Market Insight Product Selection

An adaptive, evidence-first workflow for turning messy "trend" signals into a decision-ready shortlist using multi-source triangulation + VoC.

How to Use (Adaptive)

Use this skill to produce a repeatable selection pipeline from messy signals. If the user asks for "what to sell / trending products / hot selling products", follow this order:

1. Trend Radar → build a trend-driven candidate list (10–30) with evidence
2. Top Picks → rank and select Top3–8 with an explicit decision rule
3. Deep Dive Packs → deep analysis for each Top pick (as comprehensive as evidence allows)
4. Decision & Action Guide → Go / Cautious / No-Go + next steps

Core Principle: Evidence Density (Why Your "Reasons" Become Better)

• User intent overrides the skill: if any guidance here conflicts with what the user explicitly asked for (platform/channel, market, timeframe, constraints, format), follow the user request first.
  • Example: user asks for TikTok hot products → the primary candidate pool and examples should be TikTok products (TikTok Shop / TikTok trends / creator velocity).
  • You may still use other channels (YouTube/Reddit/Amazon/etc.) for supporting VoC / triangulation, but don't replace the requested platform with "defaults".
• Evidence first: don't write deep "reasons" until you have enough concrete evidence.
• Question-driven collection: collect enough information based on the user's question (e.g., action cameras: specs/capabilities + pricing/positioning + competitors + VoC).
• Concrete product/model deep dive requires a pack: whenever analysis touches a specific product/model/listing (user-provided OR discovered during category analysis), collect a Product Deep Dive Evidence Pack (below).
• If evidence is thin: explicitly output Data Gaps + Next Collection Step instead of inventing confident reasons.

Routing Shortcut

• Category trend → Demand & Scope → Radar → (optional) Deep Dive 3–8 products/models → Final
• Specific candidates → Deep Dive → Brain → (optional) Scorecard → Final

Dependencies

• matplotlib/seaborn for chart generation

Steps

1) Demand & Scope (Optional; recommended for broad category questions)

Purpose: Conduct deep, multi-source data analysis with cross-validation. Generate actionable selection recommendations with clear conclusion, analysis logic, and data evidence.

Data Collection

Maximize data source calls (≥3 sources):

• Web search: General market intelligence, news, and social phenomena
• Sales platforms: Amazon, Alibaba, Shopee, Shein (via web_search or specialized tools)
• Trend data: Google Trends, Amazon search trends
• Social media: YouTube, Reddit

Deep dive with web_fetch tool: When search results return promising URLs, use the web_fetch tool to extract detailed information from web pages (e.g., full article content, detailed product specs, in-depth reviews)

Deep Search Protocol

Core flow: Search → Extract entities → Search DEEPER → Check saturation → Continue until complete

⚠️ Deep search is NOT limited to 2-3 rounds. Continue until saturation signals detected.

Iterative Search Rounds

Round1 - Broad (3-6 queries): Break topic into dimensions (market size, consumer preferences, competition, technology, community feedback).

Round2 - Deep (2-4 queries): From R1 findings, drill into specific brands, technologies, pain points, suppliers.

Round3+ - Until saturation: New entities → explore | Unknowns → fill | Claims → verify | Contradictions → resolve

Extract & Map (Entity Extraction)

After each web_fetch, extract entities and relationships to identify next investigation points.

Entity extraction prompt template:

Extract from this content:
1. Key entities: products, brands, companies, technologies, suppliers
2. Relationships: [Product] uses [Component] supplied by [Vendor]
3. Next investigation: what should be explored based on these findings

Example: Found "Xiaomi Band 9 uses Goodix PPG sensor" → Next searches: "Goodix sensor specs", "Goodix competitors", "PPG sensor supply chain"

Contrarian Search & Source Quality

Contrarian search: When finding positive claims, search for opposing views to avoid echo chamber.

• Growth forecast → search "risks / challenges / bearish outlook"
• Product advantage → search "drawbacks / complaints / competitor strengths"

Source quality (prioritize high-quality sources for web_fetch):

Stop Conditions (Saturation Detection)

✅ STOP when ANY condition met:

• Saturation: No new entities/data for 2 consecutive rounds
• Closure: All initial unknowns filled and verified
• Verification complete: Key claims verified by 2+ sources

❌ DO NOT stop when: Important entities unexplored | Single-source claims | Unresolved contradictions

Minimum Requirements

Comprehensive analysis: expect 15-30+ searches, 20-40+ web_fetch.

Analysis Requirements

Cross-source validation: Connect insights from multiple sources to form structured judgments

• Example: "Google Trends shows 2x search growth + Amazon data shows $60-80 trail shoes have high ratings but limited supply + Alibaba shows only X suppliers = blue ocean opportunity with 40% margin potential"
• Example: "Search volume rising + Alibaba B2B has only X qualified suppliers = true blue ocean"

Trend trajectory judgment:

• State whether trend is emerging/explosive/mature/declining
• Identify if it's short-term hype or sustainable demand
• Note seasonality factors and 6-12 month outlook

Trigger event analysis: Identify sudden growth drivers (e.g., viral YouTube video, celebrity endorsement, KOL review, social media trend)

Product categorization (aggregate by sales/growth/reviews): The classification criteria are shown in the table below

Hit product feature extraction:

• Analyze common traits of top-selling products
• Price band distribution and gaps (identify blank price zones)
• Must-have features vs. nice-to-have features
• Most complained features

User pain point mining: Extract insights from reviews using NLP analysis when data volume is sufficient

Competitive landscape analysis:

• Market concentration (top players' market share)
• Brand vs. white-label ratio
• Entry barriers and differentiation opportunities
• Pricing strategy patterns across competitors

Supply chain feasibility assessment:

• Identify "easy-entry variants" based on supplier availability
• MOQ and pricing structure analysis
• Supplier concentration and risk assessment
• Margin potential calculation (e.g., "B2B cost structure suggests 40% gross margin for white-label")

Social & cultural trends:

• Lifestyle shifts affecting demand
• Sustainability/ethical consumption preferences
• Regional/cultural preference variations
• Macro trends influencing category evolution

Demand restatement: Include 1-2 sentences restating your understanding of buyer needs

Output Requirements

Text conclusions: Include reasoning chains that connect data to insights.

Visual evidence (include whenever possible to enhance user understanding):

• Images from search results: When tool results return images with reference IDs, display 3-6 representative images that best illustrate the analysis (e.g., trending styles, product examples, market snapshots)
• Trend visualization: Show how trending styles/colors apply to products
  • Example: "Q4 best-seller is multi-function contouring stick" → show product images
  • Example: "Next season's trending color is lemon yellow" → show color palette + application across product categories
• Code-generated charts (prioritize consolidation and significance):
  • Consolidate: Draw related curves/metrics on a SINGLE chart for comparison (e.g., multiple trend lines on one plot)
  • Simplicity: Do NOT generate charts for simple information; use tables or text instead
  • Types: Trend charts (multi-line sales/search growth), Price distribution (histogram/box plot), Category comparison (multi-variable radar/bar), Word cloud (review keywords)
  • Use matplotlib, seaborn, or wordcloud to create professional visualizations
• Hot-selling product display:
  • Hot-selling products returned by web_search or product_supplier_search
  • You MUST display top hot-selling products using product card widgets
  • Display at least 3-5 top-performing hot-selling products to help users quickly identify market opportunities

Comparison Table with Decision Notes (MANDATORY when ≥3 candidates):

• Category: Classify each product using the 4 types (Safe bets/High-potential/Red ocean/False trends)
• Decision Note: [positioning] + [evidence] + [risk/action]

Value explanation: Use tangible examples to help users understand product value (e.g., "This multi-function beauty tool replaces 3 separate devices, saving counter space and $50")

Case studies (when available):

• Provide "industry gold standard" examples as benchmarks
• Real marketing/viral phenomena (social media trends, KOL reviews, successful brand strategies)
• Actionable "story lines" or content directions users can reference and adapt

Source citations:

• Link to original web pages and data sources
• Reference to source sales data tables with key metrics

Selection Recommendation (provide structured recommendations with three components):

• Conclusion: Specific products/categories grouped by strategy (Safe Bets, High-Potential, etc.)
• Analysis Logic: Framework used (e.g., "Blue Ocean Model") + key indicators (supply gap, social buzz, search trend, complaint ratio)
• Data Evidence: Verifiable sources (TikTok views, Amazon listings, Google Trends index, Alibaba supplier count)

Step2: Product Search (Optional - SKIP unless user explicitly requests)

⚠️ Trigger condition: User explicitly asks for specific product cards/listings after seeing analysis results.

• Input: Selection recommendations from Step1
• Action: Search products using product_supplier_search (intent_type='product')
• Output Requirements:
  • ≥4 product cards per strategy dimension
  • MUST display clickable cards (not just text)
  • Comparison table for cross-evaluation

Step3: Supplier Search & Inquiry (Optional - SKIP unless user explicitly requests)

⚠️ Trigger condition: User explicitly mentions supplier/manufacturer/factory/vendor keywords. "Find products" or "trending products" does NOT trigger this step.

• Input: User explicitly requests suppliers or inquiry
• Action: Search suppliers using product_supplier_search (intent_type='supplier'), assess suppliers, draft inquiry emails
• Output: Supplier recommendations + draft inquiry

Step4: Summary & Action Guide (Mandatory)

• Input: All analysis completed
• Action: Synthesize findings into actionable recommendations
• Output: Executive summary table, opportunities/risks, specific next steps

Output Format

Analysis Output Structure

1. Demand Understanding
  - Restate user needs (1-2 sentences)

2. Market Analysis & Selection Recommendation (Step1)
  - Text conclusions with reasoning chains
  - Visual evidence (images, charts, hot-selling product cards)
  - Comparison Table with Decision Notes (categorization + decision note per product)
  - Selection recommendation (conclusion + analysis logic + data evidence)

3. Product Cards (Step2, only if user explicitly requests products)
  - Structured by recommendation dimension
  - ≥4 clickable cards per strategy dimension
  - Comparison table for cross-evaluation

4. Supplier Options (Step3, only if user explicitly requests suppliers)

5. Summary & Action Guide (Step4)
  - Opportunities & risks table
  - Specific next steps
  - Data limitations

6. Next Task Suggestions (if applicable)

Chart Design

Prefer charts when you have enough data: if the dataset is sufficiently rich (≥5 data points or ≥3 categories), include charts to improve clarity. The more data you have, the more you should rely on charts (and less on pure long text).

⚠️ CRITICAL: If you need to generate ANY chart, you MUST read the chart design guide FIRST:

• File path: ./chart-design-guide.md (relative to this SKILL.md)
• Why: Charts generated WITHOUT reading this guide will have poor styling, wrong chart types, and deprecated API usage
• Contains: Tool selection (Seaborn vs Matplotlib), code patterns, styling, chart type selection, Seaborn v0.12+ API updates

Checklist

✅ Strongly Recommended:

• For broad category questions: define segment + timeframe first (Demand & Scope), then build the candidate pool
• For specific products: include concrete product inputs (image, price, link/ID, variant) and do a full deep dive (not a quick verdict)
• Use multi-source evidence: YouTube + Reddit (VoC) + at least one trend/marketplace signal
• Always output Trend Candidate List → Top Picks → Deep Dive Packs → Final Recommendation
• For specific products: build a Specific Product Evidence Pack (listing + competitors + VoC + momentum) before writing "reasons"
• Use scenario routing (Red/Blue/Ghost) to decide validation depth
• In Red Ocean: base differentiation on recurring, fixable mixed-sentiment pain point patterns (benefit + complaint)
• Include sources in the final output

❌ Avoid:

• Treating all products with the same depth (violates "adaptive")
• Ignoring VoC (reviews/comments)
• Listing data without a decision rule