Buying principleA research round is not a referendum. It exposes evidence that can change the next product decision.
A UX research agency should start with the decision
The brief should name one decision, the uncertainty behind it and what the team can change. “Understand our users” is too broad to guide recruitment or a session.
Imagine a subscription product with a steep drop between plan selection and payment. The useful question is not whether people like the checkout. It is whether new buyers understand the commitment, trust the payment step and can recover from an error.
The GOV.UK Service Manual recommends turning assumptions into research questions and choosing activities that answer them with the least time, effort and cost. A proposal should therefore connect every session to a named decision owner and a date when the evidence will be used.
If the real uncertainty is already visible in analytics or support logs, start there. An external agency adds value when observation, neutral facilitation or access to participants is missing, not when the team simply wants a more expensive opinion.
Recruit participants who can expose the actual risk
Recruitment quality matters more than a neat demographic spread. Participants must be actual or likely users who have faced the situation being tested.
For a business invoicing tool, five random office workers cannot represent finance managers who reconcile exceptions every Friday. The screening brief should cover role, recent behaviour, device, domain experience and any access needs that affect the task.
GOV.UK says qualitative rounds such as interviews or usability tests will often use four to eight participants. That is a planning range, not a universal proof threshold. The right response to unclear evidence is often another adjusted round, not a larger first batch.
Ask the agency who recruits, who screens, how substitutions work and what happens after a no-show. For specialist or disabled participants, allow more lead time and budget for interpreters, travel or assistive technology.
Match the method to the question before buying sessions
A polished lab does not rescue the wrong method. Choose the lightest activity that can expose the behaviour or belief behind the decision.
If customers abandon a form, watch representative people attempt the task. If the team does not understand how approvals happen at work, observe the workflow or run contextual interviews. If it needs a population estimate, a handful of conversations cannot provide one.
| Decision question | Useful method | Evidence produced | Common mistake |
|---|---|---|---|
| Can people complete this journey? | Moderated usability test | Observed actions, errors and recovery | Explaining the interface during the task |
| How does work happen today? | Contextual research | Tools, handoffs and constraints in context | Relying only on remembered behaviour |
| Why do people choose or avoid it? | In-depth interview | Language, motives and decision history | Treating stated preference as future behaviour |
| How common is this pattern? | Survey or behavioural analysis | A distribution across a larger sample | Generalising from a small qualitative round |
Consent and recordings belong in the delivery plan
Research creates personal data before it creates insight. The proposal should state what will be collected, who can see it, where it is stored, how long it remains and how withdrawal works.
A participant who shares their screen may reveal customer names, browser tabs or private messages. “The call is recorded” is not enough. The introduction must explain observers, recording, intended use and the participant's right to stop.
GOV.UK guidance requires informed consent and deletion of research data if consent is withdrawn. Your contract should identify the data controller, any recruitment or transcription suppliers, retention periods and the deletion process.
Do not ask for recordings by default. A clipped observation with consent, structured notes and a traceable finding may be safer and easier for the product team to use.
Make the product team watch before it receives a deck
The fastest route from research to action is direct observation. Product, design, engineering and content owners should watch the same task fail, then join the analysis while the context is fresh.
For example, a designer may see a confusing label while an engineer notices that the recovery state loses entered data. The combined observation leads to a better fix than a screenshot labelled “users found this difficult.”
Plan observers, note-taking roles and a short debrief after each session. The GOV.UK planning guide recommends protecting analysis time and involving the team. An agency should make this participation easy, not hide the work until presentation day.
When stakeholders cannot attend, require short evidence clips or annotated observations linked to the decision log. Remove participant identifiers unless they are genuinely necessary.
Turn observations into ranked product decisions
A finding should connect evidence to consequence and action. “Three users missed the button” is an observation. The decision is whether the journey, copy or visual hierarchy changes, who owns it and how the next round will test it.
Use a compact record with the research question, participant context, observed pattern, confidence, affected journey, recommendation and owner. Separate a repeated failure from one person's preference. Preserve dissent when evidence conflicts.
Accessibility research needs the same restraint. W3C warns that input from one disabled participant does not represent everyone with that disability, and that user evaluation does not replace standards testing. Combine involvement from users with disabilities with a conformance review when accessibility is in scope.
The final readout should show what changed, what did not, what remains unknown and the next test. This keeps research honest and lets a later team reconstruct the decision.
External research is the wrong purchase when nothing can change
Do not commission a study when the release is already locked, the team will not expose the prototype, or no owner can act on the evidence. Research becomes theatre when the answer cannot alter scope, sequence or design.
An agency is also unnecessary when a small internal test can answer the question safely. A product manager and designer can often run a practice session on a low-risk prototype, provided recruitment and consent are handled properly.
External support earns its cost when the participant group is hard to reach, the topic needs experienced moderation, internal politics need a neutral facilitator or the team lacks a repeatable research practice.
If the issue is known but unprioritised, buy an evidence-led UX audit or delivery support instead. If the product question is still undefined, start with a product discovery workshop.
UX research agency checklist before you sign
- Name the product decision, owner and date the evidence will be used.
- List the assumptions and research questions in priority order.
- Define actual or likely users, screening criteria and access needs.
- Match each question to a method instead of buying a fixed number of interviews.
- Confirm incentives, no-show rules, consent, retention and deletion.
- Invite the delivery team to observe and join analysis.
- Require evidence-linked findings, confidence and named owners.
- Keep analytics evidence separate from what qualitative research can prove.
- Pair disabled-user sessions with a structured accessibility audit when conformance matters.
If you have one product decision that needs real user evidence, send Seuil the decision and the user group affected. That is enough to judge whether research, an audit or a smaller internal test is the sensible next step.