The primary market research best practices worth adopting first

The case opens with a causal question

A subscription business is concerned that poor onboarding drives churn. Its proposed research combines interviews with a survey of current and lapsed customers. This is a plan, not a record of completed fieldwork, so the scenario can establish only what the design checks support, not whether onboarding actually causes churn.

The decision was fixed before recruitment

The team first writes down the decision, its owner, the evidence gap, and the action threshold. If no possible finding would change the next move, recruitment has no defensible purpose. The quality of commissioned research therefore starts with a real decision, not with the questionnaire.

The method check separated explanation from proof

The statement that poor onboarding causes churn is causal. Interviews may reveal reasons, while a survey can estimate their prevalence only if its sampling frame is defensible. An experiment or A/B test could examine a specified effect under tested conditions, but would not explain why a change worked.

Qualitative work explains possible mechanisms. Interviews map possible reasons, while a survey of current and lapsed customers can assess how common those reasons are within its covered group. Because the plan specifies no experiment, neither output establishes that onboarding causes churn.

The frame changed the scope of the claim

The next check asks who each recruitment source can describe. A sample represents the frame from which respondents were drawn, and a customer file excludes people outside it. The team records whether recruitment uses a list, panel, intercept, or customer file, then names groups that source cannot reach.

If the survey uses the customer file, it describes customers in that file, not the whole market or people who never bought. A panel of app users similarly describes app users only. Neither limitation invalidates the work if the report names it; burying it in an appendix invites overclaiming.

State exclusions in the report itself, where readers can judge coverage before sign-off. A primary market research checklist makes source, coverage, and exclusions visible before conclusions are approved; a narrow sample is workable when the report keeps its claim within its actual reach.

The participant route was inspected end to end

The team checks both eligibility and whether people can take part safely and knowingly. Capacity to consent, dependency on the client, language, disability, and safety at the venue shape who can participate and whose experience enters the evidence from the outset.

The HHS Office for Human Research Protections' vulnerable populations guidance index points to additional protections for specified groups in covered human-subjects research; it is not a commercial-research checklist. Where health, minors, or other covered ground is involved, take the question to the institution's review board rather than treating that guidance as clearance.

Recruit leavers and non-users alongside satisfied customers, rather than relying only on people who remain engaged. Accessibility checks alone do not prove inclusion: recruitment, support, language, venue safety, analysis, and later use of findings all affect who is represented. For teams new to commissioning, primary research planning includes an accessible recruitment and withdrawal route.

Explain the purpose and activity, what data is collected and who will see it, how long it is retained, and how withdrawal works. Keep research separate from promotion and lead generation, minimize personal data, and require suppliers with access to follow the same terms throughout the study.

For the privacy check, the FTC's guide to protecting personal information offers a practical reference. Questions about California consent or CPRA belong with qualified counsel, not the research lead; a research method is not a substitute for legal advice.

The World Wide Web Consortium describes perceivable, operable, understandable, and reliable principles for accessible web content. Conformance applies to content, not sample fairness. Test the instrument with disabled participants and assistive technology instead of relying only on automated checking as part of the participant journey.

Pilot every stage from recruitment through deletion, not just the questionnaire. Check the handoffs across the full journey, including the items below; fix failures, freeze a version, and log every later change before wider fieldwork begins, so the test remains reproducible from one stage to the next.

The checks produced limits, not churn estimates

No interview or survey outcomes are specified in this case. The checks therefore produce an evidence boundary, not a churn result: a customer-file survey describes covered customers, not people who never bought or the whole market, while interview and survey evidence alone does not establish cause.

Before analysis, define coding, exclusions, missing-data handling, weighting, and subgroup comparisons for consequential questions, and retain original recordings and transcripts. Distinguish what participants said from interpretation, recommendation, and management decision. A written protocol also guards against selective exclusions that can turn weak evidence into confident claims.

The team acted only within those limits

The defensible move is to use interviews to map reasons and a survey of current and lapsed customers to test how common those reasons are, without presenting the combination as causal proof. The report should make the population and limits visible before a decision is made.

Report the population, method, field dates, sample source, uncertainty, and unreachable groups. Record the decision and what happened next, separating participant statements from interpretation and recommendation. Those details let readers see which decision the evidence can support and which questions remain unresolved.

Cost the plan using the incentive per session multiplied by sessions, plus panel or list cost per complete multiplied by completes, then add tool licensing and analysis days. Compare the total with the available budget; price changes neither what a method can establish nor the population the sample represents.

Rules carried forward from this case

Keep the claim within the sample's frame, choose methods that answer the question asked, and write down the decision before collecting evidence.

Make exclusions and analysis traceable, then state in advance how a finding could change a real action. These habits keep the evidence useful without asking it to prove more than it can.

An accessible interface alone does not make a study inclusive. Recruitment, support, language, safety, analysis, and later use all affect representation. Share negative findings internally with the same care about method and limitations as favorable ones: preventing an unsupported launch can be a return on the budget, not a failed study.

Protect the team from pressure to confirm a preferred plan, and protect participants from uses they were not told about. Set an expiry date when commissioning the work; retire evidence when the market, product, or population has changed enough to weaken its predictive value, and revisit the date at each planning cycle.

Further reading and references: