Customer Interview Synthesizer

Customer Interview Synthesizer

> Turns hours of qualitative interviews into a tight, evidence-anchored set of opportunities.

What this skill is

A structured workflow for synthesizing interview transcripts: code utterances, theme them, score saturation, extract Jobs-to-be-Done (JTBD) signals and forces of progress, identify opportunity hypotheses, and produce a readout that distinguishes evidence from interpretation. Designed to remove the analyst's narrative-fitting bias and keep the customer's voice central.

What it solves

• "What did we learn?" readouts that summarize whichever interview was freshest
• Cherry-picking quotes that support a pre-existing hypothesis
• No theme saturation check (claiming a theme on a sample size of one)
• Conflating what the customer said with what the analyst inferred
• Failure to translate findings into prioritized hypotheses with confidence levels

When to invoke

• Concluding a discovery or generative interview round
• Synthesizing switching interviews after a churn wave
• Win-loss analysis from a sales cycle
• Sensemaking after a competitive deep-dive
• Feeding the Opportunity Solution Tree (OST) with fresh evidence

Phase 1: Catalog the corpus

For each interview, log:

• Participant pseudonym, segment, role
• Date, duration, interviewer
• Recording or transcript location
• Recruitment source (and any selection bias)
• Whether incentives were offered
• Whether the participant is current, former, or prospective customer

Minimum saturation guideline: 5-8 interviews per segment for generative discovery; more for major decisions.

Phase 2: Code the utterances

Read each transcript line by line. Tag passages with codes - short labels for the topic discussed.

Two-pass approach:

1. Open coding - generate codes inductively from the data
2. Axial coding - group related codes into categories

Each code must include:

• A verbatim quote from the transcript
• Interview Identifier (ID) and timestamp
• Whether it's a report (something they did or experienced) or opinion (something they said they'd do)

Reports far outweigh opinions in predictive value. Flag the distinction.

Phase 3: Theme the codes

Group codes into themes. For each theme:

• Theme title in the customer's voice
• Definition
• Number of participants who mentioned it (saturation count)
• 3-5 representative verbatim quotes
• Segments where the theme is strongest
• Disconfirming evidence (interviews where the theme didn't appear)

Saturation rule: a theme appearing in less than 30% of interviews in a segment is a hypothesis, not a finding.

Phase 4: Extract JTBD signals

For passages describing what the customer was trying to accomplish:

• Write the job statement: When [situation], I want to [motivation], so I can [outcome].
• Classify functional / emotional / social
• Identify the job stage (Define / Locate / Prepare / Confirm / Execute / Monitor / Modify / Conclude)

Phase 5: Forces of progress

For switching, churn, and win-loss interviews, map the four forces per participant:

| Force | Evidence from interview | |-------|------------------------| | Push of current situation | What was broken or frustrating? | | Pull of new solution | What did they hope for? | | Anxiety of new solution | What worried them about switching? | | Habit of old | What kept them on the old solution? |

Aggregate across interviews to identify which force dominates the decision.

Phase 6: Generate opportunity hypotheses

Convert themes into opportunity hypotheses:

> We believe [customer segment] experiences [pain or unmet need] when [situation], because [evidence]. > > Confidence: [High / Medium / Low] based on [number of interviews] and [type of evidence]. > > If we solved this, we'd expect to see [observable outcome].

Each hypothesis links back to:

• The theme(s) it came from
• The verbatim quotes that support it
• The disconfirming evidence (where it didn't show up)

Phase 7: Separate observation from interpretation

The synthesis output has two tiers, clearly labeled:

• What we heard (verbatim, quoted, sourced) - high confidence
• What it might mean (analyst interpretation) - explicitly labeled as inference

Never blur the line. Stakeholders will treat them as the same unless you separate them.

Output

• Coded transcript library (referenceable for any future synthesis)
• Theme list with saturation counts and supporting verbatims
• JTBD inventory: functional, emotional, and social jobs by segment
• Forces-of-progress diagram for switching or churn signals
• Opportunity hypothesis register with confidence levels and source evidence
• 1-page readout with the 3 most decision-grade findings
• Explicit list of disconfirming evidence
• Recommended next-round research questions

Operating rules

Always

• Quote verbatim before paraphrasing
• Distinguish reports from opinions
• Score theme saturation (count of interviews / total interviews per segment)
• Surface disconfirming evidence explicitly
• Label observation versus interpretation

Never

• Claim a theme from a single interview
• Cherry-pick quotes that confirm a hypothesis
• Use the analyst's words as if they were the customer's
• Skip the recruitment-bias note
• Merge wildly different segments into one theme

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: varunk130
• Source: varunk130/claude-code-skills
• License: MIT

Install and usage instructions live in the source repository linked above.