客户调研验证

Trigger conditions:

• User asks to validate a product idea, persona, or market assumption
• User mentions "customer research", "validate assumption", "talk to users"
• User requests Reddit/forum mining, competitor analysis, or sentiment analysis
• User wants to generate surveys or interview scripts
• User asks about customer pain points, needs, or jobs-to-be-done

Purpose

Pre-pipeline validation for DaVinci Enterprises products. Ensures marketing strategy is built on real customer signal, not assumptions. Prevents building features nobody wants.

What It Does

1. Reddit/Forum Mining — Extract threads, comments, sentiment from subreddits and forums
2. Survey Generation — Convert research questions into structured surveys
3. Interview Scripts — Generate customer interview guides with probing questions
4. Persona Validation — Test persona assumptions against real user behavior
5. Competitor Review Scraping — Aggregate reviews from G2, Trustpilot, Reddit
6. Sentiment Analysis — Aggregate and score customer sentiment across sources

Usage

Quick Start

# Validate a product hypothesis via Reddit mining
scripts/reddit-miner.sh --subreddit "personalfinance" --query "FIRE calculator" --limit 50

# Generate a customer interview script
scripts/interview-generator.sh --persona "FIRE enthusiast" --problem "retirement planning tools"

# Scrape competitor reviews
scripts/competitor-scraper.sh --product "Personal Capital" --sources "g2,trustpilot,reddit"

Integration with Marketing Pipeline

This skill feeds into the content strategy workflow:

1. Discovery → Run customer research to identify pain points
2. Validation → Test persona assumptions against real data
3. Strategy → Build content pillars around validated needs
4. Execution → Ogilvy creates content targeting real customer language

Output format: JSON reports to data/research/ for downstream consumption.

Scripts

reddit-miner.sh

Fetch Reddit threads matching keywords, extract sentiment, output structured JSON.

Usage:

./scripts/reddit-miner.sh --subreddit SUBREDDIT --query "search terms" [--limit N] [--sentiment]

Output: data/research/reddit-{subreddit}-{timestamp}.json

interview-generator.sh

Generate customer interview script from persona + problem statement.

Usage:

./scripts/interview-generator.sh --persona "description" --problem "pain point"

Output: Markdown interview guide to stdout

competitor-scraper.sh

Aggregate reviews from multiple sources, extract themes and sentiment.

Usage:

./scripts/competitor-scraper.sh --product "Product Name" --sources "g2,trustpilot,reddit"

Output: data/research/competitor-{product}-{timestamp}.json

Output Schema

All scripts output to data/research/ with consistent JSON schema:

{
  "meta": {
    "skill": "customer-research",
    "script": "reddit-miner",
    "timestamp": "2026-03-22T00:43:00Z",
    "query": {...}
  },
  "findings": [
    {
      "source": "reddit",
      "source_id": "thread_abc123",
      "text": "I wish there was a FIRE calculator that...",
      "sentiment": 0.65,
      "themes": ["pain point", "feature request"],
      "metadata": {...}
    }
  ],
  "summary": {
    "total_sources": 47,
    "avg_sentiment": 0.42,
    "top_themes": ["complexity", "cost", "trust"],
    "key_insights": ["Users want transparency", "Price sensitivity high"]
  }
}

Dependencies

• jq — JSON processing
• curl — HTTP requests
• Reddit API access (optional: can scrape public threads without auth)
• OpenClaw LLM access for sentiment analysis

Example Workflow

Scenario: Validate demand for FIRE Sim product

1. Mine Reddit pain points:

   bash

   ./scripts/reddit-miner.sh --subreddit "financialindependence" \
     --query "retirement calculator problems" --limit 100 --sentiment
   

2. Scrape Personal Capital reviews:

   bash

   ./scripts/competitor-scraper.sh --product "Personal Capital" \
     --sources "g2,trustpilot,reddit"
   

3. Generate interview script:

   bash

   ./scripts/interview-generator.sh \
     --persona "30-40 tech worker, $200K income, aiming FIRE by 45" \
     --problem "existing retirement tools too conservative or too complex"
   

4. Analyze findings:

   • Review JSON outputs in data/research/
   • Identify recurring themes, pain points, language patterns
   • Validate/invalidate persona assumptions
   • Feed insights into content strategy

5. Document learnings:

   • Update projects/davinci-enterprises/customer-insights.md
   • Flag validated needs for product roadmap
   • Inform Ogilvy content pillars with real customer language

Quality Gates

• Minimum sample size: 30+ sources per research question
• Sentiment confidence: Only report sentiment scores with >50 samples
• Theme validation: Themes must appear in ≥3 independent sources
• Source diversity: Mix Reddit, review sites, forums (not just one platform)

Anti-Patterns

❌ Don't:

• Build features based on one Reddit comment
• Cherry-pick data to confirm existing beliefs
• Skip competitor analysis (reinventing the wheel wastes time)
• Ignore negative sentiment (it's the most valuable signal)

✅ Do:

• Let data challenge your assumptions
• Track quotes verbatim (real customer language = gold for content)
• Cross-reference findings across sources
• Document what you disproved, not just what you confirmed

Integration Points

• Content Strategy: Feed validated pain points to Ogilvy for pillar creation
• Product Roadmap: Link research findings to JIRA/task tickets
• Persona Database: Update persona definitions based on validation results
• Marketing Copy: Extract customer language for landing pages, ads

Maintenance

• Research data retention: 90 days (then archive to cold storage)
• Re-run validation quarterly for active products
• Update scripts when Reddit/review site APIs change
• Log failed scrapes to logs/customer-research-errors.log

Next Steps After Running Research:

1. Review findings in data/research/
2. Update persona docs with validated/invalidated assumptions
3. Create content strategy tasks based on identified pain points
4. Schedule customer interviews if online research raises questions
5. Document learnings in project-specific CONTEXT.md