分析、增长与迭代
衡量真正能复利的事物:漏斗、群组、实验与留存 —— 别把忙碌当作进步。
Explainer
上线是「诚实回路」的第 0 天。获取可以掩盖留存问题;仪表盘可以一片绿,而 cohort 曲线却在衰减。Growth PM 痴迷于找出瓶颈阶段、验证实验是在推动持久行为而非新鲜感点击、并保护护栏指标免受隐性伤害。分析就是你证伪「乐观叙事」的方式。
从虚荣指标到决策指标
流量、下载、曝光、席位数、原始注册数 —— 它们让自尊膨胀的速度比让认知增长的速度快。优先选择与「可重复的用户价值创造」相关的指标:激活完整度、习惯性使用、按有意义细分的变现、cohort 留存的平台化。
- 为每个 KPI 指出它本周改变的决策。
- 把任何重要的东西按维度切分 —— 聚合指标会在某个细分里藏起背叛。
- 在趋势 + cohort 没有印证之前,把绝对数字当作假设。
AARRR:在打磨落地页之前,先找到漏水的那一段
Dave McClure 的海盗漏斗(Acquisition → Activation → Retention → Referral → Revenue)是助记符,不是魔法。增长工作从「相对基准,漏斗在哪里掉得最快」开始 —— 而不是从「情绪上最痛的地方」开始。
Cohort 分析与对幸存者偏差的谦逊
cohort 把共享某个有意义起始特征的用户(注册周、套餐、渠道)定格下来。比较 cohort 把上线狂欢与习惯性使用区分开。当你只回顾性地分析仍然留下的用户时,幸存者偏差就出现了 —— 它会美化 onboarding 的改动,同时掩盖 drop-off 的悬崖。
实验成熟度
当新鲜感效应淹没信号、事后扭曲比率、样本大小追求统计显著而非有商业意义的提升、跳过护栏、或一旦显著就缩短 runtime 时,A/B 测试就会失效。
增长系统
漏斗用来诊断;loop 用来复利。获取 loop(付费、病毒、SEO)、留存 loop(习惯、网络效应)和变现桥梁都应该被显式建模 —— 哪怕只是粗糙的图 —— 因为当瓶颈移动时,它们会改变你应优先做什么。
Framework atlas
Reference cards for each method in this mission
Expand a card for when to deploy it, misuse patterns, sequencing guidance, and (where relevant) shorthand formulas.
Funnel diagnostics · Dave McClureAARRR (Pirate Metrics)(AARRR)
Five macro stages spanning how users arrive, activate, retain, amplify, and pay. Use them to localize the weakest transition before prescribing tactics.
When to use
- Growth triage workshops.
- Explaining bottleneck thinking to executives.
When not to
- Discrete enterprise pipeline stages unrelated to activation self-serve funnels.
- When acquisition is sales-led and never touches a product surface.
How to apply
- Define stage boundary events with timestamps.
- Measure conversion ratios between neighbouring stages weekly.
- Benchmark against analogous products or historic internal cohorts.
- Ship experiments only at the current limiting stage.
- Advance focus when ratios normalize.
Engagement diagnosticsCohort retention curve
Plot % of cohort still active versus days/weeks since start. Healthy products often plateau; leaky products decay toward zero smoothly.
When to use
- Evaluating onboarding or activation changes.
- Comparing seasonal launches.
When not to
- One-off enterprise pilots with n<20 — noise dominates signal.
- When you cannot define a consistent activation event across users.
How to apply
- Pick a meaningful recurring action (Weekly Active, habitual session).
- Align cohort boundaries (signup week vs first value event).
- Compare curves before/after product changes overlaid—not just single-day retention KPIs.
Learning velocityExperiment planning canvas
Hypothesis statement, metric, minimum detectable lift, statistical power assumptions, segmentation plan, rollout + kill thresholds—before instrumentation.
When to use
- Growth teams juggling multiple concurrent tests.
When not to
- Qualitative discovery where quant tests would be premature.
- When legal/compliance forbids holding out a control cell.
How to apply
- Write reversible hypothesis sentences.
- Lock primary + guardrail metrics.
- Pre-commit runtime or sample—not optional peeking tweaks.
- Peek-and-stop inflates false positives dramatically.
Product Psychology
Cognitive biases that distort product decisions
Metric cherry-picking
Choosing the slice of data that validates the rollout while ignoring regressions elsewhere.
Product Risk
Ship regressions flagged by guardrail metrics silently while celebrating an irrelevant north-star blip.
Research Countermove
Mandatory guardrail dashboards and pre-registration of slices + metrics before launch.
Novelty effect blindness
Short-term uplift from UI freshness mistaken for sustained behavior change.
Product Risk
Roadmaps reorder around cosmetic wins that evaporate.
Research Countermove
Hold out cells; extend bake time; cohort users by exposure count.
Interactive lab
These instruments implement the textbook formulas loosely—use them to stress‑test judgments, compare frameworks on the same backlog, then document evidence and decisions.
← swipe to see all frameworks →
AARRR
Funnel bottleneck finder
Enter absolute counts across your AARRR funnel. Rates are naive ratios—use cohort definitions aligned to your product semantics.
Stage conversions
- Top of funnel reach (sessions) → Account created / signup8.40% (8,400 / 1,00,000)Tightest transition — inspect this constraint first
- Account created / signup → Hit activation milestone38.10% (3,200 / 8,400)
- Hit activation milestone → Return within window (cohort habitual)56.25% (1,800 / 3,200)
- Return within window (cohort habitual) → Successful referral fired12.22% (220 / 1,800)
- Successful referral fired → Converted revenue events409.09% (900 / 220)
Resources / Case Studies
Curated reading for this mission
Reforge / Brian Balfour
Deep dives on retention, loops, funnel math, channels, growth models, PMF checkpoints.
The canonical textual growth syllabus many PM influencers reference implicitly.
Amplitude
North Star + metric taxonomy frameworks, onboarding analytics narratives, taxonomy cookbooks.
Closes the gap between 'we track events' and 'we steer strategy with behavioural truth'.
Elena Verna (Substack RSS)
RSS feed for Elena’s operating notes bridging growth, experimentation, SaaS benchmarks.
Up-to-date operator perspective on pragmatic metrics & org design—not theory-only academia.
Casey Winters
RSS essays on marketplace growth loops, onboarding quality, experimentation culture.
Connects pirate metrics narratives to nuanced network effects realities.
Operator-written essays on monetization loops, onboarding, experimentation programs, hiring growth PMs.
Benchmark thinking for experimentation velocity beyond generic funnel charts.
Product Coalition
Medium publication RSS aggregating pragmatic PM narratives (strategy, experimentation, stakeholder craft).
Broad surface area perspectives beyond a single author's lens.
Evan Miller
Readable explanation of variance, statistical power calculators, pitfalls of naive significance.
Stops debates where people quote p-values without sample planning literacy.
Ron Kohavi et al.
Industrial-scale A/B learnings across Microsoft/Bing pedigree—novelty bias, surrogate metrics.
The reference when your org grows past spreadsheets into experiment platforms.