Customer research

Contents

File	Path	When to load
Research design method	${CLAUDE_PLUGIN_ROOT}/skills/product-design-shared/references/research-design.md	Hypothesis + null + MECE + findings
Pyramid composition	${CLAUDE_PLUGIN_ROOT}/skills/product-design-shared/references/pyramid-principle.md	Where findings get composed into arguments
Canonical Working Backwards reference	${CLAUDE_PLUGIN_ROOT}/skills/product-design-shared/references/working-backwards.md	Listen + Define stage context
Research methods	references/research-methods.md	Interview protocols, affinity clustering, qual/quant taxonomy
Customer-insights role	${CLAUDE_PLUGIN_ROOT}/skills/working-backwards/references/roles/customer-insights.md	Researcher agent for Listen-stage execution
Research plan template	templates/research-plan.md	Hypothesis, null, MECE questions, methods, findings (fill-in)
Problem statement template	templates/problem-statement.md	"Today [customers] have to..."
Customer journey map template	templates/customer-journey-map.md	Phase / action / touchpoint / thought / feeling / opportunity

Customer Research

The discipline of framing a testable hypothesis, gathering evidence via MECE sub-questions, and synthesizing findings into a shape the Pyramid Principle can compose. This is the research bottom-up half of Minto's core instruction — the upstream of working-backwards and any narrative composition.

When to Use

• You have a vague customer pain and need to know if it's real, at scale, and root-caused correctly.
• You need to produce a problem statement for a PR/FAQ or long-form narrative.
• You need to build a customer journey map for a design review.
• You have interview transcripts or survey data and need to synthesize them into decision-ready findings.
• You need to design a research plan before committing to a study (interviews, A/B test, survey).

When NOT to use:

• The customer problem is already well-framed and data-backed → go straight to skills/working-backwards/ or skills/product-discovery/.
• You need writing critique for an existing prompt or document → skills/meta-prompt-optimizer/.

Core Concepts

The method: research-design

The load-bearing reference is ${CLAUDE_PLUGIN_ROOT}/skills/product-design-shared/references/research-design.md. Five steps:

1. Hypothesis — one testable claim in one sentence.
2. Null hypothesis — the outcome that would disprove it. Pre-commit before gathering.
3. MECE sub-questions — 3–5 questions that collectively test the hypothesis without overlap.
4. Methods per question — behavioral, attitudinal, pain-point texture, segmentation, scale, causal.
5. Findings in Pyramid-base shape — per sub-question: Finding + Evidence + So what + Confidence.

The shape of your output is not optional. Pyramid-base-shaped findings feed pyramid-principle.md Phase 1 by construction.

Pyramid connection

This skill produces the Pyramid base. Downstream skills compose it:

• working-backwards uses the findings for the 5CQ's Q1, Q2, Q4 and the PR's Situation + Complication.
• product-discovery uses the findings for its research-scout role.
• A narrative-writing workflow of your choice composes a long-form document or one-pager from the findings.

Keeping findings in a structured shape means downstream composition becomes mechanical translation, not improvisation.

Three artifacts ship from this skill

Artifact	Template	When
Research plan	templates/research-plan.md	Before gathering — pre-commit the hypothesis and null
Problem statement	templates/problem-statement.md	After Define — the Listen output used by WB Stage 2
Customer journey map	templates/customer-journey-map.md	When you need to visualize current-state pain points across a flow

Quick Reference

A typical customer-research run:

1. Capture intent. What is the fuzzy question? Who is the primary customer segment?
2. Frame the hypothesis + null. Use templates/research-plan.md. Pre-commit in writing.
3. Decompose MECE. 3–5 sub-questions. Test: "Can I answer Q1 without answering Q2?"
4. Pick methods per question (see references/research-methods.md for the taxonomy).
5. Execute — either yourself for small corpora (synthesize from support tickets, survey data, past research) or via the customer-insights role of the researcher agent.
6. Synthesize per research-design.md step 7 — one finding per sub-question, each with Evidence + So what + Confidence.
7. Evaluate the hypothesis — Supported / Refuted / Inconclusive. Report honestly.
8. Produce the downstream artifact — problem statement, journey map, or hand off to working-backwards.

Composition with other skills

Upstream	Downstream
Raw customer data (tickets, interviews, surveys, analytics)	skills/working-backwards/ (Listen + Define + Invent)
User-research transcripts	skills/product-discovery/ (PRD personas + user stories)

Anti-Patterns

• Skipping the null hypothesis. Research becomes confirmation-gathering without it. Write it down before you gather.
• Non-MECE sub-questions. Overlapping questions muddle the findings. Re-decompose before executing.
• Using interviews to prove causation. Pick methods by question type — causal claims need experiments, not stories.
• Synthesizing during gathering. Forces confirmation bias into the evidence. Separate execution and synthesis in time.
• Reporting only supporting evidence. The antagonist test downstream depends on you seeing the counter-evidence.
• Moving goalposts when the null fires. When the null fires, document it — don't redefine the hypothesis to rescue the conclusion.

References

• references/research-methods.md — interview protocols, affinity clustering, qual/quant/subjective metric taxonomy, common VoC resource categories
• Shared methodology refs in ${CLAUDE_PLUGIN_ROOT}/skills/product-design-shared/references/:
  • research-design.md — the method
  • pyramid-principle.md — the downstream composition discipline
  • working-backwards.md — the end-to-end product-discovery process this skill plugs into
  • methodology-selection.md — when to use this skill vs. others