How to Analyze User Interviews: A Wall of Sticky Notes Is Not an Insight
Two common extremes after interviews are equally unreliable: writing conclusions from memory or clustering hundreds of sticky notes, photographing the wall, and declaring "insights complete." Real analysis separates raw evidence, interpretation, and recommendations so others can trace every conclusion.
Two common extremes after interviews are equally unreliable: writing conclusions from memory or clustering hundreds of sticky notes, photographing the wall, and declaring "insights complete." Real analysis separates raw evidence, interpretation, and recommendations so others can trace every conclusion.
01 Separate Five Levels: Quote, Observation, Finding, Insight, and Action
These terms are often mixed in projects, making it impossible to tell what users said and what researchers inferred.
| Level | Example | Watch For |
|---|---|---|
| Quote | "I usually reconcile all invoices at the end of the month." | Preserve context; do not quote selectively |
| Observation | The participant opened email first, then copied information manually into Excel | Describe observed behavior without evaluation |
| Finding | Multiple finance employees search for evidence across different channels | A pattern across participants, not a summary of one sentence |
| Insight | Scattered evidence prevents Finance from confirming the completeness of monthly accounts | Explain why the pattern affects the objective |
| Action | Create a unified billing center with missing-document alerts | A team choice, not a research fact |
If an analysis table places all of these in one "User Feedback" column, team assumptions easily become user facts later.
02 Begin Analysis Immediately After Every Interview Round
Do not wait until all ten sessions end. Spend 15–30 minutes after each interview reviewing what happened, what was surprising, and what the next session should continue testing. This allows the guide to evolve and prevents the final participant from dominating memory.
Analysis should include the moderator, note-taker, and at least one observer. Collaborative organization reduces one researcher's bias only when everyone returns to the record instead of debating meeting memories.
03 Step One: Break Records into Evidence Units
One evidence unit communicates one understandable fact while preserving participant, context, and source. For example:
"P03 | Finance Manager | Month-End Reconciliation | Must download evidence from email, WeChat groups, and three systems, and worries about missing temporary refunds."
Do not write "The finance process is complex." That is already an interpretation and loses the specific source of complexity.
Recommended fields include:
- Participant ID and role.
- Scenario or task stage.
- Exact quote, behavior, or interface action.
- Evidence type: said, did, data record, or document.
- Researcher note in a separate column from facts.
- Audio or video timestamp or original-material location.
04 Step Two: Coding Is Not Decorating Sentences with Labels
Coding adds comparable meaning to evidence. The first pass can stay close to the words, such as "month-end batch reconciliation," "cross-channel evidence search," and "fear of missing refunds." The second pass combines related codes into themes such as "scattered reconciliation information."
Do not force records into predefined categories at the outset. If everything must fit Efficiency, Experience, or Features, new problems cannot emerge.
A Small Coding Example
| Evidence | Initial Code | Possible Theme |
|---|---|---|
| Process the Entire Month's Invoices at Month-End | Concentrated Batch Processing | Work Rhythm and Batching |
| Search Email for "Refund" | Cross-Channel Search | Scattered Information |
| Avoid Deleting Processed Attachments | Missing Completion Confirmation | Insufficient Traceability |
| Historical Evidence Is Lost After a Colleague Leaves | Materials Depend on Individuals | Organizational Risk |
05 Step Three: Preserve Contradictions During Clustering
An affinity map should not force every sticky note into "consensus." The same feature may be a shortcut for experienced users and a risk for beginners; large customers may need strict approval, while small teams see approval as a delay.
Contradictions may come from:
- Different roles and permissions.
- Task-frequency differences.
- Different expertise or organization size.
- Different markets, devices, or channels.
- A gap between the participant's idealized process and actual behavior.
Preserve applicability and counterexamples under each theme rather than claiming "80% of users need this." Qualitative interviews primarily explain reasons and reveal patterns; small-sample frequency should not impersonate market proportion.
06 Step Four: Assess Evidence Strength Instead of Counting Mentions Alone
Evaluate four dimensions:
| Dimension | Stronger Evidence | Weaker Evidence |
|---|---|---|
| Proximity to Behavior | Observed real action or authentic materials | Verbal prediction about the future |
| Sample Coverage | Appears across roles, contexts, or rounds | Comes from one extreme participant |
| Clear Consequence | Causes time, error, risk, or abandonment | Only a personal preference |
| Traceability | Exact quote, recording, log, or document | Only a researcher's memory remains |
A problem mentioned once that can cause a major financial error may deserve more priority than a color preference mentioned by six people.
07 Step Five: Write Findings with Boundaries
Vague finding:
"Users think billing is difficult to use."
Better wording:
"Finance employees performing monthly reconciliation cannot confirm all payment, refund, and invoice states in one place, so they manually recheck email and chat records at month-end."
It contains role, task, friction, and consequence without prescribing a solution prematurely.
Use this template:
"When [role] completes [task] in [context], [behavior or consequence] occurs because [reason]. This judgment is based on [evidence scope], while [uncertainty] still needs validation."
08 Step Six: An Insight Must Explain Why Current Behavior Is Rational
Strong insights do not label users; they explain why apparently inefficient behavior makes sense.
Continued Excel use, for example, may not indicate resistance to a new system. Excel may quickly combine data from several channels while preserving familiar reconciliation traces. A new product that copies the spreadsheet's appearance without solving cross-channel integration and traceability will still fail to attract migration.
This type of explanation helps teams avoid superficial features.
09 Step Seven: Turn Opportunities into Discussable Questions
Do not jump from "users cannot find evidence" to "build an AI assistant." Write opportunity questions first:
- How might we help Finance confirm evidence completeness at month-end?
- How might we retain processing state without adding data-entry work?
- How might we preserve historical traceability after an employee handoff?
Opportunity questions preserve several solution directions for product, design, and technology teams to evaluate together.
10 Analysis Outputs Should Enable Action
Include at least:
- Research objective and sample scope.
- Key findings, each with evidence and applicability boundaries.
- Contradictions and unvalidated questions.
- Opportunities and business impact.
- Recommended action: fix immediately, explore a solution, continue research, or defer.
- An index of raw evidence for traceability.
Do not place every interview record in the presentation. The body should explain what decisions require; appendices retain the evidence chain.
11 Run an Analysis Meeting That Is More Than a Researcher Presentation
Have observers write evidence independently before joint clustering. Discuss what happened first, what it means second, and solutions last. If product leaders repeatedly say "we cannot build this" during clustering, the team deletes real problems too early.
A 90-minute analysis meeting can use:
- 15 minutes: Review objective and sample.
- 25 minutes: Extract evidence independently.
- 25 minutes: Cluster and name themes.
- 15 minutes: Write findings and contradictions.
- 10 minutes: Confirm actions and questions to validate.
Complex projects usually need several rounds, not one meeting for all analysis.
Frequently Asked Questions
Must Every Interview Be Transcribed Word for Word?
Not necessarily. Full transcription suits high-risk, sensitive research or work requiring exact quotations. General projects may use detailed notes, key clips, and timestamps. The choice depends on analysis depth and privacy cost.
How Many Participants Must Mention Something for It to Be a Finding?
There is no fixed threshold. Consider the sample, behavioral evidence, consequence, and research objective. Do not turn qualitative work into small-sample voting.
Can AI Summarize Interviews Directly?
It can assist transcription, initial coding, and clip retrieval, but privacy and authorization must be addressed, and researchers must verify against original materials. AI can flatten contradictions, lose tone, and mistake frequent words for important problems.
What If Interview Findings Conflict with Analytics?
First check whether definitions and samples agree. Analytics show what happened; interviews often explain why. The conflict itself may reveal a segment or measurement problem and should not be resolved by choosing whichever result confirms expectations.
12 Insight Is Valuable When It Is Traceable and Changes a Decision
Valuable interview analysis does not produce more statements that "users want" something. It creates a clear path from evidence to interpretation to action. Even when the team disputes a recommendation, everyone can return to the same evidence instead of letting the loudest voice decide the product.
| Service | View |
|---|---|
| UI/UX Design Services | View Service Details |
| Project Inquiry | Contact JVDS Design Studio |
| Design and Website Development Articles | Read More Articles |
Link copied
