product-analyst
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseProduct Analyst
产品分析师
Strategic product analytics expertise for data-driven product decisions — from metrics framework selection to experimentation design and impact measurement.
为数据驱动的产品决策提供专业的产品分析支持——从指标框架选择到实验设计及影响衡量。
Philosophy
核心理念
Great product analytics isn't about tracking everything. It's about measuring what matters to drive better product decisions.
The best product analytics:
- Start with decisions, not data — What will you do differently based on this metric?
- Instrument once, measure forever — Invest in solid event tracking upfront
- Balance leading and lagging — Predict outcomes, don't just report them
- Make data accessible — Self-serve dashboards beat SQL queues
- Experiment before you ship — Validate hypotheses with real users
优秀的产品分析并非追踪所有数据,而是衡量关键指标以推动更优的产品决策。
最佳产品分析实践:
- 从决策出发,而非数据——基于这个指标,你会做出哪些不同的决策?
- 一次埋点,永久衡量——前期投入扎实的事件追踪工作
- 平衡领先与滞后指标——预测结果,而非仅做事后报告
- 让数据触手可及——自助仪表盘优于SQL排队等待
- 上线前先实验——通过真实用户验证假设
How This Skill Works
本技能的工作方式
When invoked, apply the guidelines in organized by:
rules/- — Frameworks (AARRR, HEART), KPI selection, metric hierarchies
metrics-* - — Conversion analysis, drop-off diagnosis, optimization
funnel-* - — Retention analysis, segmentation, lifecycle tracking
cohort-* - — Adoption tracking, usage patterns, feature success
feature-* - — A/B testing, hypothesis design, statistical rigor
experiment-* - — Event tracking, data modeling, collection best practices
instrumentation-* - — Visualization, stakeholder reporting, self-serve analytics
dashboard-*
调用本技能时,将应用目录下按以下类别组织的指南:
rules/- — 框架(AARRR、HEART)、KPI选择、指标层级
metrics-* - — 转化分析、流失诊断、优化
funnel-* - — 留存分析、用户分群、生命周期追踪
cohort-* - — 功能采用追踪、使用模式、功能成功度
feature-* - — A/B测试、假设设计、统计严谨性
experiment-* - — 事件追踪、数据建模、采集最佳实践
instrumentation-* - — 数据可视化、利益相关者报告、自助分析
dashboard-*
Core Frameworks
核心框架
AARRR (Pirate Metrics)
AARRR(海盗指标)
| Stage | Question | Key Metrics |
|---|---|---|
| Acquisition | Where do users come from? | Traffic sources, CAC, signup rate |
| Activation | Do they have a great first experience? | Time-to-value, setup completion, aha moment |
| Retention | Do they come back? | DAU/MAU, D1/D7/D30 retention, churn |
| Revenue | Do they pay? | Conversion rate, ARPU, LTV |
| Referral | Do they tell others? | NPS, referral rate, viral coefficient |
| 阶段 | 问题 | 关键指标 |
|---|---|---|
| 获客 | 用户从何处而来? | 流量来源、CAC(客户获取成本)、注册转化率 |
| 激活 | 他们是否拥有良好的首次体验? | 价值实现时间、设置完成率、“惊喜时刻” |
| 留存 | 他们是否会回访? | DAU/MAU(日活/月活)、D1/D7/D30留存率、流失率 |
| 变现 | 他们是否付费? | 转化率、ARPU(每用户平均收入)、LTV(用户生命周期价值) |
| 推荐 | 他们是否会推荐他人? | NPS(净推荐值)、推荐率、病毒系数 |
HEART Framework (Google)
HEART框架(谷歌)
| Dimension | Definition | Signal Types |
|---|---|---|
| Happiness | User attitudes, satisfaction | NPS, CSAT, surveys |
| Engagement | Depth of involvement | Sessions, time-in-app, actions/session |
| Adoption | New users/features uptake | New users, feature adoption % |
| Retention | Continued usage over time | Retention curves, churn rate |
| Task Success | Efficiency and completion | Task completion, error rate, time-on-task |
| 维度 | 定义 | 信号类型 |
|---|---|---|
| 满意度 | 用户态度、满意度 | NPS、CSAT(客户满意度评分)、调研 |
| 参与度 | 参与深度 | 会话数、应用内时长、每会话操作数 |
| 采用率 | 新用户/新功能的使用率 | 新用户数、功能采用率 |
| 留存率 | 持续使用情况 | 留存曲线、流失率 |
| 任务成功率 | 效率与完成度 | 任务完成率、错误率、任务耗时 |
The Metrics Hierarchy
指标层级
┌─────────────────┐
│ North Star │ ← Single metric that matters most
│ Metric │
├─────────────────┤
│ Primary │ ← 3-5 key performance indicators
│ KPIs │
├─────────────────┤
│ Supporting │ ← Diagnostic and health metrics
│ Metrics │
├─────────────────┤
│ Operational │ ← Day-to-day tracking
│ Metrics │
└─────────────────┘ ┌─────────────────┐
│ 北极星指标 │ ← 最关键的单一指标
│ │
├─────────────────┤
│ 核心KPI │ ← 3-5个关键绩效指标
│ │
├─────────────────┤
│ 支撑指标 │ ← 诊断性及健康指标
│ │
├─────────────────┤
│ 运营指标 │ ← 日常追踪指标
│ │
└─────────────────┘Retention Analysis Types
留存分析类型
┌───────────────────────────────────────────────────────────┐
│ RETENTION VIEWS │
├───────────────────────────────────────────────────────────┤
│ N-Day Retention │ % who return on exactly day N │
│ Unbounded │ % who return on or after day N │
│ Bracket Retention │ % who return within a time window │
│ Rolling Retention │ % still active after N days │
└───────────────────────────────────────────────────────────┘┌───────────────────────────────────────────────────────────┐
│ 留存分析视角 │
├───────────────────────────────────────────────────────────┤
│ N日留存 │ 第N天确切回访的用户占比 │
│ 无界留存 │ 第N天或之后回访的用户占比 │
│ 区间留存 │ 特定时间窗口内回访的用户占比 │
│ 滚动留存 │ N天后仍活跃的用户占比 │
└───────────────────────────────────────────────────────────┘Experimentation Rigor Ladder
实验严谨性层级
| Level | Approach | When to Use |
|---|---|---|
| 1. Gut | Ship and hope | Never for important features |
| 2. Qualitative | User research, feedback | Early exploration |
| 3. Observational | Pre/post analysis | Low-risk changes |
| 4. Quasi-experiment | Cohort comparison | When randomization hard |
| 5. A/B Test | Randomized control | Optimization, validation |
| 6. Multi-arm Bandit | Adaptive allocation | When speed > precision |
| 层级 | 方法 | 适用场景 |
|---|---|---|
| 1. 直觉判断 | 上线后听天由命 | 重要功能绝对不要用 |
| 2. 定性分析 | 用户研究、反馈 | 早期探索阶段 |
| 3. 观察性分析 | 前后对比分析 | 低风险变更 |
| 4. 准实验 | 群组对比 | 难以实现随机分组时 |
| 5. A/B测试 | 随机对照实验 | 优化、验证 |
| 6. 多臂老虎机 | 自适应分配 | 速度优先于精度时 |
Metric Selection Criteria
指标选择标准
| Criterion | Question | Good Sign |
|---|---|---|
| Actionable | Can we influence this? | Direct lever exists |
| Accessible | Can we measure it reliably? | <5% missing data |
| Auditable | Can we debug anomalies? | Clear calculation logic |
| Aligned | Does it tie to business value? | Executive cares |
| Attributable | Can we trace changes to causes? | A/B testable |
| 标准 | 问题 | 良好表现 |
|---|---|---|
| 可行动性 | 我们能否影响该指标? | 存在直接可控的杠杆 |
| 可获取性 | 我们能否可靠地衡量它? | 数据缺失率<5% |
| 可审计性 | 我们能否排查异常? | 计算逻辑清晰 |
| 一致性 | 它是否与业务价值挂钩? | 管理层关注该指标 |
| 可归因性 | 我们能否将变化追溯到原因? | 可进行A/B测试 |
Anti-Patterns
反模式
- Vanity metrics — Tracking what looks good, not what drives decisions
- Metric overload — 50 dashboards, zero insights
- Lagging only — Measuring outcomes without predictive indicators
- Silent failures — No alerting on data quality issues
- HiPPO-driven — Highest-paid person's opinion beats data
- P-hacking — Running tests until you get significance
- Ship and forget — Launching features without success criteria
- Segment blindness — Looking only at averages, missing cohort differences
- 虚荣指标——追踪看起来好看但无法驱动决策的指标
- 指标过载——50个仪表盘,零洞察
- 仅用滞后指标——只衡量结果而无预测性指标
- 静默故障——未设置数据质量问题告警
- HiPPO驱动——高薪者的意见凌驾于数据之上
- P值篡改——反复测试直到获得显著性结果
- 上线即遗忘——发布功能但未设定成功标准
- 分群盲区——仅看平均值,忽略群组差异