interview-to-jtbd

Turn messy customer interviews into JTBD maps, opportunity areas, product briefs, and hypothesis backlogs - with evidence, not vibes.

Overview

interview-to-jtbd is a universal agent skill for analyzing user interviews, research notes, and open-ended customer feedback. It helps an AI agent turn raw materials into an evidence-backed JTBD (Jobs to be Done) map, opportunity map, product brief, hypothesis backlog, and plan for subsequent interviews.

The core idea is not to build a separate SaaS or CLI as the main entry point, but to package the methodology into a portable skill that different AI agents can use: Claude Code, OpenAI Codex, Cursor, Gemini CLI, and any agent that can read SKILL.md-style instructions.

What It Generates

• JTBD map: jobs, situations, motivations, expected outcomes, and supporting quotes.
• Opportunity map: unmet needs, current workarounds, severity, and product angles.
• Product brief: problem framing, target users, constraints, risks, and evidence.
• Hypothesis backlog: build-measure-learn hypotheses grounded in research.
• Follow-up interview guide: questions for the next discovery round.

Quick Demo

Input excerpt

"I keep interview notes in three docs, then spend Friday copying quotes into a spreadsheet. By the time I write the roadmap update, I am not sure which quote supports which decision."

Agent output

## Job: Trace research evidence to product decisions

**Statement:** When I prepare a roadmap or product brief after customer interviews,
I want to connect each decision to the exact evidence behind it,
so I can defend priorities without manually rebuilding the research trail.

**Evidence:**
- "spend Friday copying quotes into a spreadsheet"
- "not sure which quote supports which decision"

**Opportunity:** Reduce the manual work of turning interview notes into traceable product artifacts.

**Hypothesis:** If PMs can generate an evidence table and JTBD map from raw notes,
then they will produce clearer product briefs faster and with fewer unsupported claims.

See a fuller example in examples/demo-output/customer-research-synthesis.md.

Prerequisites

• An AI agent that can read repository files: Claude Code, Codex, Cursor, Gemini CLI, or a similar tool.
• Git, if you want to clone the repository.
• Customer interview transcripts, research notes, survey open-ends, support tickets, or other qualitative feedback.
• Optional: an approved local or hosted LLM setup for confidential research data.

No package manager, API key, database, or SaaS account is required by the skill itself.

Install

1. Clone the skill

git clone https://github.com/lowwwbank/interview-to-jtbd.git
cd interview-to-jtbd

2. Open it with your agent

Use whichever agent you already work with:

# Claude Code
claude

# OpenAI Codex
codex

# Gemini CLI
gemini

You can also open the folder in Cursor or any editor where your AI agent can read SKILL.md and the references/ directory.

3. Point the agent at your research

Put transcripts or notes in a folder such as:

research-notes/
|-- interview-01.md
|-- interview-02.md
`-- survey-open-ends.csv

Then ask the agent:

Use SKILL.md as your operating instructions.
Analyze ./research-notes.
Generate an evidence-backed JTBD map, opportunity map, product brief,
hypothesis backlog, and follow-up interview guide.
Do not invent quotes or claims. Mark unsupported assumptions explicitly.

Alternative: install inside another agent workspace

If you already keep agent skills in a project folder, vendor this repository as a subfolder:

mkdir -p .agents/skills
git clone https://github.com/lowwwbank/interview-to-jtbd.git .agents/skills/interview-to-jtbd

Then tell your agent to use .agents/skills/interview-to-jtbd/SKILL.md.

Alternative: copy-paste only

If your agent does not support local skill folders, copy the contents of SKILL.md into the agent's custom instructions or prompt window, then paste or attach your research notes.

Product Positioning

Category

Agent Skill / AI research synthesis workflow.

Target Audience

• Product managers doing their own discovery.
• Founders and indie hackers needing to quickly understand customers.
• AI builders who don't want to manually process interviews.
• UX researchers in small teams without budget for Dovetail/Marvin/Condens.
• Product bootcamps, students, product communities.

Value Proposition

Enterprise research repositories solve the problem of centralized storage and teamwork. interview-to-jtbd solves a narrower problem: quickly and reproducibly turning a set of interviews into actionable product artifacts, while preserving the link between every conclusion and its evidence.

Positioning Formula: Dovetail-style research synthesis for AI agents and markdown-first teams.

Market Context

While existing products (Dovetail, Marvin, Condens, Looppanel, Aurelius, Productboard) address broad enterprise/workspace scenarios, interview-to-jtbd fills the open niche:

• open-source
• skill-first
• markdown/file-first
• local/privacy-first
• no workspace, billing, onboarding, or SaaS account required
• focuses on JTBD, rather than general qualitative analysis.

Workflow

Raw interviews      Evidence extraction      JTBD synthesis       Product artifacts
      |                     |                      |                       |
      v                     v                      v                       v
Transcripts  -->  Quotes, observations  -->  Jobs, outcomes  -->  Briefs, hypotheses,
Notes             workarounds, risks         opportunities        follow-up questions

Methodology

Based on:

• JTBD: Users "hire" products to make progress in a specific situation.
• Outcome-Driven Innovation: Focus on job, desired outcomes, unmet needs, and opportunity areas.
• Human-in-the-loop: When using LLMs in qualitative research, human oversight, evidence grounding, and bias mitigation are crucial.

Design Constraints

The skill is universal and does not depend on a specific AI platform.

• SKILL.md is the portable core.
• No agents/openai.yaml required in the core.
• Does not rely on OpenAI-only metadata.
• No app installation needed for basic usage.
• No Node/Python needed for basic usage.
• No API key required at the skill level (provider determined by the user's agent).
• Does not store interviews in a third-party service.
• AI synthesis is never presented as fact without evidence.

Repository Structure

.
|-- SKILL.md
|-- references/
|   |-- jtbd-methodology.md
|   |-- output-schemas.md
|   |-- privacy-and-ethics.md
|   `-- research-synthesis-rubric.md
|-- assets/templates/
|   `-- research-summary.md
`-- examples/demo-output/
    `-- customer-research-synthesis.md

Quality Principles

• Every claim must be traceable to source material.
• Observations, inferences, assumptions, and recommendations must be separated.
• Sparse data should produce lower confidence, not stronger conclusions.
• Personally identifiable information should be anonymized by default.
• The final synthesis should help a human product person make a better decision, not replace judgment.