Build an AI-powered user research synthesis workflow

The Problem

You have ten user interviews sitting in your inbox. Each one is forty-five minutes of rich qualitative data. Your team conducted them over two weeks, with careful recruitment, thoughtful discussion guides, and genuine empathy. And now they sit unprocessed, because synthesis takes longer than the interviews themselves.

This is the dirty secret of user research: teams that conduct regular user interviews synthesize fewer than 40% of them into actionable insights, according to a 2023 study by User Interviews. The rest decay in shared drives, remembered imperfectly, cited selectively, and eventually forgotten.

The problem is not motivation. It is throughput. A single interview generates 8,000-12,000 words of transcript. Synthesizing across ten interviews means pattern-matching across 100,000 words while maintaining fidelity to individual participant contexts. This is cognitively exhausting work, and most teams do not have dedicated researchers to do it.

Nielsen Norman Group reports that 85% of usability issues can be identified from just five user tests, but only if those tests are properly synthesized. Without synthesis, five tests produce five anecdotes, not five data points.

The Synthesis Bottleneck

Research without synthesis is just conversation. The value is not in the interview. It is in the pattern that emerges across interviews. When synthesis is slow, the roadmap gets built on the last interview the PM remembers, not the full body of evidence. This is how teams end up building for one vocal user instead of a genuine market need.

How This Prompt Works

This prompt creates an AI-powered research synthesis workflow that processes interview transcripts into structured insights without losing the nuance of qualitative data.

The workflow operates in four stages:

• Transcript ingestion: Feed raw transcripts or interview notes into the system. The AI extracts verbatim quotes, behavioral observations, pain points, and unmet needs from each interview.
• Code generation: The AI applies thematic coding across all transcripts, identifying recurring themes, contradictions, and outliers. It uses both deductive codes (from your research questions) and inductive codes (emergent from the data).
• Pattern synthesis: Coded data is clustered into insight statements, each grounded in specific evidence. Every insight links back to the participants and quotes that support it.
• Opportunity mapping: Insights are translated into opportunity statements compatible with Teresa Torres' Opportunity Solution Tree framework, ready for prioritization.

The output is a structured synthesis document with evidence chains, not AI-generated summaries that lose provenance.

When to Use It

• You have a backlog of unprocessed interviews and need to catch up
• Your team conducts regular interviews but struggles to turn them into roadmap inputs
• You want to scale research synthesis without hiring a dedicated researcher
• You need to present research findings to stakeholders and want rigorous evidence chains

Common Pitfalls

• Trusting the AI blindly: AI is excellent at pattern matching but poor at detecting sarcasm, social desirability bias, and context-dependent meaning. Always review the coded data against your memory of the conversation.
• Skipping the raw review: Read at least 20% of transcripts yourself before running synthesis. You need intuition about the data to evaluate whether the AI's patterns are real.
• Synthesizing too early: Wait until you have at least five interviews before synthesizing. Fewer than five and you are finding patterns in noise.

Further Reading

• The State of User Research Report - User Interviews
• How Many Test Users in a Usability Study - Nielsen Norman Group
• Continuous Discovery Habits: Opportunity Solution Trees - Product Talk