A problem is a costly situation, not a requested feature
Problem discovery seeks evidence that a recognizable person repeatedly encounters an obstacle while trying to make progress. An interviewee saying a tool 'would be nice' is not the same as describing a recent task, failed attempt, consequence, and workaround. Begin with the situation: what triggered action, what the person tried, what made the result inadequate, and what happened next. Content opportunities emerge from unresolved decisions and constraints, not from asking people to invent article titles or predict whether they would buy an unspecified solution.
Evidence: Government Digital Service; Pew Research Center
What people say and what they did answer different questions
A participant's language is valuable evidence of meaning, expectations, and remembered experience. It is not a direct recording of behavior and can be influenced by memory, politeness, question wording, or the desire to help the researcher. GOV.UK recommends focusing on stories and real examples rather than how things should happen. Treat 'I always compare prices' as a claim to examine; a recent account of opening three tabs, abandoning checkout, and asking a friend provides a more specific sequence. Neither alone proves market frequency.
Evidence: Government Digital Service
Neutral structure protects discovery from the solution
An interview guide should state the research question without embedding the hoped-for answer. Begin broadly, ask about the last relevant episode, and use follow-ups such as what happened, when, who was involved, and what made that difficult. Pew's methods material shows that wording and order can influence responses, a principle that matters even though an interview is not a probability survey. Show a concept only after problem exploration, and label reactions to it as concept feedback rather than evidence that the original problem existed.
Evidence: Pew Research Center; Oxford Academic
The evidence table keeps observation separate from interpretation
Create columns for anonymous participant code, recruitment criterion, recent situation, trigger, verbatim phrase, observed or reported action, workaround, time or money consequence, frequency claim, confidence, contradiction, researcher interpretation, and next test. One row represents one episode, not an entire person. Keep direct language distinguishable from analytical labels. If a participant supplies only a general opinion, record that limitation instead of rewriting it as an event. This structure allows another editor to see how a content problem was derived and where inference entered.
Evidence: Government Digital Service
A constructed interview fragment shows the distinction
Suppose someone says, 'Choosing a standing desk is confusing.' A leading follow-up—'Would a comparison calculator help?'—moves immediately to the researcher's solution. A behavioral follow-up asks about the last purchase attempt. The person recalls comparing weight limits, discovering that return shipping was unclear, and postponing the order. The testable content problem is not generic confusion; it is estimating fit and return downside before buying. This example illustrates analysis and is not a claim about a real TenMultigure interview.
Evidence: Government Digital Service; Pew Research Center
Triangulation reduces confidence in a convenient story
GOV.UK analysis guidance advises extracting what was seen or heard before deciding findings and involving several observers to limit individual bias. Compare interviews with support questions, search patterns, community discussions, or legitimate behavioral records. Seek a deviant case: a person who solved the task easily or chose a different workaround. Agreement across methods strengthens a hypothesis only within the sampled context; disagreement may reveal distinct situations rather than bad data. Small qualitative rounds do not estimate population prevalence.
Evidence: Government Digital Service; Oxford Academic
The next step is a falsifiable content problem
Write one statement containing audience situation, desired progress, current workaround, costly barrier, and the evidence that would disconfirm it. Then choose a small content test that helps with the decision without pretending to validate demand by itself. Protect consent and remove unnecessary identifiers from notes. Interviews cannot establish market size, willingness to pay, or future behavior on their own. Stop when the team can decide which problem deserves the next evidence round, not when every possible user has been interviewed.
Evidence: Government Digital Service; Government Digital Service
Sources and further reading
These references informed this article. A source supports a claim; it does not imply endorsement of TenMultigure or any future product reference.
- Using in-depth interviewsGovernment Digital Service · Accessed August 10, 2026
Supports using recent stories and open, neutral follow-ups to distinguish lived obstacles from hypothetical feature enthusiasm.
- Analyse a research sessionGovernment Digital Service · Accessed August 10, 2026
Provides the observation-to-finding method used to keep participant words separate from the editorial interpretation.
- Writing Survey QuestionsPew Research Center · Accessed August 10, 2026
Grounds the warning that question wording and order can shape answers even when the session feels conversational.
- The Effect of the Question on Survey Responses: A ReviewOxford Academic · Accessed August 10, 2026
Adds independent evidence on response effects, reinforcing the need to label interview claims by their elicitation context.
Reviewed by TenMultigure Editorial Review. See an error or a source that has changed? Tell the editorial team.
Review method: AI-assisted desk research with editorial checks. Reviewed ; next scheduled review . Defined interview-based problem evidence around recent behavior, added an anonymized episode table and constructed desk example, and bounded qualitative claims with contradiction and triangulation.