution or purchase amount during that period.

The observation window, treatment of returns, definition of a purchase and currency must be stated. RFM is most appropriate where repeat transactions are meaningful and transaction history is sufficiently complete. It is less informative for infrequent durable goods, subscriptions with identical periodic charges or customers whose purchases occur through unobserved channels.

A common procedure is to rank customers separately on R, F and M and assign scores from 1 to 5. For recency, the most recent customers receive the highest score; for frequency and monetary value, the highest observed values receive the highest score. A customer’s code may be written as RFM = 5-4-5. The codes are ordinal rather than interval measurements: a score of 5 is not necessarily five times a score of 1.

Worked Example: RFM Scoring

Suppose a retailer uses quartiles and scores customers from 1 to 4. For recency, 1–30 days receives 4, 31–90 days receives 3, 91–180 days receives 2 and more than 180 days receives 1. For frequency, 1 purchase receives 1, 2–3 receives 2, 4–6 receives 3 and more than 6 receives 4. For monetary value, the ranges are set from the organization’s distribution.

Customer A purchased 14 days ago, made 7 purchases in the observation window and spent £620. If £620 belongs to the highest monetary band, the customer receives R = 4, F = 4 and M = 4, producing RFM 4-4-4. Customer B purchased 75 days ago, made 2 purchases and spent £180 in the second monetary band, producing RFM 3-2-2.

A simple aggregate index is:

For Customer A, the index is 4 + 4 + 4 = 12. For Customer B, it is 3 + 2 + 2 = 7. This index is easy to communicate, but it assumes equal importance for the three dimensions. An organization may instead use weights:

where the weights sum to 1 if the result is intended to remain on the same score scale. Weights should be justified by evidence or decision objectives, not chosen merely to produce a preferred ranking.

RFM groups might include champions, loyal customers, promising new customers, large but inactive customers and lapsed customers. These labels are hypotheses requiring validation. A “champion” group should be checked for profitability, returns, channel cost and response to treatment.

4. Clustering and Predictive Scoring

4.1 Clustering

Clustering algorithms group observations so that members within a group are relatively similar and members across groups are relatively different according to selected variables. K-means is a widely used method for numeric data. Given k clusters, it seeks to minimize within-cluster squared Euclidean distance:

Here, Cⱼ is cluster j, xᵢ is the vector of measurements for customer i, and μⱼ is the centroid of cluster j. K-means assumes that numeric variables are meaningfully scaled, that means are suitable representatives and that roughly compact clusters are useful. It is sensitive to outliers and initial centroid selection.

A standard workflow is:

1. Define the business population and observation window.
2. Select variables linked to a decision, such as frequency, margin, category breadth and digital engagement.
3. Clean missing values and investigate outliers.
4. Transform skewed variables where justified and standardize variables with different units.
5. Estimate several candidate values of k.
6. Compare statistical diagnostics with managerial interpretability.
7. Profile clusters and test whether they differ on outcomes not used to create them.
8. Deploy only if the clusters are stable, reachable and actionable.

Standardization is often expressed as:

where μ is the sample mean and σ is the sample standard deviation. It produces a unitless value and prevents a variable measured in large currency units from automatically dominating distance calculations. Standardization does not solve all problems: highly skewed variables may still require a justified transformation, and irrelevant variables can dilute meaningful patterns.

The number of clusters should not be chosen solely by a statistical index. A very large number may fit noise and create unmanageable campaigns; a very small number may conceal important differences. Stability over time, business interpretability, incremental performance and operational cost matter as much as mathematical fit.

4.2 Predictive Scoring

A score estimates the probability or expected value of a future event. For binary response, logistic regression is a common interpretable model:

Here, pᵢ is the predicted probability for customer i, β₀ is the intercept, βⱼ are estimated coefficients and xᵢⱼ are customer features. The model assumes that the log-odds are linearly related to the predictors, unless transformations or nonlinear terms are added. Predictions should be assessed on data not used to fit the model.

Scoring is useful for prioritizing retention calls, identifying likely responders or estimating churn risk. It does not automatically create a segment. A score becomes a targeting segment when a threshold or set of bands is connected to a differentiated action and the consequences of errors are understood.

4.3 Segmentation and Causal Decision-Making

A customer who is likely to buy is not necessarily a customer who needs an advertisement. Predictive response models estimate what may happen, whereas uplift or incremental-response analysis asks how an intervention changes behavior relative to no intervention. Where randomized control groups are available, incremental impact can be estimated as:

This difference is interpretable as a causal estimate only when treatment assignment and measurement support a valid comparison, such as randomization or a credible quasi-experimental design. Targeting customers solely because they have high predicted purchase probability can waste budget on people who would have purchased anyway.

5. Evaluating Segment Validity and Attractiveness

A segment is analytically credible when it is internally coherent, meaningfully different from other segments and sufficiently stable. It is commercially useful when an organization can reach it, serve it and measure the effect of doing so.

5.1 Validity Criteria

Important criteria include:

• Identifiability: the segment has measurable characteristics and a reproducible rule.
• Distinctiveness: its needs, behavior or expected response differ materially from other segments.
• Stability: membership and defining characteristics do not change arbitrarily because of minor data fluctuations.
• Reachability: the firm can identify members in operational systems and access them through suitable channels.
• Actionability: a specific product, price, message, service or channel decision can be applied.
• Completeness: relevant customers are not systematically omitted because of missing data or channel fragmentation.
• Ethical acceptability: the segment does not encode unjustified discrimination or exploit vulnerability.

5.2 Attractiveness Criteria

Attractiveness includes current and future size, growth, margin, competitive intensity, strategic fit, cost to serve, risk and resource requirements. A large segment may be unattractive if it is expensive to serve or highly contested. A small segment may be valuable if it has high margin, strong referral potential or strategic importance.

A practical evaluation table can score each candidate segment using a defined scale. For example, a firm may rate expected contribution, growth, reachability and strategic fit from 1 to 5, document the evidence and apply explicit weights. Weighted scores are decision aids, not objective truths. Sensitivity analysis should test whether a different weighting would change the target choice.

6. Targeting Strategies

Four broad approaches are common:

Targeting should consider not only expected return but also fairness, customer experience and operational feasibility. A differentiated strategy is not automatically superior. Excessively many offers can confuse customers, increase production costs and make brand meaning inconsistent.

7. Personas and Differentiated Value Propositions

A persona is a research-based representation of a meaningful customer type. It should combine evidence with a concise narrative: context, goals, barriers, behaviors, decision criteria, preferred channels and relevant service needs. A persona is not a fictional biography designed from stereotypes.

A robust persona normally includes:

• a neutral name or label;
• segment membership rule and evidence base;
• goals and desired outcomes;
• frustrations, barriers and perceived risks;
• behaviors and triggers;
• decision criteria and alternatives considered;
• preferred touchpoints and accessibility needs; and
• appropriate offer, message and success measures.

Demographic details should be included only when relevant to the decision. A persona should state uncertainty where evidence is limited. For example, “often compares delivery times before purchase” is stronger than “is busy” if the former is supported by observed behavior.

A differentiated value proposition connects a customer problem with a credible benefit. A useful structure is:

For [target segment], who [need or situation], our [offering] provides [primary benefit] because [reason to believe], unlike [alternative], which [relevant limitation].

Consider an illustrative online grocery business:

• Time-constrained planners: value reliable scheduled delivery and simple repeat ordering. The proposition emphasizes predictability and reduced planning effort.
• Value-seeking families: value total-basket savings and transparent promotions. The proposition emphasizes affordability without hiding conditions.
• Exploratory food enthusiasts: value discovery, provenance and recipe inspiration. The proposition emphasizes curation and product knowledge.

These propositions should be tested through experiments, qualitative feedback and behavioral outcomes rather than assumed to be correct because they sound appealing.

8. Applied Case: Northstar Fitness (Illustrative)

Background

Northstar Fitness is an illustrative digital subscription business offering workout programs, nutrition content and connected-device integration. Management observed high acquisition but declining three-month retention. It had demographic data, app events, subscription history, support records and campaign exposure data.

Problem

The company sent the same renewal messages to all subscribers. Management needed to determine whether distinct customer groups required different onboarding and retention actions without creating unfair or intrusive personalization.

Analysis

The analytics team defined the observation window as the first 60 subscription days and created features including completed workouts, active days, program variety, device connection, support contacts, trial-to-paid conversion and early content engagement. Monetary value was represented by contribution margin rather than subscription revenue alone, because refunds and support costs varied.

The team first used descriptive lifecycle and RFM-style measures. It then applied standardized clustering to engagement and margin variables, testing two through six clusters. A four-cluster solution was selected because it was reasonably stable across resamples and produced interpretable operational groups:

1. Routine adopters: frequent activity, strong retention and moderate support needs.
2. Explorers: broad content use but inconsistent weekly activity.
3. Device-dependent users: high engagement when the device was connected and elevated support contacts when synchronization failed.
4. Quiet trialists: low activity after registration and high risk of early cancellation.

The clusters were profiled using variables not used in formation, including cancellation outcomes and satisfaction survey results. This external profiling helped determine whether the clusters had meaningful differences rather than merely reflecting the chosen inputs.

Approach and Solution

Northstar introduced differentiated interventions. Routine adopters received progress recognition and advanced content. Explorers received structured weekly plans and reminders based on declared goals. Device-dependent users received proactive synchronization guidance and an easy support route. Quiet trialists received a short onboarding sequence focused on one achievable first workout rather than a large content catalogue.

The company used a holdout control group for each intervention. The primary outcome was retained paid subscription at day 90; secondary outcomes included workout completion, support resolution and unsubscribe rate. It avoided using inferred health conditions or sensitive demographic characteristics for targeting.

Outcome

In this illustrative case, the differentiated program produced a higher day-90 retention rate than the generic-message control, especially among quiet trialists and device-dependent users. The result was not interpreted as proof that clustering itself caused the improvement. The causal claim depended on the experimental comparison, implementation quality and absence of major contamination between treatment and control groups.

Learning Points

The case demonstrates five principles. First, a segment must lead to a different action. Second, operational variables can be more useful than demographic stereotypes. Third, clusters require validation against outcomes not used to create them. Fourth, experiments are needed to distinguish targeting value from selection effects. Fifth, privacy and restraint should be designed into the segmentation process rather than added after deployment.

9. Implementation Architecture and Measurement

Segmentation becomes useful only when it can be operationalized. A typical architecture links a customer data platform or warehouse to identity resolution, feature tables, model outputs, campaign systems and measurement environments. Definitions should be documented in a data dictionary, including event timestamps, consent status, source systems, refresh frequency and retention rules.

A segment activation process should specify:

1. eligibility rules and exclusions;
2. refresh schedule and membership history;
3. channel permissions and contact frequency;
4. treatment, control and suppression rules;
5. offer economics and budget limits;
6. outcome metrics and attribution method; and
7. review criteria for model drift or harm.

Useful measures include reach, conversion, incremental conversion, contribution margin, retention, churn, customer satisfaction, complaint rate and opt-out rate. A campaign’s return on marketing investment may be expressed as:

The numerator must use incremental contribution rather than total sales if the purpose is to evaluate the campaign’s added effect. Results should be compared with a suitable control or counterfactual whenever feasible.

10. Ethical, Legal and Governance Considerations

Customer segmentation uses personal data and can affect opportunities, prices, service levels and visibility. Key principles include purpose limitation, data minimization, accuracy, security, transparency and respect for consent and lawful processing requirements. Applicable privacy and consumer-protection laws vary by jurisdiction; organizations must obtain appropriate legal advice rather than treating analytics practice as a substitute for compliance.

Important risks include:

• Proxy discrimination: variables such as location or browsing behavior may reproduce the effect of protected characteristics.
• Exclusion: data-poor customers may receive worse service because they are not recognized by the model.
• Inference risk: sensitive traits may be inferred without the customer knowingly providing them.
• Manipulation: personalization may exploit urgency, vulnerability or limited choice.
• Surveillance: collecting every possible event can exceed what is necessary and reasonable.
• Automation opacity: customers and employees may not understand why a treatment was assigned.

Governance should include documented purposes, access controls, retention limits, model validation, fairness testing, human escalation and an appeal or correction route where decisions have significant effects. Generative AI can assist with persona drafting or qualitative coding, but generated summaries must be traced to evidence, checked for stereotypes and protected from leakage of confidential data. AI-generated personas should never replace customer research or governance review.

11. Limitations and Best Practices

Segmentation can fail because of poor identity resolution, incomplete channels, changing behavior, biased samples, unstable clusters or untested assumptions. The ecological context also matters: a customer may belong to different meaningful segments for different products or occasions. Segments are not permanent psychological truths; they are analytical representations created for a purpose and time period.

Best practice is to begin with the decision, not the algorithm. Define what action will differ, then select data that can support it. Use a combination of qualitative research and behavioral evidence. Examine distributions and missingness before modeling. Separate model-development data from validation data. Prefer interpretable segment rules when performance is comparable. Test incremental outcomes with controls. Monitor drift, profitability, customer experience and fairness after launch. Retire segments that no longer produce differentiated value.

Chapter Summary

Segmentation organizes heterogeneous customers into groups that can support differentiated decisions. Demographic, behavioral, needs-based, value-based and lifecycle approaches answer different questions and are strongest when combined appropriately. RFM provides a transparent framework for transactional analysis, while clustering discovers patterns and predictive scoring estimates future outcomes. Neither method is automatically useful: segments must be identifiable, distinct, stable, reachable, actionable, attractive and ethically acceptable.

Targeting selects segments in light of value, strategic fit, cost and capability. Personas translate evidence into a human-readable account of goals, barriers and behaviors, while value propositions connect segment needs to credible benefits. Activation requires operational rules, measurement and control groups. Finally, privacy, fairness, transparency and governance are essential because segmentation can influence customer opportunities and experiences.

Key Terms

Segmentation: Dividing a heterogeneous market into relatively homogeneous groups.

Targeting: Selecting segments for differentiated resource allocation.

Positioning: The intended place of an offering in the customer’s mental model.

Persona: A research-based representation of a meaningful customer type.

Value proposition: A statement of customer benefit, problem solved and reason to believe.

Demographic segmentation: Grouping by characteristics such as age, income or household status.

Firmographic segmentation: Grouping organizations by industry, size, revenue or similar attributes.

Behavioral segmentation: Grouping by observed actions, usage or engagement.

Needs-based segmentation: Grouping by desired outcomes, motivations or problems.

Value-based segmentation: Grouping by current or expected economic contribution.

Lifecycle segmentation: Grouping by relationship stage or customer journey position.

RFM: Recency, frequency and monetary value analysis.

Clustering: An unsupervised method for grouping similar observations.

Predictive scoring: Estimating a future outcome or probability for an individual.

Incremental response: The difference in outcome caused by an intervention relative to a counterfactual.

Customer lifetime value: The expected present value of future customer contribution.

Reachability: The ability to identify and contact members of a segment.

Actionability: The extent to which a segment supports a feasible differentiated decision.

Proxy discrimination: Unfair effects produced by variables that indirectly represent protected characteristics.

Model drift: Declining relevance or accuracy as data or customer behavior changes.

Review Questions

Short Answer Questions

1. Distinguish segmentation from targeting and positioning.
2. Define recency, frequency and monetary value in RFM analysis.
3. Why should revenue not be used as the sole measure of customer value?
4. State four criteria for evaluating segment validity.
5. What is the purpose of a control group in a targeting experiment?

Long Answer Questions

1. Compare demographic, behavioral, needs-based, value-based and lifecycle segmentation, including the strengths and limitations of each.
2. Explain how an organization could use RFM and clustering together in a customer analytics project.
3. Discuss how personas should be constructed from evidence rather than stereotypes.
4. Explain the difference between predictive response and incremental response and why the distinction matters for targeting.
5. Evaluate the ethical and governance risks associated with personalized digital segmentation.

Analytical / Application Questions

1. A retailer identifies customers with high purchase frequency but low contribution margin because of returns and discounts. Recommend a revised value-based segmentation approach and explain which metrics should be monitored.
2. Design a segmentation and targeting plan for a subscription app with high trial registration but low activation. Specify data inputs, possible segments, interventions, control design and success measures.
3. A clustering project produces eight statistically distinct groups, but the marketing team can operationally support only three differentiated campaigns. Explain how you would select or consolidate segments without losing important customer differences.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes segmentation from targeting?

A. Segmentation selects media channels, while targeting calculates profit.

B. Segmentation creates customer groups, while targeting selects groups for action.

C. Segmentation is always predictive, while targeting is always descriptive.

D. Segmentation creates advertisements, while targeting defines accounting rules.

Correct Answer: B. Segmentation creates customer groups, while targeting selects groups for action.

Question 2

In RFM analysis, recency usually measures:

A. The monetary value of the latest basket

B. The number of categories purchased

C. The time elapsed since the most recent purchase

D. The probability of responding to an offer

Correct Answer: C. The time elapsed since the most recent purchase

Question 3

A customer has R = 4, F = 3 and M = 2. Using RFMIndex = R + F + M, what is the index?

A. 6

B. 7

C. 8

D. 9

Correct Answer: B. 7

Question 4

Why are variables commonly standardized before k-means clustering?

A. To ensure every customer belongs to every cluster

B. To prevent variables with larger measurement scales from dominating distance calculations

C. To convert all categorical variables into valid causal effects

D. To guarantee that the chosen number of clusters is optimal

Correct Answer: B. To prevent variables with larger measurement scales from dominating distance calculations

Question 5

Which is a validity criterion for a useful segment?

A. It must contain exactly the same number of customers as every other segment.

B. It must be based only on demographic variables.

C. It should be identifiable, distinct and actionable.

D. It must remain unchanged permanently.

Correct Answer: C. It should be identifiable, distinct and actionable.

Question 6

What does customer lifetime value attempt to estimate?

A. Only the customer’s most recent order value

B. Expected present value of future customer contribution, subject to model assumptions

C. The number of advertisements shown to a customer

D. The average age of customers in a segment

Correct Answer: B. Expected present value of future customer contribution, subject to model assumptions

Question 7

Which statement about personas is most appropriate?

A. A persona should be based primarily on stereotypes because they are easy to communicate.

B. A persona is a research-based representation of a meaningful customer type.

C. A persona replaces the need for behavioral data.

D. A persona must describe an individual real customer by name.

Correct Answer: B. A persona is a research-based representation of a meaningful customer type.

Question 8

If a treated group has a response rate of 12% and a randomized control group has a response rate of 8%, the estimated incremental response is:

A. 4 percentage points

B. 8 percentage points

C. 12 percentage points

D. 20 percentage points

Correct Answer: A. 4 percentage points

Question 9

Why can targeting customers with the highest predicted purchase probability be inefficient?

A. High-probability customers may purchase without the intervention.

B. Predicted probabilities cannot be calculated from customer data.

C. Control groups always have higher response rates.

D. Purchase probability is identical to contribution margin.

Correct Answer: A. High-probability customers may purchase without the intervention.

Question 10

Which practice best addresses ethical risk in automated segmentation?

A. Collect every available customer attribute regardless of purpose.

B. Hide segment rules so customers cannot question decisions.

C. Use data minimization, fairness testing, transparency and human escalation.

D. Use demographic proxies whenever protected attributes are unavailable.

Correct Answer: C. Use data minimization, fairness testing, transparency and human escalation.