The research design specifies the procedures by which the information required for a decision is obtained.
02 / 27
The Research Design
A research design is the plan that specifies the procedures for collecting and analyzing the information needed to address the marketing research problem.
It follows directly from the problem definition and the research questions established in the first step of the research process.
It specifies the type of study, the information required, the measurement procedures, the sample, and the plan for analysis.
The choice of design determines which conclusions the findings can support.
03 / 27
01
Part 1: Types of Research Design
Research designs are classified by their purpose into three types: exploratory, descriptive, and causal.
Each type answers a different kind of question.
The type is selected according to what is already known about the problem and what the decision requires.
04 / 27
Exploratory Research
Exploratory research is conducted when the problem is not yet well understood.
Its purpose is to define the problem more precisely, to identify relevant variables, and to develop hypotheses.
Its methods include expert interviews, reviews of secondary data, pilot studies, and qualitative research.
Its procedures are flexible, and its samples are small and not representative of the population.
Its findings are therefore treated as tentative. They inform further research rather than support final decisions.
05 / 27
Descriptive Research
Descriptive research is conducted to describe the characteristics of a market or a group, such as the profile of customers, their usage behavior, or their perceptions of a brand.
It requires a clear statement of the problem and specific hypotheses before data collection begins.
The design specifies who is studied, what information is obtained, when and where the data are collected, and how.
Its methods include surveys, panels, and the analysis of existing records.
Descriptive research can establish that two variables are associated, but it cannot establish that one causes the other.
Stores
06 / 27
Cross-Sectional and Longitudinal Designs
Descriptive designs differ in how they treat time.
A cross-sectional design collects data from a sample of the population once.
A single cross-sectional design draws one sample. A multiple cross-sectional design draws two or more samples, each measured once.
A longitudinal design measures the same sample repeatedly over time.
Longitudinal designs are typically implemented through panels, in which the same households or individuals report their purchases or media use at regular intervals.
Cross-sectional designs describe a population at one point in time. Longitudinal designs reveal how individual behavior changes over time, such as switching between brands.
Exercise · Sort into groups
Sort each study by the way its design treats time.
Tap a card, then tap the group it belongs to.
07 / 27
Causal Research
Causal research is conducted to determine whether a change in one variable produces a change in another.
It addresses questions such as whether a price reduction increases sales, or whether a new package design increases purchase intention.
Its principal method is the experiment, in which the researcher manipulates one or more independent variables and measures their effect on a dependent variable.
The researcher controls other variables that could influence the outcome, so that the observed effect can be attributed to the manipulation.
Causal research requires the greatest prior knowledge of the problem and the greatest control over the research setting.
08 / 27
Selecting a Research Design
Exploratory research is appropriate when the problem is ambiguous and little is known about it.
Descriptive research is appropriate when the problem is defined and the decision requires an accurate description of a market or a group.
Causal research is appropriate when the decision depends on the effect of a specific marketing action.
The three designs are frequently combined in sequence: exploratory research generates hypotheses, descriptive research estimates the magnitude of a phenomenon, and causal research tests the effect of a proposed action.
The sequence is not fixed. A well-defined problem may begin with descriptive or causal research.
Exercise · Which one is it?
Each case describes a decision situation. Name the research design that is appropriate at this stage.
1 / 6
A publisher's newsletter subscriptions have stagnated, and no one in the firm can suggest why.
A car-sharing firm needs to know the age, income, and trip frequency of its current members before planning a campaign.
An online retailer must decide whether free shipping on orders above 40 euros would increase the average order value.
A hotel group wants to understand what business travellers mean when they describe a stay as 'restful', before it drafts any survey.
A beverage firm needs an estimate of the share of households in each region that buy sparkling water every month.
A cereal brand has two package designs and must choose the one that produces higher purchase intention.
A publisher's newsletter subscriptions have stagnated, and no one in the firm can suggest why.
The problem is ambiguous and little is known about it, so exploratory research is appropriate.
The problem is defined and the decision requires an accurate description of a group: descriptive research.
The decision depends on the effect of a specific marketing action, so causal research is appropriate.
The aim is to define the problem and develop hypotheses, which is the purpose of exploratory research.
Estimating the magnitude of a phenomenon in a market is the task of descriptive research.
Testing whether a package design changes purchase intention requires manipulating the design: causal research.
Which concept does this case show?
09 / 27
Discussion: Designing the Study
Discussion: A manufacturer of electric bicycles observes that inquiries from its website rarely result in purchases. Management has no explanation for this pattern. Which research design is appropriate at this stage, and how might the design change as the problem becomes better understood?
10 / 27
02
Part 2: Secondary Data
Data are classified as primary or secondary according to the purpose for which they were collected.
Secondary data are examined before any primary data are collected.
The analysis of secondary data is often called desk research.
11 / 27
Primary and Secondary Data
Primary data are collected by the researcher for the specific purpose of the current research problem.
Secondary data were collected previously, for a purpose other than the current problem.
01
Secondary data can be obtained more quickly and at lower cost than primary data.
02
Secondary data can clarify the problem, suggest hypotheses, inform the research design, and in some cases answer the research question without further data collection.
03
Because secondary data were collected for another purpose, they may be incomplete, outdated, or defined in ways that do not correspond to the current problem.
12 / 27
Evaluating Secondary Data
Secondary data are evaluated before they are used.
01
Methodology: how the data were collected, including the sample, the response rate, and the questionnaire.
02
Accuracy: the extent of error in the data, which is assessed by comparing several sources.
03
Currency: when the data were collected and whether conditions have changed since then.
04
Purpose: why the data were collected, since data collected to support a particular position may be biased.
05
Definitions: how key variables were defined and measured, such as the age groups used or the definition of a household.
06
Dependability: the expertise, credibility, and reputation of the source.
Exercise · Match up
Match each criterion for evaluating secondary data with the concern it addresses.
Criterion
Concern about a source
Tap a term, then the example it goes with.
13 / 27
Internal Secondary Data
Internal secondary data are generated within the firm in the course of its operations.
Sources include sales records, customer relationship management systems, loyalty program records, website and application analytics, and customer service logs.
Internal data are readily available, inexpensive, and specific to the firm's own customers.
Internal data describe only the firm's own customers, and therefore provide no information about competitors' customers or about people who have never purchased.
14 / 27
External Secondary Data: Published Sources
Published external data are made available by organizations outside the firm.
Government sources include censuses, labor statistics, and trade statistics.
Nongovernment sources include trade associations, industry reports, academic journals, and the financial reports of publicly listed companies.
Many government sources are available free of charge and are based on large samples, although they are often published with a delay.
15 / 27
External Secondary Data: Syndicated Services
Syndicated data are collected by commercial research firms according to a standard procedure and sold to multiple clients.
01
Consumer panels record the purchases or media use of the same households over time.
02
Retail tracking services record the sales of products through retail outlets, often from scanner data at the point of sale.
03
Media audience measurement services estimate the size and composition of the audience for television, radio, and digital media.
Syndicated data allow a firm to compare its performance with that of competitors, at a lower cost than collecting the same data independently.
Because the data are standardized, they cannot be adapted to the specific needs of a single client.
Exercise · Which one is it?
Each case describes a firm's information need. Name the syndicated service that meets it.
1 / 6
A yoghurt brand wants to know whether the same households that tried its new flavor bought it again in the following months.
A detergent manufacturer wants weekly unit sales of its brand and its competitors' brands across supermarket chains.
An advertiser wants to know how many viewers aged 25 to 44 watched a sports broadcast in which its commercial appeared.
A snack firm wants to compare the sales of its products in pharmacies and convenience stores after a competitor's price cut.
A publisher selling advertising needs an independent estimate of the audience for its digital video channel.
A coffee brand wants to see which brands its former buyers switched to over the past year.
A yoghurt brand wants to know whether the same households that tried its new flavor bought it again in the following months.
Repeat purchase by the same households over time is recorded by a consumer panel.
Sales through retail outlets, from point-of-sale scanner data, come from a retail tracking service.
The size and composition of a television audience are estimated by a media audience measurement service.
Product sales by type of retail outlet are recorded by a retail tracking service.
Audience size for digital media is estimated by a media audience measurement service.
Following the purchases of the same households over time is the function of a consumer panel.
Which concept does this case show?
16 / 27
03
Part 3: First-, Second-, and Third-Party Data
Customer data are also classified by their relationship to the firm that uses them.
The classification determines how reliable the data are, how specific they are to the firm's customers, and which legal obligations apply to their use.
17 / 27
Types of Customer Data
First-party data are collected by the firm directly from its own customers, such as purchase histories, account information, and interactions with its website.
Zero-party data are a subset of first-party data that customers provide intentionally, such as stated preferences in a profile or answers to a product quiz.
Second-party data are another organization's first-party data, obtained directly from that organization through a partnership, such as data shared between an airline and a hotel group.
Third-party data are collected by organizations that have no direct relationship with the consumers concerned, and are aggregated from many sources and sold to multiple buyers.
First-party data are generally the most accurate and specific to the firm. Third-party data offer the greatest scale but the least transparency about how they were collected.
Exercise · Which one is it?
Each case describes data used by a firm. Name the type of customer data.
1 / 6
An online bookshop analyzes the order histories of its registered customers.
A skincare brand's website asks visitors to complete a skin-type quiz, and uses their answers to recommend products.
A car rental firm receives booking data directly from a rail operator under a partnership agreement.
A furniture retailer buys a list of 'recent home movers' from a data broker that compiles it from many websites.
A supermarket records which products members scan with its loyalty app.
A streaming service's subscribers select their favorite genres in their profile settings.
An online bookshop analyzes the order histories of its registered customers.
The firm collected the data directly from its own customers: first-party data.
Customers provide the answers intentionally, so they are zero-party data, a subset of first-party data.
Another organization's first-party data, obtained directly through a partnership, are second-party data.
The broker has no direct relationship with the consumers and sells aggregated data to many buyers: third-party data.
The supermarket collects the records directly from its own customers: first-party data.
Stated preferences that customers enter intentionally in a profile are zero-party data.
Which concept does this case show?
18 / 27
The Decline of Third-Party Data
01
Privacy regulation now restricts the collection and use of personal data, as in the European Union's General Data Protection Regulation.
02
Several web browsers block third-party cookies by default, and mobile operating systems require user permission for tracking across applications.
As a result, the availability and reliability of third-party data have declined.
Firms have responded by investing in first-party data, for example through loyalty programs, customer accounts, and direct relationships with customers.
19 / 27
04
Part 4: AI-Assisted Desk Research
AI tools are now widely used to locate, summarize, and compare secondary sources.
They reduce the time required for desk research considerably.
They also introduce errors that are difficult to detect, because their output is fluent and appears authoritative.
20 / 27
Applications of AI in Desk Research
01
AI can be used to search for relevant reports, articles, and statistics.
02
AI can be used to summarize long documents, such as industry reports and annual reports.
03
AI can be used to extract figures and tables from documents and to compare them across sources.
04
AI can be used to translate sources published in other languages.
Tools that retrieve sources from the web and cite them in their answers make verification easier, but they do not ensure that the cited source supports the claim.
21 / 27
Fabricated References and Figures
Language models can generate references to reports and articles that do not exist, with plausible titles, authors, and publication years.
They can generate statistics that are not reported in any source, or attribute a real statistic to the wrong source, year, or population.
A summary can omit qualifications that the original source stated, such as the sample size or the definition of the market.
Such errors are most frequent for specialized topics, recent events, and precise numerical values.
22 / 27
Verifying AI-Assisted Desk Research
01
The researcher first locates the original source for each reference, independently of the AI tool.
02
The researcher then confirms that the source exists and that its authors, publisher, and date are as stated.
03
The researcher then confirms that the cited figure or claim appears in the source.
04
The researcher then examines the context of the figure, including its definitions, sample, and date.
05
Finally, the researcher evaluates the source against the criteria for secondary data: methodology, accuracy, currency, purpose, definitions, and dependability.
Internal secondary data are uploaded to an external AI tool only when the firm's data policies and the terms of the tool permit it.
Exercise · Put it in order
An AI tool has returned a market-size figure with a citation. Put the researcher's verification steps in order.
1
2
3
4
5
Tap a step, then the number where it belongs.
23 / 27
Discussion: Desk Research With AI
Discussion: An analyst preparing a market entry report asks an AI tool for the size of the plant-based dairy market in three countries. The tool returns three precise figures, each with a citation. Which steps should the analyst take before the figures are included in the report, and what should the report disclose about their origin?
24 / 27
05
Part 5: Research Designs for Studying AI Agents
The research designs introduced in Part 1 also apply when the subject of research is an AI agent rather than a consumer.
Two designs predominate: descriptive audits of agent outputs and causal experiments on agent choice.
25 / 27
Descriptive Audits of AI Agents
An audit systematically records the outputs of an AI agent across a large set of queries, such as the brands an assistant recommends in response to 500 product questions.
Audits describe what agents do: which brands appear, how often, and in which position.
Audits of AI shopping agents have documented position effects, in which an alternative's place in a list influences whether the agent selects it (Wadi & Ma, 2026a).
An audit observes only the final output. Because the queries differ in many respects at once, an audit cannot attribute a pattern in the outputs to a particular cause.
26 / 27
Causal Experiments on AI Agents
A causal experiment manipulates one factor while all other elements of the prompt, the products, and the environment are held constant.
For example, the same hotel listing is presented with and without a sponsorship label, and the agent's choices are compared across the two conditions.
The order of alternatives and attributes is randomized across sessions, to control for position effects.
Each condition can be repeated many times at low cost, and each session can be run independently of the others, so that no carryover occurs between conditions.
A descriptive audit identifies a pattern in agent behavior; a causal experiment tests whether a specific factor produces it.
Exercise · Sort into groups
Sort each study of AI agents by its research design.
Tap a card, then tap the group it belongs to.
27 / 27
Key Takeaways
01
The research design specifies how the information required by the marketing research problem is obtained and analyzed.
02
Exploratory research clarifies the problem, descriptive research describes a market or a group, and causal research tests the effect of a marketing action.
03
Descriptive designs are cross-sectional or longitudinal, depending on whether the same respondents are measured once or repeatedly.
04
Secondary data are examined before primary data are collected, and are evaluated for methodology, accuracy, currency, purpose, definitions, and dependability.
05
Secondary data are internal or external, and external data are published or syndicated.
06
Customer data are classified as first-, second-, or third-party, and the decline of third-party data has increased the value of first-party data.
07
AI tools accelerate desk research, but every reference and figure they provide must be verified against the original source.
08
Descriptive audits record what AI agents recommend and select, while causal experiments manipulate one factor at a time to establish why they do so.