Audience research
A protocol for planning audience research for a paid social agency offer, from populations to decisions
A research ethics and methods protocol for audience research for an England paid social agency offer, from populations and sampling to consent and decisions.
| Case | Unit to recruit | Useful evidence | Does not establish |
|---|---|---|---|
| Commissioning organisations | Enterprise, public body or charity | Contracting conditions and service fit | What every individual buyer thinks |
| Budget and approval roles | Named roles within organisations | Authority, approval steps and risk concerns | Day-to-day account operation |
| Platform operators | People configuring accounts and permissions | Operational friction and hand-offs | Purchasing authority |
| Creative and measurement users | People producing assets or reconciling reports | Workflows, use and failure modes | Consumer response |
| People exposed to adverts | People in the defined consumer population | Comprehension, recognition, accessibility and response | Agency buying behaviour |
| Research participants | Identifiable people invited to take part | Evidence supplied under an agreed research relationship | Permission for promotion or sales contact |
Commissioning organisations
Commissioning organisations are the contracting-side population, but the right unit depends on the decision: it may be a head office, procurement body or local operating site. Decide whether the question concerns potential contracting, service fit or buying conditions before treating enterprises as interchangeable cases.
Use a population register to keep organisations distinct from the people who work in them. A marketing manager, finance approver and legal reviewer within one organisation may have different evidence about the same proposed service, so count organisations and respondents separately.
Budget and approval roles
Budget and approval roles can authorise expenditure, claims or risk decisions, but their responsibilities may not overlap. Recruit the roles needed to answer the specific question, rather than assuming one marketing contact can speak for finance, procurement and legal review. Treat each account of a decision as role-specific evidence.
For example, a budget approver can describe the last agency commitment and who signed it, while a legal reviewer may explain which claim needed approval. Record which role supplied each observation. A broad competitor research finding cannot substitute for evidence about an individual organisation's approval route.
Platform operators
Platform operators configure accounts, permissions, billing and day-to-day delivery. Their descriptions are valuable evidence about access problems and working practices, but they do not demonstrate that the operator can commission an agency or approve a budget. Keep operational testimony separate from buying authority in both recruitment and analysis.
Ask about the most recent relevant change or task, such as a permission update, and record what happened. Where a buyer and operator describe the same process differently, preserve both accounts rather than smoothing the disagreement into a single organisational view.
Creative and measurement users
Creative and measurement users can explain how assets are produced, reports interpreted and results reconciled. Their accounts can identify failure modes and hand-offs, but they should not answer on behalf of people who see advertising.
Keep the question tied to the work they performed, not an assumed view of the whole campaign.
A measurement user may describe which records were reconciled for the last report, while a creative user may explain the production task. Record the work context and evidence source for each response so that one user's experience is not presented as another population's finding.
People who see adverts
People exposed to advertising can discuss comprehension, recognition, accessibility and response. They are not agency buyers merely because they use a social service.
Ofcom's Online Nation 2025 provides UK-wide evidence about online services and user behaviour, but it does not identify English recipients of a particular campaign or establish an England buying population.
Define consumer eligibility directly for the research question and keep it separate from the organisation and role samples. Do not turn UK-wide user evidence into an England-specific estimate, or assume that platform use means a person has seen the advertising being studied.
Identifiable participants
Research participants are the identifiable people invited to take part, rather than another label for buyers or consumers. Participation creates a research relationship with its own information, privacy and withdrawal requirements. It does not by itself permit promotional quotation, testimonial use or later sales contact.
An invitation should name the researcher and commissioning body, explain the subject and duration, describe recording and data use, state confidentiality limits and give a withdrawal route and accessibility contact. Explain why the person was approached and whether their employer will learn who took part.
Tie each study to a choice
Begin with one choice that a named owner must make. Examples include whether to offer a client-owned managed service, which hand-off causes approval delays or whether a buyer can support a proposed measurement design. Avoid an open-ended brief such as understanding paid social.
Falsifiable decision statement
Can four of five teams document ownership?
Proceed with three-month English retail pilot
Do not proceed; revise offer or evidence
For each question, specify the decision, evidence needed, population, method and stopping rule. The Government Analysis Function's questionnaire design guidance starts from data requirements and distinguishes questionnaire design from the wider survey method. It is under review, so retain its status and publication date when adapting its recommendations.
A falsifiable planning rule could be to proceed with a three-month English retail pilot only if four of five eligible commissioning teams can document account ownership, budget authority and a measurement owner. That threshold is not a market statistic. Change it only before fieldwork, or disclose the change and its timing.
Fix the frame before recruitment
Choose an England geography that fits the decision: head-office location may suit budget research, an English local unit may suit regional delivery, and residence in England may define consumer research. Record the selected rule on every sample row, along with the unit being counted.
The ONS UK business activity dataset gives registered enterprises and local units by English region, industry and size, and explains that the units differ. The DBT Business Population Estimates 2025 adds estimates of unregistered private businesses and uses head-office geography. Neither source identifies agency buyers, so do not treat either as a buyer list.
Record the frame source edition and extraction date, population unit and geography, sector and size eligibility, duplicates, missing contacts and other exclusions. Also log numbers invited, eligible, started and completed, alongside contact attempts and fieldwork dates. These records make coverage and non-response visible.
For role-based work, sample organisations first, then recruit the specified roles within them. Otherwise, one large employer with several respondents can outweigh several smaller accounts. Keep organisation and individual respondent denominators distinct in sampling, analysis and reporting.
Match the instrument to the claim
A survey can estimate prevalence only for a defined frame and a disclosed method. A convenience link shared on social media has an unknown invitation denominator and self-selection bias. It can collect exploratory comments, but it cannot support an England percentage.
Use a question-to-decision map. The Analysis Function guidance recommends inclusive design, short and unambiguous questions, cognitive testing and a limited quantitative pilot before launch. Specify recall periods, use if any where zero is valid, and avoid combining creative, media and measurement in one response item.
Report each proportion as:
reported proportion = eligible respondents giving the response / eligible respondents shown the question
State missing answers, routing, duplicates, partial completions, weighting and field dates. Confidence intervals do not correct coverage or non-response bias. Where recruitment is purposive or quota-based, call the result sample evidence, not a population estimate.
Keep invitation, consent and contact distinct
Agreement to take part, a UK GDPR lawful basis, promotional reuse and permission for marketing contact are separate choices. Government Service Manual guidance on informed consent for user research says participants should understand the research and agree to take part, with consent revisited during a session where needed.
Government Social Research ethical assurance guidance is written for government work, but offers principles on sound conduct, informed participation, access and minimising harm. These sources do not determine the lawful basis for a private agency project.
Document the lawful basis and any additional condition, then have a named UK privacy professional and research-ethics reviewer approve the actual plan before invitations go out. Keep agreement to participate and be recorded, promotional reuse permission, and later marketing permission distinct from that legal assessment.
Where relevant, make four choices explicit:
- Agreement to take part and be recorded.
- The documented UK GDPR lawful basis and any additional condition.
- Permission to reuse a quotation, image or account name for promotion.
- Permission for later marketing contact through the proposed channel.
Refusing one choice should not quietly cancel or grant another. ICO valid-consent guidance distinguishes research participation from UK GDPR consent as a lawful basis. ICO electronic-mail marketing guidance says consent must be specific and people must be able to change their mind; PECR and data-protection analysis for later contact remains separate.
Ask about work already done
Use customer interview probes anchored to the most recent completed or attempted task. Ask an approver to describe the last agency commitment, an operator to walk through the last permission change, and a measurement user to identify the records reconciled for the last report.
Do not request passwords, unpublished audience lists, raw customer files or unnecessary commercial terms. If screen sharing is useful, plan redaction and stop recording before credentials or unrelated personal data appear. The session lead, note-taker and security contact should know how to halt the session.
Probe contradictions without arguing. Ask what happened next or which record would show it, and capture missing evidence and non-use as well as smooth accounts. A claimed standard process is not observed practice until a permitted record or demonstration supports it.
Keep summaries tied to coded evidence
A persona can summarise a recurring decision context when coded records support it. Do not add a name, portrait, quotation, budget or motivation that participants did not supply. Separate observed facts from interpretation and unanswered questions, so a useful shorthand does not turn into a fictional account.
An evidence summary should state the population and English geography, supporting case count, observation dates, common task, variation, contradictory cases, missing roles and decision implication. Link it to de-identified notes or coded survey fields under controlled access, rather than circulating identifiable source material.
Where coding requires judgement, have more than one analyst review it. Keep a codebook, log changes and explain how disagreements were resolved. Small qualitative samples can reveal mechanisms and vocabulary, but they cannot establish national prevalence.
Test tools without exposing live records
Choose research software only after checking service scope, contract and data terms, storage geography, security material, accessibility evidence, ownership, export and deletion. Supplier pages are first-party claims, not independent assurance, so product convenience should not decide the method or ethical safeguards.
Use synthetic records for acceptance testing. Test an invitation, consent route, withdrawal, branching, screen-reader path, small-screen completion, export, restricted analyst access and deletion. Do not upload live participant or advertising-account data just to compare products.
The NCSC supplier assurance questions offer a UK framework for examining governance, data protection, incidents and contractual responsibilities. Completing a questionnaire is not proof that a specific deployment is safe; assess the intended configuration and access arrangements separately.
Separate research records from campaign operations
Keep participant contact details, recordings and transcripts out of advertising account notes and sales systems by default. Limit access by role, encrypt transfers, log exports and set deletion dates. Credentials and recovery codes do not belong in research recordings or notes.
If an agency processes personal data on a client's instructions, the ICO's current controller and processor contract guidance sets out required contractual topics and is marked under review following the Data (Use and Access) Act. Determine the parties' roles from their actual decisions, not from the label agency.
Promotional reuse needs an advertising review as well. CAP Code Section 2 requires marketing communications to be recognisable, while the CMA's reviews and endorsements collection addresses genuine and transparent commercial material. Neither source turns research participation into permission to publish a testimonial.
Close on evidence and limitations
End fieldwork on the recorded date or when the predetermined saturation or sample rule is met. Preserve refusals, non-response and missing population groups, and compare evidence by role rather than averaging incompatible answers. Record the limits before a decision owner sees the recommendation.
A measurement report should show the question, frame, recruitment, participant counts, exclusions, raw denominator, method, coded themes, uncertainty, dissenting cases, privacy and ethics sign-off, and action owner. Set an update trigger, such as a platform access change, a new buyer role or a finding that fails in the pilot.
Accessibility, privacy, security, advertising, platform and commercial approval remain independent gates. A promising interview cannot compensate for a failed data route. Assign the writer, fact-checker, UK privacy and PECR professional, research-ethics reviewer, statistician, security reviewer, accessibility specialist and advertising reviewer before publication.
In this guide
- Defining a paid social agency target audience by organisation and decision roleDefine a paid social agency target audience in England by organisation, decision role and research population, with explicit sampling limits.
- Paid social agencies buyer personas: coded records to cardsCode records into one evidence card for a paid social buyer persona: card fields, a hypothetical worked card, second-analyst checks and expiry triggers.
- Follow-up probes for paid social customer interviews, task by taskFollow-up probes for paid social customer interviews: what to ask after each opening question, how to weigh one answer, and when to pause the recording.
- Researching rival paid social agencies with a dated evidence registerA dated evidence register for rival paid social agencies, separating identity, claims, public adverts, reviews, authorised controls and unknowns.
