Data Analyze Refined Skill

outputs/

Core Principles

• Market perspective first: Explain the current state, opportunities, strategy, and risks in business language, avoiding academic expressions and model-oriented terminology.
• Automatically determine the analysis direction: When fields are missing, produce alternative insights rather than stopping because the data is incomplete.
• Charts follow insights: The number and type of charts are determined by the insights. Do not create charts merely for the sake of visualization.
• Conclusions first, details later: Begin with conclusions that management can easily understand, then present the supporting evidence and details.

Quick Usage

# 默认输出到数据同级 outputs/
node src/index.js /path/to/data.xlsx --sheet "导出数据表" --core-dimension "申报企业名称" --value-column "协议采购量" --charts-mode on

# 自定义输出目录
node src/index.js /path/to/data.xlsx --sheet "工作表" --output-dir "/path/to/outputs"

# 图表策略：auto/on/off（默认 auto）；截图需要时加 --enable-screenshot

• Python 3.8+ and

  pip install -r requirements.txt

  (pandas/numpy/matplotlib/seaborn/openpyxl/python-docx, etc.)
• Node 18+ is optional and used only for the Node.js entry point; run

  npm install

  first (the incorrect dependency has been removed). If npm reports an error, you can directly use

  python3 src/main.py ...

  .

• --sheet

  : Excel worksheet name (optional for CSV)

• --core-dimension

  : Core entity column (brand/company preferred; inferred automatically if unspecified)

• --value-column

  : Core metric column (purchase volume/quantity/total volume preferred)

• --target-brand

  : Optional target brand/applicant company used for white-space and opportunity analysis (if omitted, the leading brand or company name may be used)

Automatic Analysis Routing (Industry-Agnostic)

Workflow (Automatic Direction Detection → Market Insights)

1. Reading and preprocessing
   • Remove whitespace from column names and normalize aliases:

     品牌/企业/厂家/供应商 -> 品牌名称

     ;

     地区/地市/区域 -> 城市

     (for hierarchical fields, extract the city level according to the delimiter);

     目录/品类/类目 -> 产品大类

     .
   • Attempt to convert key numeric columns to numeric values:

     数量/销量/订单量/GMV/金额/合同量/需求量/消耗量

     .
2. Automatically identify core dimensions and metrics
   • Core dimension priority: brand/applicant company/manufacturer → hospital/channel/customer → city/region → other categorical columns with low uniqueness.
   • Core metric priority:

     --value-column

     > quantity/sales/order volume/GMV/amount/contract volume/demand > the first numeric column.
3. Noise-reduction rules
   • Use categorical TopN only when a column has 2–50 unique values and is not at least 90% unique; skip columns whose values are entirely identical or entirely unique.
   • Correlation requires a value of at least 0.3 and at least two numeric columns; otherwise, do not output it.
   • When there is no time or price column, explicitly state that trend or price-volume analysis is not being performed, and use structure, coverage, or opportunity analysis instead.
4. Analytical perspectives
   • Field profiling, numerical statistics, categorical distribution, and concentration (Top 1/3/5, number of entities required for 80%/90% coverage, and outlier thresholds).
   • Share and cross-tab analysis: brand/company, city, channel, category/SKU; brand × city, brand × channel, city × category, and top products within each brand.
   • Coverage and customer analysis: number of covered customers/average volume per customer for the core dimension, top customers, and target-brand customer white spaces (low share × high capacity).
   • Opportunities: target-brand white spaces (city total × target-brand share), with opportunity priorities calculated from impact × feasibility × investment efficiency. If time, price, or structural fields are available, add insights on price bands, structural mix, and time trends.
5. Output generation
   • CSV: core dimension/city/product category/major category/coverage, city-brand distribution, city Top 3, city white spaces, opportunity priorities (city/hospital or customer), product distribution, brand-product distribution, city-product distribution, top products within brands, customer TOP, customer white spaces, etc. (field overview/numerical statistics are for reference only).
   • Excel:

     outputs/excel/数据分析汇总.xlsx

     (table-of-contents sheet + links back to it, containing the business tables above; field overview/numerical statistics are excluded by default).
   • Charts: Generate PNG + SVG files in

     outputs/figures/{png,svg}

     based on charts-mode (core dimension share, city, product category, major category, coverage, etc.); automatically determine whether each chart is meaningful, and base the chart count on the number of insights.
   • HTML:

     outputs/html/market_analysis_infographic.html

     (core conclusions/opportunity summary + embedded core charts) and

     outputs/html/charts_gallery.html

     (chart gallery).
   • Word report:

     outputs/reports/<数据名>_分析报告.docx

     , written from the marketing team’s perspective with core conclusions, white spaces/opportunities, strategy, and an action plan; explicitly identify data gaps even when fields are missing.

Minimum Deliverables (Still Required for Sparse Data)

• Always output at least: a one-page conclusion section + at least one meaningful chart + at least one opportunity-list CSV.
• If there is no time column: remove trend analysis and strengthen structural, coverage, and opportunity insights.
• If there is no amount column: remove price bands and retain quantity share and concentration analysis.
• If there is no region column: remove regional white-space analysis and switch to channel/customer white spaces.
• If there is only one numeric column: do not perform correlation analysis; analyze distributions, shares, concentration, and TopN only.

Conclusion Page Structure (Required)

• Current-state interpretation: Key facts such as market size, structure, concentration, and coverage.
• Insights: Growth sources, structural mismatches, white-space/opportunity lists, and channel/regional differences.
• Future strategy: Actionable paths such as expanding coverage, deepening existing accounts, adjusting product mix or price bands, and reallocating budget.
• Risk analysis: Data gaps, structural mismatch risks, channel dependency, excessive concentration, and other relevant risks.

Chart Selection Rules (Insight-Driven)

• Share/concentration: Bar chart, stacked bar chart, or Pareto chart.
• Structure/price bands: 100% stacked bar chart or grouped bar chart.
• Opportunities/white spaces: Scatter or bubble chart (volume × share/coverage).
• Trends (only when a time column exists): Line or area chart.
• Channel efficiency/funnel: Funnel chart or conversion-rate bar chart.

Opportunity Priority Rules (New)

• Impact score: Potential incremental scale (total volume × share gap).
• Feasibility score: Existing foundation (target-brand volume or existing coverage), with share gap used as a proxy when necessary.
• Investment efficiency score: Approximates investment intensity; larger scale generally requires more investment and therefore receives a lower score.
• Overall priority score: Rank by combining “impact × feasibility × investment efficiency” and output high priority, medium-high priority, medium priority, or watchlist.

Marketing Team Report Highlights

• Core conclusions: Total volume, CR3/CR5, leading brands and shares, primary product categories/cities, and coverage (number of hospitals/cities covered and average volume per hospital).
• Products/structure: Top categories/SKUs and flagship products or mix within each brand when product or structural fields are available.
• Opportunities: Target-brand city white spaces and customer white spaces (low share × high capacity), supplemented with price, time, or structural insights when those fields exist.
• Action plan: Priority target list, long-tail/white-space pilots, product/mix/pricing tactics, and recommendations for additional fields such as time, channel, and price.

Quick Reference

Common Mistakes and Corrections

• Too academic: significance tests or regression models are used → replace them with business metrics and comparison criteria.
• Forced charts: charts are created without supporting insights → retain only charts that support the conclusions.
• Stopping when data is missing: omit output because time or price fields are unavailable → produce structural, coverage, and opportunity insights and disclose the gaps.
• Conclusion too brief: provide only one sentence → complete it using the “current state–insights–strategy–risk” structure.

Rationalizations and Corrections

Red Flag Checklist (Correct Immediately When Present)

• Using academic or model-oriented terminology instead of business language
• Fixing the chart count or using templated conclusions
• Stopping output because fields are missing
• Providing conclusions without strategy or risk

Example (Python)

# 根据字段可用性选择洞察与图表（示意）
fields = {"日期", "城市", "渠道", "产品", "数量", "金额"}
insights = []
charts = []

if "城市" in fields and "数量" in fields:
    insights.append("城市机会与白区")
    charts.append("城市份额条形图/白区散点")
if "渠道" in fields and "金额" in fields:
    insights.append("渠道效率与预算建议")
    charts.append("渠道ROI条形图")
if "日期" in fields:
    insights.append("增长来源与趋势")
    charts.append("时间趋势折线")

report_sections = ["现状解读", "洞察体现", "未来策略", "风险分析"]

Resources

• Code:

  src/main.py

  (pipeline),

  src/data_analyzer.py

  (preprocessing/analysis),

  src/chart_generator.py

  (charts),

  src/infographic_generator.py

  , and

  src/screenshot_generator.py

  .
• Reference:

  references/playbook.md

  (SOP, field mapping, and noise-reduction key points).
• Example structure:

  examples/

  (sample inputs and report-format references).