Claude Code Plugins

Community-maintained marketplace

Feedback

Product management data analysis skill for defining metrics, building funnels/cohorts, diagnosing growth/retention, designing experiments, and translating insights into product actions. Use for tasks like metric trees, event taxonomy, dashboard requirements, SQL-style analysis plans, and decision memos.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name data-analyse
description Product management data analysis skill for defining metrics, building funnels/cohorts, diagnosing growth/retention, designing experiments, and translating insights into product actions. Use for tasks like metric trees, event taxonomy, dashboard requirements, SQL-style analysis plans, and decision memos.

data-analyse

Use this skill for PM 视角的数据分析:把问题变成可量化指标与可执行动作。

Outputs (choose what the task needs)

  • Metric tree / North Star Metric decomposition
  • Event taxonomy(埋点口径:事件名/属性/触发时机/去重规则)
  • Analysis plan(要回答的问题、需要哪些数据、SQL/口径)
  • Funnel / cohort definitions(留存、转化、激活)
  • Experiment plan(假设、指标、样本量、guardrails)
  • Insight → action memo(结论、证据、建议、风险)

Workflow

  1. Clarify the decision
  • 业务问题是什么?要做哪个决策?“做/不做/怎么做”。
  • 时间窗口:今天要结论,还是一周内可迭代?
  1. Define metrics and guardrails
  • North Star Metric + supporting metrics.
  • Guardrails:错误率、退款/投诉、延迟、合规等。
  • 明确口径:分母/分子、去重、时间窗、过滤条件。
  1. Define data collection (events + properties)
  • 事件命名一致、可组合。
  • 属性尽量有限且可枚举;避免高基数字段(如 user_id 作为属性)。
  • 明确客户端/服务端各自负责哪些事件。
  1. Choose analysis methods
  • Funnel:识别掉点(按渠道/人群/版本分群)。
  • Cohort:留存与复购(D1/D7/D30)。
  • Segmentation:新老用户、付费/非付费、不同入口。
  • Causal thinking:区分相关与因果,避免幸存者偏差。
  1. Recommend actions
  • 针对最大掉点给 1–3 个高性价比改动。
  • 每个改动配:预期影响、实现成本、风险与验证方式。
  1. Experiment design (if needed)
  • 先写假设:如果 X 改为 Y,会让指标 Z 提升,因为 ...
  • 定义主指标、辅助指标、guardrails。
  • 明确随机化单位(用户/会话/企业)和实验周期。

Templates

埋点条目

  • Event: event_name
  • When: 触发时机
  • Props: key/type/allowed values
  • Dedup: 去重规则
  • Owner: client/server

Insight memo

  • Question:
  • Data:
  • Findings:
  • Interpretation:
  • Recommendation:
  • Risks:
  • Next measurement: