Marketing attribution audience research: six populations

Marketing attribution audience research should start with a buying decision, not a fictional description of a typical marketer. Define which organisations could face the decision, which people can report on it, what evidence would change the next step and what the study cannot establish. Until real fieldwork is completed, every proposed segment and persona remains a hypothesis.

This guide covers research planned for organisations operating in England. An England office, a UK registered company and a respondent who happens to live in England are not interchangeable geography rules. UK guidance is labelled as such. Research was completed on 6 September 2026 by desk review; nobody was recruited, interviewed or surveyed.

What to take away

Start attribution audience research with a real buying decision, not a fictional persona.
Keep eligible organisations, participants, respondents, users, buyers and affected people as separate populations.
Record which role supplied each claim, because one person may hold several roles.
Treat every proposed segment and persona as a hypothesis until real fieldwork is completed.
Research participation and direct marketing permission are separate decisions that need separate records.

Keep six populations separate

Attribution work affects more people than the person who buys a reporting tool. A study needs an explicit relationship between these units.

Six populations, six evidence trails

Population

Eligible organisation: England, activity, recent decision
Research participant: Invited, screened, agreed
Respondent: Answers in the analysis
System user: Configures, checks, reads
Buyer or decision-maker: Specification, budget, approval
Affected person: Data or journey represented

Definition

Eligible organisation: Frame source, eligibility, geography
Research participant: Invitation, consent, status
Respondent: ID, role, completed items
System user: Task, permissions, interaction
Buyer or decision-maker: Stage, authority, approvers
Affected person: Relationship, effect, rights route

Evidence to retain

Eligible organisation
Research participant
Respondent
System user
Buyer or decision-maker
Affected person

Population	Practical definition	Evidence to retain
Eligible organisation	An organisation meeting the declared England, activity and recent-decision criteria	Frame source, eligibility result and geography field
Research participant	A person who was invited, screened and agreed to a specific research activity	Invitation route, information provided, consent record and participation status
Respondent	The participant whose answers appear in the analysis	Respondent ID, role evidence, completed items and missing answers
System user	A person who configures, checks or reads an attribution process	Recent task, permissions and actual system interaction
Buyer or decision-maker	A person with an evidenced role in specification, budget, approval or contract	Decision stage, authority boundary and other approvers
Affected person	Someone whose data, work or customer journey may be represented in the system	Data relationship, possible effect and route for questions or rights

One person may occupy several roles, but the researcher must record which one supplied each claim. A platform operator can explain a workflow and still be unable to speak for procurement. A budget holder may approve a contract without understanding event definitions. Website visitors whose identifiers feed a model are not research respondents unless they knowingly join the study.

The GOV.UK Service Manual introduction to user research urges government teams to understand all types of users rather than asking only what is popular. Its public-service setting is not a sample design for commercial buyers, but the distinction is useful: research the people doing, approving and experiencing the task.

Turn the buyer decision into a protocol

Write the decision in one sentence. For example: decide whether to retain a manual channel-credit process, procure a service or commission an incremental test for a defined campaign class. This wording avoids assuming software is the answer.

Make audience research useful for platform decisions

Where a platform decision is in scope, record the products and attribution workflows participants actually use or are considering. Google Analytics 4 and Adobe Analytics are analytics platforms; HubSpot and Salesforce include CRM and marketing products; Mixpanel and Amplitude focus on product analytics; Segment is a customer data platform. These are different product categories, not interchangeable alternatives, so use research to uncover requirements rather than claim that interviews establish which one performs best. Record whether those requirements concern models such as last-click, linear or data-driven attribution.

Then complete a research record before recruitment:

decision owner and decision deadline;
organisation population and England geography rule;
respondent roles needed to answer each question;
sampling frame, exclusions and known coverage gaps;
recruitment message, channels, screener and version;
field dates and planned contact attempts;
interview guide or questionnaire version and question order;
data fields, recording method, storage, access and deletion date;
incentives, supplier relationships and other conflicts;
analysis approach, uncertainty and stopping conditions.

The GOV.UK guidance on planning user research tells service teams to turn untested assumptions into research questions and match activities to the questions. A commercial attribution study can adopt that discipline without presenting government guidance as a legal standard for private research.

Define England eligibility from observable facts

A workable organisation criterion might require a contracting or operating site in England, a specified type of marketing decision during a recent period, and evidence that the respondent took part in it. Record the source for each field. Do not guess sector, buying authority or organisational size from a job title, postcode or online profile.

Is England eligibility observable?

Can you cite a source for each eligibility field?

Yes

record source and screen on recent task

do not guess from title, postcode or profile

Official business data helps explain frame coverage. The ONS UK Business: activity, size and location dataset uses the Inter-Departmental Business Register and provides regional enterprise and local-unit tables.

It gives no contact list of attribution buyers or demand evidence. If the study frame comes from existing customer records, a professional association or a recruitment panel, describe that source instead.

The person-level screener should ask about a recent task in neutral language. Establish whether the participant specified events, reconciled reports, interpreted a model, approved spend, handled privacy review or signed a contract. Do not infer authority from seniority alone. Keep screen failures, unreachable invitations and refusals in the denominator so the final report shows who was lost.

Recruit without disguising promotion as research

Research participation and direct marketing permission are separate decisions. The ICO's current guidance for identifying direct marketing says genuine market research is not direct marketing, but a message becomes direct marketing if it includes promotion or the exercise is intended to support later marketing to the people involved. Calling a sales approach a survey does not change its purpose.

Research or direct marketing?

Genuine market research

Message content: No promotion
Later intent: No later marketing
Consent: Participation decision
Labelling: Honest invitation

Direct marketing

Message content: Includes promotion
Later intent: Supports later marketing
Consent: Marketing permission
Labelling: Survey label does not change purpose

For the actual personal-data processing, a named UK data-protection reviewer must document purpose, lawful basis, transparency, minimisation, retention, recipients and rights. Ethical agreement to participate is not automatically the organisation's UK GDPR lawful basis, and a lawful basis does not remove the need for an honest invitation.

The GOV.UK participant-privacy guide recommends explaining what data will be collected and how it will be used, limiting recruitment data and deleting participant details when no longer needed. It addresses government service research. A private organisation must adapt the procedure to its own role and obtain qualified advice.

Record each incentive and when it was disclosed. An incentive may affect who responds and what they say. It must not be presented as evidence of enthusiasm for the proposed offer. Provide an accessible information route and a non-recorded alternative where appropriate rather than excluding people because one tool or session format creates a barrier.

Use interviews for process evidence

An interview can reveal how a recent attribution decision unfolded. It cannot estimate prevalence without a suitable quantitative design. Ask for a concrete event: the last time two reports disagreed, a model result was challenged, or a campaign needed an incremental answer.

The discussion guide should separate prompts from possible follow-ups. Record its version, pilot changes and question order. Ask what happened before asking for explanations or preferred solutions. Distinguish the participant's words, an observed artefact, the researcher's interpretation and an unresolved point in the notes.

The GOV.UK in-depth interview guide advises government researchers to recruit current or likely users, plan around research questions and obtain informed consent. For attribution work, eligibility should be proven through the recent task rather than a broad claim of interest.

No interview findings or quotations exist for this draft. A future report must state the sampling frame, invited, screened, eligible and completed counts; field dates; interviewer; guide version; incentive; recordings; withdrawals; role mix; England rule; and reasons the evidence may not transfer to other buyers.

Design surveys around the intended estimate

A survey is more than a web form. The target population, frame, selection procedure, invitation, questionnaire, response process, weighting and uncertainty determine what can be inferred.

The Government Analysis Function's questionnaire design guidance distinguishes a questionnaire from a survey and says data quality begins with design. The page is marked under review, so editors must recheck it before publication. Use it as methodological guidance, not a guarantee that a particular question works.

Ask one idea at a time and define technical terms. Randomise response choices only when the design calls for it and retain the randomisation record. Pilot the complete route with synthetic data and eligible reviewers. Freeze a numbered questionnaire before fieldwork; any later change creates a version break that should be reported.

The response table needs at least four denominators: frame records selected, invitations delivered, organisations eligible and usable responses for each item. Item non-response is not a negative answer. A convenience sample of newsletter readers or volunteers may reveal issues but cannot support an England prevalence estimate without a defensible inference design.

ONS describes sample design and estimation as covering frames, selection, weighting, non-response adjustment, outliers, variances and other quality measures. Those principles expose the work required before a survey percentage can be generalised. They do not supply a ready-made attribution sample.

Keep personas as evidence records

A persona should summarise a supported task pattern, not invent a named person. Give it an ID such as HYP-PROCUREMENT-01 and label it as a working hypothesis. Include the population definition, field dates, evidence IDs, number and type of sources, contradictory observations, missing groups, confidence rationale and review date.

Avoid photographs, personal names, invented quotations, demographic decoration. Protected characteristics may be relevant to inclusive recruitment or a genuine barrier, but never guess or use them as shorthand for behaviour.

The GOV.UK guide to learning about user needs says government user needs should rest on research evidence, not assumptions; apply that evidence principle to the commercial study's narrower scope.

An evidence card can separate four columns: observed event, participant statement, analyst interpretation and unknown. If several roles share a task, do not merge them merely to produce a neat profile. If findings conflict, preserve the conflict and design the next study around it.

Analyse without manufacturing certainty

Prepare an analysis plan before reading the results. For interviews, use coded observations tied to source IDs and have a second researcher examine material decisions. Report negative cases and themes that arose only after probing. A quotation, if later published with permission, illustrates one participant's statement; it does not establish frequency.

For surveys, calculate results against the correct item denominator. Publish missingness, response rate, frame coverage, weights and uncertainty. Compare respondent roles only when the groups were defined before analysis and the achieved sample permits it. Do not transform a stated intention to evaluate software into adoption or revenue.

The Office for Statistics Regulation's Code of Practice quality page calls for sources, methods, strengths, limitations and uncertainty to be explained. The code governs official statistics, not a private buyer study, but it supplies a sound disclosure standard.

Set stop and correction rules

Pause recruitment if the invitation becomes promotional, eligibility cannot be verified, consent materials differ from the collection route, participants face an unplanned risk, or the intended sample is no longer reachable. Stop analysis if the question version cannot be identified, respondent and organisation units were mixed, or raw evidence cannot be traced to a finding.

Retire a persona when later evidence contradicts its defining pattern or its population no longer matches the decision. Withdraw a survey estimate when a denominator or weighting error changes its meaning. Keep a correction log with owner, date, affected output and notification route.

The final decision record should say proceed, narrow, conduct more research, retain the existing process or stop. A study is useful when it removes an option as well as when it supports one. Before publication, named UK user-research, statistics, data-protection, PECR, accessibility, security and employment reviewers must examine their respective boundaries, and every source should be reopened.

Before you act

Write the buying decision in one sentence before recruitment.
Define the England eligibility rule from observable facts.
Record the source for every eligibility field.
Keep screen failures and refusals in the denominator.
Document purpose, lawful basis, retention and rights with a reviewer.
State sampling frame, counts and field dates in the final report.

Common questions

What counts as an eligible organisation for this research?

A workable criterion might require a contracting or operating site in England, a specified marketing decision during a recent period, and evidence the respondent took part in it. Record the source for each field. Do not guess sector, buying authority or organisational size from a job title, postcode or online profile.

Can an interview estimate how common an attribution problem is?

No. An interview can reveal how a recent attribution decision unfolded, but it cannot estimate prevalence without a suitable quantitative design. Ask for a concrete event, such as the last time two reports disagreed or a model result was challenged, and separate prompts from follow-ups.

When does a research invitation become direct marketing?

The ICO guidance says genuine market research is not direct marketing, but a message becomes direct marketing if it includes promotion or the exercise is intended to support later marketing to the people involved. Calling a sales approach a survey does not change its purpose.