Audience research for creative software teams, from the decision to be made to defensible personas

Start with the decision that could change, then follow the branch that matches the uncertainty. Each route connects a research choice to what the evidence can support. If the result exposes a gap, treat that gap as the next question, rather than smoothing it over with a confident-sounding persona or market claim.

If a polished persona is already doing the rounds

If a persona looks settled before anyone has tested its claims, reopen it as a set of hypotheses. Labels such as maker, commissioner, approver and administrator can help organise recruitment, but invented biographies, names or quotations do not make assumptions into evidence.

For each proposed need, connect it to dated notes, indicate how many relevant participants described it, and retain examples that do not fit. Keep the claim conditional if evidence is thin, and remove a pattern that does not affect a real design or buying decision. Persona evidence should make uncertainty visible rather than disguise it with a finished profile.

Rule of thumb: Keep a persona only if it can help change a decision.

If nobody can say what research might change

If the team cannot name a decision that could move, pause before selecting a method. Work out whether the uncertainty is about a workflow, an audience, a buying obstacle or whether to continue at all.

A question about lost approval context during a handoff gives research a clearer purpose than asking whether people want easier software.

Write down the intended decision, current belief, people closest to it, evidence required and result that would stop or narrow the work. Use strategy failure checks to spot a break in the reasoning. Consider software selection only when choosing software is the decision the evidence needs to inform.

Rule of thumb: State the decision before settling on the research method.

If the uncertainty is about frequency

If earlier work has established the behaviours and answer options, a survey can help describe how often a behaviour appears within a defined recruited sample. It is not a sound discovery tool when the team has guessed the answer space before understanding an unfamiliar workflow.

Keep questions focused on one idea, allow responses such as not applicable or cannot disclose, and test an accessible form on relevant devices and assistive technologies. Compare survey tools with the research and data plan, then check current controls and account terms before fieldwork.

Rule of thumb: Measure frequency only when the possible answers are understood.

If several roles touch the same task

If authority is split across a workflow, recruit people according to recent actions and the decisions they were entitled to make.

A job title is an unreliable substitute: the same label can mean different things across organisations. Ask what each person did and when, rather than treating their role name as evidence.

Include contrasts that could change the account, such as occasional versus high-volume production or fixed versus flexible tool environments. Ask who else took part and whether the process can be discussed without restricted material. Team roles can suggest useful distinctions, but do not turn them into permanent audience categories. Qualitative recruitment should reveal differences, not imitate a statistical sample.

Rule of thumb: Recruit by recent actions and decision authority, not title.

If a public number appears to describe the market

If a public figure is being used to frame recruitment or a market claim, establish its coverage and definition first. Record the source, geography and limitations alongside the figure. Business population estimates can provide context, but they do not establish who uses creative software.

The ONS bulletin UK business: activity, size and location counts registered enterprises and local units by industry and geography, including English regions, using VAT and PAYE registrations; some very small activity is absent, and software use is not measured. The IPA Agency Census 2025 describes participating member agencies across the UK, while Ofcom's Online Nation research concerns people's use of online services, not creative teams' software purchasing.

Rule of thumb: Use public figures to show boundaries, not to invent totals.

If supplier claims are standing in for user evidence

If competitor research is being used to explain customer needs, separate what a supplier says it offers from what participants report experiencing.

Supplier material can document an offer, but it cannot prove user needs or product outcomes. Advertising standards are a different question from evidence about a workflow or buying decision.

For an identity check, use a Companies House company record, then connect the organisation to a product through its own legal or privacy record. Note the product page, access date, edition, intended customer and documented capability. Mark unsupported claims as unverified instead of scoring them as proven, and keep participant evidence distinct from a preferred competitive story.

Rule of thumb: Treat supplier statements as descriptions of an offer, not proof of outcomes.

If participants describe an ideal process instead of a real one

If accounts stay vague, anchor discussion in a recent event or observe a task where it is safe. Use synthetic or cleared material for demonstrations, and distinguish what the participant did from what the researcher inferred.

Ask what arrived first, where a decision moved, how the approved version was recognised and what happened when work was rejected.

Customer interviews work best when prompts do not steer people towards approving a feature. Do not invite disclosure of client credentials, unreleased advertising, customer data or confidential supplier terms. A shared discussion guide supports comparison while leaving room for follow-up; if observation is unsuitable, reconstruct a specific event instead of relying on general descriptions.

Rule of thumb: Test a safe, recent task before trusting a general account.

If personal or campaign material may enter the study

If research could involve names, contact details, employer information, access needs, recordings or campaign material, decide what is necessary before recruitment. Explain the purpose, data collected, recipients, recording, retention, withdrawal and complaint routes. Separate scheduling details from research notes, restrict access and set a deletion point.

Name relevant processors and collect only what is needed to answer the question. A qualified privacy reviewer should help the accountable organisation document its role, purpose and lawful basis. Consider additional assessment where special-category data, children, vulnerable people, monitoring or high-risk profiling may be involved. UK GDPR sets no universal retention period; the controller must justify its own storage limit. See data protection planning.

Rule of thumb: Reduce sensitive material before inviting participants.

If findings point in different directions

If accounts conflict, separate concrete observations from interpretations and keep examples that challenge the apparent pattern. Segment results where role, authority or working environment changes what a finding means. Show which evidence supports each conclusion and where that support stops, rather than forcing a single tidy audience story.

Choose whether the next move is a bounded test, a narrower group or workflow, another kind of evidence, or stopping. Do not report a survey percentage without its questionnaire, recruitment source, field dates, response count, denominator, exclusions and geographic coverage. A convenience sample describes its respondents, not every English advertiser; repeated qualitative evidence does not establish prevalence.

Rule of thumb: Let the limits of the evidence determine the next move.

If the findings are ready to become a persona

Build the persona from the notes in four steps, so each claim can be traced.

Here is an illustrative example, not a real study. Persona: handoff approver. Task: approving final creative from freelancers. Evidence: 5 of 8 interviewees lost approval context at handoff, 2 had a working comment trail, and 1 did no handoffs. Obstacle: the approved version is hard to recognise. Decision informed: whether to build a version and approval trail first. Limits: eight English interviews, so it is not a prevalence estimate. Review date: six months.

If fieldwork is ready to begin

If the decision and method are clear, prepare a brief before contacting participants. Record groups included and excluded, recruitment source, screening criteria, consent and withdrawal routes, data controller, processors, storage locations, deletion point, access arrangements and the person who owns the decision.

Add restrictions on client material, synthetic alternatives, the analysis method, the decision rule and the source cut-off. The record should let another reviewer see what evidence will be gathered, how it will be handled and which choice it informs, without implying more certainty than the research can support.

Rule of thumb: Make the evidence traceable before recruitment starts.