Market Research, Audience Insight and Segmentation

Introduction

Advertising and media management depend on decisions made under uncertainty. Managers must determine whom to reach, what those people value, how they behave across media and markets, and whether communication activity has produced a meaningful response. Market research supplies evidence for these decisions, but evidence is not automatically insight. Insight emerges when reliable information is interpreted in relation to a managerial problem and converted into an actionable proposition.

This chapter examines the research process from problem definition to strategic application. It integrates qualitative and quantitative methods, audience measurement, customer journeys, personas, segmentation, targeting and insight development. It also evaluates sampling error, non-response, bias, data quality, privacy and the limitations of contemporary data systems, including artificial intelligence.

Learning Objectives

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

1. Distinguish market data, research findings, audience insight and strategic implication.
2. Select and combine qualitative and quantitative methods appropriately.
3. Explain sampling, validity, reliability and common sources of research bias.
4. Interpret audience measurement and campaign performance indicators.
5. Construct customer journeys and evidence-based personas.
6. Design actionable segmentation and targeting strategies.
7. Develop insights while recognising ethical, methodological and technological limitations.

1. From a Management Problem to an Insight Problem

Research should begin with a decision, not with a convenient dataset. A management problem may be expressed as “sales are declining,” whereas a research problem asks why the decline is occurring and what evidence would distinguish among possible explanations. Possible causes include reduced category relevance, a competitor’s stronger distribution, ineffective creative, changing media habits or price sensitivity.

A useful research brief identifies:

• the business decision to be supported;
• the population or market of interest;
• the information required;
• the proposed methods and their limitations;
• the timetable, budget and ethical safeguards; and
• the decisions that will follow from alternative findings.

The distinction among four concepts is important:

An insight is not merely an interesting demographic fact. It normally connects a human tension, a behavioural explanation and an opportunity for action. For example, “young professionals use food-delivery apps” is descriptive. “Young professionals order delivery not only for convenience but to protect scarce evening time; messages promising reduced decision effort may therefore be more persuasive” is closer to an actionable insight. The latter still requires validation.

2. Research Paradigms and Method Selection

Research choices reflect assumptions about knowledge. Positivist or post-positivist approaches generally seek measurable patterns, comparison and probabilistic inference. Interpretivist approaches seek meaning, context and the way consumers construct their experiences. Pragmatic research selects methods according to the decision rather than treating one paradigm as universally superior.

Qualitative research

Qualitative research is appropriate when the researcher needs to explore meanings, language, motivations, lived experience or an unfamiliar category. Common methods include:

• depth interviews, which allow detailed exploration of individual experience;
• focus groups, which reveal interaction, social norms and competing interpretations;
• ethnography and participant observation, which examine behaviour in context;
• mobile diaries, which capture experiences close to the time of occurrence;
• social listening and discourse analysis, which examine naturally occurring language, subject to platform and sampling limitations; and
• projective or elicitation techniques, used cautiously to help participants express difficult-to-articulate associations.

Qualitative samples are usually purposive rather than statistically representative. Their value lies in depth and conceptual development, not in estimating population percentages. A focus group can reveal that a product is associated with embarrassment, but it cannot establish that embarrassment affects a known proportion of the market.

Quantitative research

Quantitative research measures incidence, prevalence, relationships, differences and change. Methods include structured surveys, experiments, panels, transaction analysis, media measurement and controlled field tests. Quantitative research supports estimation only when the sampling and measurement processes justify it.

A survey question must operationalise an abstract construct. “Brand trust” may be measured through several items concerning reliability, honesty and confidence. Single-item measures can be useful, but multi-item scales often improve coverage of a construct. Researchers should define the response scale, reference period and population clearly. “How often do you stream video?” is weaker than “On how many of the past seven days did you stream video for at least 10 minutes?”

Mixed methods

Mixed methods are valuable when explanation and measurement are both required. Exploratory interviews may identify language and hypotheses; a survey may estimate their prevalence; follow-up interviews may explain an unexpected result. The methods should be integrated rather than simply placed side by side. Researchers should state which method has priority and how findings will be reconciled when they conflict.

3. Sampling, Measurement and Data Quality

The sampling process

A population is the complete group about which a conclusion is intended. A sampling frame is the operational list or mechanism from which units are selected. The sample is the subset actually studied. These are not necessarily identical. An online panel may exclude people with limited internet access; a customer database excludes non-customers; platform analytics may omit activity outside the platform.

Probability sampling gives units a known, non-zero chance of selection. Simple random, systematic, stratified and cluster sampling are examples. Non-probability methods include quota, purposive, convenience, snowball and opt-in sampling. Non-probability samples may be useful, especially for exploratory work, but demographic weighting does not automatically remove all selection bias.

For a simple random sample estimating a proportion, a commonly used approximate sample-size expression is:

Here, n₀ is the initial required sample size, z is the standard-normal critical value associated with the desired confidence level, p is the anticipated population proportion, and e is the tolerated margin of sampling error. The formula assumes a probability sample, a sufficiently large population, independent observations and a proportion estimated under a simple random design. If no prior estimate of p is available, p = 0.5 gives the largest required sample for a fixed z and e. The expression does not correct for non-response, design effects or measurement error.

For a 95% confidence level, z ≈ 1.96. If p = 0.5 and e = 0.05, then:

The result is rounded upward to 385 completed observations under the stated assumptions. A finite-population correction may be used when the sample is a substantial fraction of a known finite population:

where N is the population size. In practice, the planned number of invitations must be larger because not everyone responds and some responses are unusable.

Reliability, validity and representativeness

Reliability concerns consistency: would the procedure produce similar results under comparable conditions? Validity concerns whether the procedure measures the intended construct and supports the claimed conclusion. A scale can be reliable but invalid—for example, a consistently misunderstood question. Internal validity concerns whether an observed effect can reasonably be attributed to the intervention rather than confounding factors. External validity concerns transferability to other populations, settings or times.

Researchers should examine:

• Content validity: whether the measure covers relevant aspects of the construct.
• Construct validity: whether relationships with other measures behave as theory predicts.
• Criterion validity: whether results correspond to an appropriate external criterion.
• Ecological validity: whether the setting resembles real behaviour.
• Data completeness: the extent of missing, duplicated or truncated observations.
• Measurement invariance: whether a scale has the same meaning across groups.

Bias and error

Sampling error is random variation arising because a sample rather than the entire population is observed. It can be quantified under probability-sampling assumptions. Bias is systematic deviation and is not solved simply by increasing sample size. Important forms include coverage error, non-response bias, response or social-desirability bias, recall error, acquiescence bias, interviewer effects, survivorship bias, algorithmic selection bias and confirmation bias in interpretation.

Weighting can improve alignment with known population margins, but it may increase variance and cannot repair unmeasured differences. Researchers should report weighting variables, response rates where meaningful, exclusions, fieldwork dates, questionnaire wording, missing-data treatment and uncertainty. A statistically precise estimate of a poorly measured construct remains misleading.

4. Audience Measurement and Media Metrics

Audience measurement translates exposure opportunities into indicators used for planning and evaluation. Definitions vary by medium and vendor, so metrics must be interpreted with their measurement rules.

• Reach is the number or percentage of distinct people exposed at least once during a defined period.
• Frequency is the average number of exposures among reached people.
• Impressions are exposure opportunities or recorded deliveries, depending on the system.
• Rating is commonly the percentage of a defined target population represented by an audience or exposure.
• Gross rating points (GRPs) are commonly calculated as reach percentage multiplied by average frequency.
• Cost per thousand (CPM) is cost divided by impressions, multiplied by 1,000.
• Click-through rate (CTR) is clicks divided by measured impressions, multiplied by 100.
• Conversion rate is defined conversions divided by the relevant denominator, such as clicks or sessions, multiplied by 100.

For example, a campaign reaching 40% of a target audience at an average frequency of 3 produces:

This does not mean that every individual received three exposures. It is an aggregate planning measure. Reach and frequency distributions are more informative when available.

If media cost is 18,000 currency units and recorded impressions are 600,000:

The unit is currency units per thousand recorded impressions. CPM does not indicate attention, comprehension, brand effect or sales efficiency by itself.

Measurement systems may differ in their treatment of viewability, invalid traffic, co-viewing, cross-device identity, duplicated users, walled gardens and privacy restrictions. “Impression,” “reach” and “view” should therefore be accompanied by a source, definition and measurement window.

5. Customer Journeys and Touchpoint Analysis

A customer journey describes how a person moves from a situation or need through consideration, choice, use and post-purchase response. It is not always linear. Consumers may enter through a recommendation, compare prices after seeing an advertisement, abandon a basket, return through search and advocate after use.

A practical journey map identifies, for each stage:

1. the customer goal and context;
2. actions and questions;
3. emotions and perceived barriers;
4. touchpoints and channels;
5. brand or service experience;
6. evidence supporting the interpretation; and
7. an opportunity and responsible owner.

Journey mapping becomes weak when it is based solely on internal assumptions. Evidence may include interviews, search behaviour, service logs, web analytics, path analysis, usability tests and transaction data. Analysts should distinguish an observed sequence from an inferred motivation. A person who visits a pricing page may be comparing alternatives, seeking reassurance or merely looking for a forgotten detail.

Attribution models should also be treated cautiously. Last-click attribution is simple but assigns excessive credit to the final recorded interaction. Multi-touch models distribute credit according to specified rules, while experiments and geo-based tests can provide stronger evidence of incrementality. Correlation between exposure and purchase does not prove that exposure caused the purchase because motivated customers may be more likely to seek advertising.

6. Personas: Useful Abstraction, Dangerous Stereotype

A persona is a research-based representation of a significant user or buyer pattern. It should include a behavioural situation, goals, barriers, information needs, media habits and evidence. Demographics may provide context but should not substitute for behaviour or motivation.

A robust persona record includes:

• a descriptive name that does not imply a stereotype;
• relevant context and task;
• category behaviour and decision criteria;
• tensions, anxieties and desired outcomes;
• media and information practices;
• accessibility or service requirements;
• representative quotations, clearly labelled as quotations; and
• source, date and confidence level.

Personas should be treated as hypotheses that guide design and communication, not as fictional individuals with predictive certainty. They require updating when behaviour, technology or category conditions change. Multiple personas should not be created merely to make a strategy appear sophisticated.

7. Segmentation, Targeting and Positioning

Segmentation divides a heterogeneous market into groups that are sufficiently similar for a particular marketing decision. Targeting evaluates those groups and selects one or more for investment. Positioning defines the intended meaning and competitive frame in the target’s mind.

Common segmentation bases include:

A segment should be identifiable, substantial or strategically valuable, reachable, differentiable, actionable and reasonably stable for the planning period. These criteria are managerial, not natural laws. A small segment may be valuable if it has high margin or strong influence; a large segment may be unattractive if it is unreachable or poorly served.

Segmentation can be designed through theory-led categories, cluster analysis, latent-class approaches or decision trees. Statistical clustering does not discover objectively true groups; results depend on variables, scaling, distance measures, algorithm settings and the chosen number of clusters. A segment solution should be assessed for stability, interpretability, size, response differences and practical usability. Cross-validation or replication with another sample is preferable to accepting the first attractive solution.

Targeting may be undifferentiated, differentiated, concentrated or customised. The choice depends on resources, category structure, brand distinctiveness, heterogeneity and the economics of serving groups. Ethical evaluation is essential when targeting concerns children, health, financial vulnerability, political persuasion or inferred sensitive characteristics.

8. Developing and Testing Insights

Insight development is an abductive process: researchers move between evidence, possible explanations and implications. A disciplined workflow is:

1. Define the decision and audience.
2. Assemble evidence from more than one relevant source.
3. Identify a recurring behaviour, contradiction or unmet need.
4. Ask why the pattern may exist and what competing explanations are plausible.
5. Formulate an insight statement linking context, tension, motivation and opportunity.
6. Check whether the statement is distinctive, humanly meaningful and actionable.
7. Test it through appropriate qualitative, quantitative or experimental evidence.
8. Translate it into proposition, creative direction, channel role and measurement plan.

A useful insight template is: “When [context], people want [desired outcome], but [barrier or tension], because [underlying explanation]. Therefore, the brand can [credible role].” This is a thinking aid rather than a formula. The proposed brand role must be supported by capability and evidence.

Triangulation strengthens interpretation when different methods converge, but disagreement is also informative. A survey may show high stated interest while observed behaviour remains low because of price, friction or social-desirability effects. The answer is not to choose the preferred result; it is to investigate the discrepancy.

9. Applied Case: Repositioning a Ready-to-Eat Breakfast Brand

The following is an illustrative case, constructed for teaching purposes.

Background

A national ready-to-eat breakfast brand has stable distribution but declining repeat purchase among urban adults aged 25–40. The brand team initially proposes targeting “millennials” through social video.

Problem

The proposed audience is broad and demographic. Management needs to know whether the issue is awareness, relevance, taste, price, availability or changing morning routines, and which occasions offer a credible growth opportunity.

Analysis

The research team begins with 20 depth interviews and four mobile diary groups. Participants frequently describe mornings as “compressed” rather than simply “busy.” They want breakfast that reduces decisions and mess, but they also fear that ready-to-eat products signal low care or poor nutrition. A survey of 800 category buyers measures occasion, barriers, brand perceptions, purchase frequency and media behaviour. The sample is quota-controlled and weighted to known age, gender and region margins; the report states that weighting does not guarantee full representativeness.

Behavioural and needs-based analysis identifies three provisional groups: routine planners, rushed improvised eaters and health-oriented preparers. The second group is not necessarily young, and the groups overlap in media use. A journey review finds that awareness is adequate, but the product page and retail shelf do not answer the questions “Will it keep me full?” and “Will it fit my morning?”

Approach and solution

The team selects rushed improvised eaters as a priority growth segment because the group reports a frequent unmet need and can be reached in relevant morning and commuting contexts. It does not claim that all people in the segment share one identity. The positioning is reframed from “fast breakfast” to “a dependable start when the morning changes.” Creative demonstrations show portability, satiety information and flexible consumption occasions. Search, retail media, audio and short-form video receive different roles rather than identical executions.

A controlled regional test compares markets with the revised campaign and markets receiving the previous communication, while distribution and price changes are monitored. Primary outcomes include incremental repeat purchase and brand association, not merely video completion.

Outcome

The test shows improved association with reliability and a modest increase in repeat purchase in exposed markets, but the effect is smaller among households with children. Follow-up interviews reveal that household preferences and preparation time remain important. Management therefore avoids claiming a universal repositioning and develops a separate family-oriented communication route.

Learning points

The case demonstrates that demographic labels can conceal occasions and tensions; qualitative research can explain quantitative patterns; segments should be linked to an opportunity; and campaign success should be tested against business and brand outcomes. It also illustrates that an insight may be useful without being universally true.

10. Ethics, Privacy and Governance

Research and audience management involve responsibilities beyond legal compliance. Consent should be informed, specific where required and proportionate to the activity. Participants should understand collection, use, retention and sharing practices. Researchers should minimise data, protect access, document processing and provide appropriate withdrawal or rights mechanisms under applicable data-protection law.

Special care is needed with children, health information, financial data, precise location, political preferences and inferred vulnerability. Combining datasets can create sensitive profiles even when individual variables appear harmless. Pseudonymisation reduces direct identifiability but does not make data fully anonymous if re-identification remains possible.

AI-assisted transcription, sentiment analysis, synthetic audiences and predictive targeting can improve speed but introduce risks. Training data may underrepresent groups; language models may misinterpret dialect and irony; automated classifications may reproduce historical discrimination; and opaque scores can be treated as facts. Governance should include documented purposes, human review, bias testing, access controls, audit trails, retention limits and a route for contesting consequential decisions.

Research ethics also includes respectful questioning, fair incentives, avoiding deception except where justified, secure handling of recordings and honest reporting of inconvenient findings. Fabricating precision or hiding limitations is a professional failure even when no law is breached.

11. Contemporary Developments and Best Practice

The fragmentation of media has made cross-platform measurement more difficult. Connected television, retail media, creator platforms, mobile environments and commerce data offer new signals but also increase duplication and governance challenges. Privacy-preserving measurement, clean rooms, contextual targeting and first-party relationship strategies are developing responses, although each has technical and commercial limitations.

Best practice includes:

• begin with a decision and a falsifiable research question;
• combine methods according to the uncertainty being addressed;
• separate descriptive data from causal claims;
• document population, frame, sample, fieldwork and weighting;
• pretest questionnaires and instruments;
• distinguish reach from attention and exposure from effect;
• use behavioural evidence alongside stated attitudes;
• treat personas and segments as revisable models;
• test stability and actionability of segmentation solutions;
• disclose uncertainty, missingness and competing explanations;
• protect privacy through minimisation and purpose limitation; and
• measure incremental business, brand or social outcomes rather than vanity metrics alone.

Challenges and Limitations

No research programme eliminates uncertainty. Qualitative findings may be context-specific; surveys may suffer from poor recall or response bias; panels may not represent non-users; platform metrics may change definitions; and transaction data may show what happened without explaining why. Segments can become stereotypes, attribution can confuse association with causation, and dashboards can create an illusion of objectivity.

Managers should therefore ask: What exactly was measured? For whom? Under what conditions? Compared with what? Which alternatives remain plausible? What decision would change if the finding were wrong? These questions convert methodological caution into better management rather than paralysis.

Chapter Summary

Market research supports advertising and media decisions when it begins with a clearly defined management problem. Qualitative methods explain meanings, contexts and motivations; quantitative methods estimate patterns and test relationships; mixed methods connect explanation with measurement. Sampling quality, validity, reliability, non-response, coverage and measurement bias determine how far findings can be trusted.

Audience metrics such as reach, frequency, GRPs, CPM, CTR and conversion rate require precise definitions and should not be confused with attention, persuasion or incrementality. Customer journeys organise evidence across contexts and touchpoints, while personas provide carefully bounded representations of behavioural patterns. Segmentation should be identifiable, reachable, differentiable and actionable, and targeting must consider both commercial value and ethical risk.

An insight links evidence to a human tension and a credible brand opportunity. It must be tested, not merely celebrated. Privacy, fairness, transparency and governance are integral to professional research, especially as AI and fragmented media systems expand the scale of data collection.

Key Terms

Market research: Systematic collection and analysis of information for a marketing decision.

Audience insight: An evidence-based explanation of an audience’s meaningful behaviour or tension.

Qualitative research: Research focused on meanings, experiences, language and context.

Quantitative research: Research that measures variables and estimates patterns numerically.

Mixed methods: Integration of qualitative and quantitative approaches in one research programme.

Population: The complete group to which a research conclusion is intended to apply.

Sampling frame: The operational source from which a sample is selected.

Sampling error: Random variation resulting from studying a sample rather than a census.

Bias: Systematic deviation from the intended population value or interpretation.

Reliability: Consistency of a measurement or procedure.

Validity: The extent to which evidence supports the intended measurement or inference.

Reach: Distinct people exposed at least once during a defined period.

Frequency: Average number of exposures among reached people.

Customer journey: The sequence of contexts, actions and experiences associated with a customer decision.

Persona: A research-based representation of a meaningful user or buyer pattern.

Segmentation: Division of a market into strategically meaningful groups.

Targeting: Selection of segments for priority marketing attention and investment.

Positioning: Intended meaning of a brand relative to alternatives.

Triangulation: Comparison of evidence from multiple methods or sources.

Incrementality: The additional outcome caused by an intervention beyond what would otherwise occur.

Review Questions

Short Answer Questions

1. Distinguish data, a research finding, an audience insight and a strategic implication.
2. Why can a qualitative focus group not establish the percentage of a population holding an opinion?
3. Define reach, frequency and GRPs.
4. What is the difference between reliability and validity?
5. Identify three common sources of bias in audience research.

Long Answer Questions

1. Compare qualitative, quantitative and mixed-method research for investigating declining brand preference.
2. Explain how sampling frame, non-response and weighting affect the credibility of a survey.
3. Discuss the role and limitations of customer journey mapping in media planning.
4. Explain the criteria for a strategically useful market segment and evaluate demographic segmentation against needs-based segmentation.
5. Discuss ethical and governance requirements for AI-assisted audience targeting.

Analytical / Application Questions

1. A campaign reaches 55% of a target audience at an average frequency of 2.4. Calculate its GRPs and explain what the result does and does not show.
2. Design a mixed-method research programme for a streaming service experiencing high trial but low renewal. Specify the research questions, methods, sampling logic and key limitations.
3. A survey reports high stated purchase intention, but transaction data show no increase in sales after advertising. Develop at least three plausible explanations and propose a method for distinguishing among them.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes an audience insight from a demographic description?

A. An insight reports the average age of the audience.

B. An insight explains a meaningful behaviour or tension and suggests an opportunity for action.

C. An insight lists the channels used by the audience.

D. An insight gives the total number of impressions delivered.

Correct Answer: B. An insight explains a meaningful behaviour or tension and suggests an opportunity for action.

Question 2

Which method is most appropriate for exploring consumers’ unarticulated feelings about a sensitive service experience?

A. A large reach report

B. A depth interview

C. A CPM calculation

D. A frequency distribution only

Correct Answer: B. A depth interview

Question 3

Under the simple random-sample assumptions stated in the chapter, what is the approximate initial sample size for 95% confidence, p = 0.5 and a 5% margin of error?

A. 96

B. 196

C. 385

D. 1,000

Correct Answer: C. 385

Question 4

A campaign reaches 40% of a target audience at an average frequency of 3. What are its GRPs?

A. 13.3

B. 43

C. 120

D. 400

Correct Answer: C. 120

Question 5

Which statement about weighting is most accurate?

A. Weighting eliminates all forms of bias.

B. Weighting can align selected sample margins with known population margins but cannot repair every unmeasured difference.

C. Weighting converts a convenience sample into a simple random sample.

D. Weighting removes measurement error from poorly worded questions.

Correct Answer: B. Weighting can align selected sample margins with known population margins but cannot repair every unmeasured difference.

Question 6

Which criterion is least relevant when evaluating whether a segment is actionable?

A. Whether the segment can be reached effectively

B. Whether the segment differs in a strategically useful way

C. Whether the segment has a decorative name

D. Whether the organisation can serve the segment

Correct Answer: C. Whether the segment has a decorative name

Question 7

Why is last-click attribution potentially misleading?

A. It always ignores the final interaction.

B. It may assign excessive credit to the final recorded interaction and overlook earlier influences.

C. It measures only qualitative interviews.

D. It cannot be calculated from digital data.

Correct Answer: B. It may assign excessive credit to the final recorded interaction and overlook earlier influences.

Question 8

Which statement best describes a persona?

A. A guaranteed prediction of every individual’s behaviour

B. A research-based representation of a meaningful user or buyer pattern

C. A legal identity record for advertising delivery

D. A synonym for a census

Correct Answer: B. A research-based representation of a meaningful user or buyer pattern

Question 9

If media cost is 18,000 currency units and impressions are 600,000, what is the CPM?

A. 3

B. 30

C. 300

D. 33,333

Correct Answer: B. 30

Question 10

Which practice most directly supports responsible AI-based audience targeting?

A. Using every available variable regardless of sensitivity

B. Suppressing documentation to protect the algorithm

C. Combining purpose limitation, bias testing, human review and audit trails

D. Treating predictive scores as verified facts

Correct Answer: C. Combining purpose limitation, bias testing, human review and audit trails