market-research（市场研究）

Market Problems Deep Research

Research a target vertical's pain points using deep research APIs. Distill findings into a numbered hypothesis set. Output is pure industry education — no email generation, no company matching.

Environment

Provider selection and credentials are handled in Step 0 of the workflow.

Workflow

Step 0: Confirm provider and learn API

1. Ask the user which deep research provider they want to use. If they're unsure, Perplexity is a common choice — see workflow below for query design patterns.
2. Fetch or read the provider's API documentation and identify:

- Chat/completions or search endpoint - Available models (pick the one with web search / citations) - Authentication method and credentials - Rate limits

1. Ask for their API credentials and confirm access before proceeding

Step 1: Define the research scope

Read the company context file if it exists (claude-code-gtm/context/{company}_context.md) for ICP and existing hypotheses.

Ask the user for:

Step 2: Run hypothesis-driven research

Do NOT run generic research. Run 3-4 focused queries, each targeting a different angle of the same problem. The queries should be specific enough to return actionable data points, not overviews.

Query design principles:

• Each query should target ONE specific aspect of the pain
• Ask for concrete data points, numbers, timelines, tool names
• Ask for workflow descriptions, not abstractions
• Ask for failure modes and workarounds
• Keep queries vertical-agnostic in structure — the vertical comes from Step 1

Run each query through the chosen provider's API (from Step 0).

Standard 3-query framework:

Query 1 — Workflow pain: "What is the specific day-to-day workflow for [role] at [company type] when they [task we solve]? What tools do they use? Where do those tools fail? How long does each step take? Give concrete examples and data points."

Query 2 — Tool/database gaps: "How well do [existing tools] cover [target segment]? What percentage of the market do they miss? Why do [target companies] fall through the cracks? What data is wrong or stale? Give specific numbers."

Query 3 — Scaling problems: "What happens when [company type] tries to scale [process] beyond the initial [easy phase]? What breaks? What are the real-world failure stories? How do they work around it? What does it cost?"

Optional Query 4 — Industry leaders and public statements: "Who are the recognized thought leaders in [vertical]? What have they said publicly about [pain area] in the last 12 months? Include quotes, conference talks, blog posts, LinkedIn posts. Focus on practitioners, not analysts."

Step 3: Distill into numbered hypothesis set

Read all research responses and extract distinct, non-overlapping pain points. Each hypothesis should be:

• Specific: tied to a concrete workflow step, tool failure, or scaling problem
• Quantified: includes at least one data point (hours, percentages, dollar amounts)
• Verifiable: the recipient can confirm it from their own experience
• Non-obvious: teaches them something they may not have measured

Format:

## Hypothesis Set: [Vertical]

### #1 [Short name]
[2-3 sentence description with data points]
Best fit: [what type of company this applies to most]

### #2 [Short name]
...

Target: 5-7 hypotheses per vertical.

Step 4 (optional): Industry Leaders

If Query 4 was run, compile an industry leaders section:

## Industry Leaders: [Vertical]

### [Leader Name] — [Title, Company]
- **Public stance on [pain area]:** [summary of their position]
- **Key quote:** "[direct quote]" — [source, date]
- **Relevance:** [why this matters for outreach or positioning]

This section helps with:

• Email personalization (referencing what a leader said)
• Positioning (aligning with or contrasting industry voices)
• Content creation (informed takes on industry problems)

Step 5: Save outputs

Save to the vertical context directory:

claude-code-gtm/context/{vertical-slug}/sourcing_research.md   — full research output
claude-code-gtm/context/{vertical-slug}/hypothesis_set.md      — distilled hypotheses
claude-code-gtm/context/{vertical-slug}/industry_leaders.md    — leaders section (if Query 4 ran)

Create the directory if it doesn't exist.

Output Consumers

The hypothesis set is consumed by:

• enrichment-design — to design enrichment columns that score/confirm hypotheses
• list-segmentation — to match companies to hypotheses and assign tiers
• email-generation — to personalize P1 openers per hypothesis
• email-response-simulation — to evaluate whether email copy aligns with research

Relationship to hypothesis-building

hypothesis-building generates hypotheses from your own knowledge (context file + user input) — fast, no API. This skill validates and enriches those hypotheses with external research. If a hypothesis set already exists at claude-code-gtm/context/{vertical-slug}/hypothesis_set.md, use it to focus research queries instead of starting from scratch.

Typical flow: hypothesis-building first (define what you think) → market-research (validate with data). Or skip this skill entirely if you know the vertical well.

When NOT to Use This Skill

• If you already have a hypothesis set for the vertical — update it, don't recreate
• If you just need quick hypotheses from existing knowledge — use hypothesis-building
• If the user just wants to write emails — use email-generation skill
• If the user wants to find companies — use list-building skill
• If the user wants to enrich a table — use list-enrichment skill
• If the user wants to match companies to hypotheses — use list-segmentation skill