Cohort Analysis

Tracking user groups over time to understand true behavioral patterns

⬢ ティア 2専門分野

+$15k-

給与インパクト

3か月

習得までの期間

普通

難易度

職種数

概要

Cohort analysis groups users by signup date or behavior and tracks retention/revenue metrics over time. Mature SaaS companies (Slack, Stripe, Mixpanel) use it to detect product improvements, identify declining user cohorts, and forecast LTV. Career path: Analyst (build cohort tables, $80-110k) → Strategist (predictive models, retention optimization, $110-150k) → Analytics Lead (company-wide cohort framework, $150-200k+) over 6-12 months. Built on time-series thinking, statistical controls for seasonality, and tools like Mixpanel, Amplitude, or SQL.

Cohort Analysisとは

Cohort analysis groups users by shared characteristics (usually signup date) and tracks their behavior over time. Unlike aggregate metrics that hide important trends, cohort analysis reveals whether your product is truly improving by comparing how different vintages of users behave. This is the most important analytical technique for SaaS, subscription, and marketplace businesses. It answers "are newer users retaining better?" and "is our product getting better over time?"

🔧 ツール＆エコシステム

MixpanelAmplitudeHeapGA4LookerTableauModeHexdbtSQL

📋 始める前に

Data Analysis Product Analytics

💰 地域別給与

地域	ジュニア	ミドル	シニア
USA	$85k	$125k	$175k
UK	£50k	£72k	£105k
EU	€55k	€78k	€115k
CANADA	C$90k	C$130k	C$180k

🎓 認定資格

Mixpanel Certified Reforge - Growth Analytics Amplitude Academy - Behavioral Analytics

🎯 Cohort Analysisを使う職種

Growth Hacker

Product Manager

Product Operations Manager

Retention Lifecycle Marketer

User Acquisition Manager Games

⚖ 比較

Churn Analysis Product Analytics Customer Segmentation

❓ よくある質問

Cohort vs Segment, what's the difference?

Cohort = group defined by a shared event date (e.g., all users who signed up in April 2025), tracked forward in time. Segment = group defined by a shared property at a moment in time (e.g., all users with plan = 'Pro' today). Cohorts show trends over time; segments show a snapshot. Use cohorts to measure product improvements ('did Nov cohort retain better than Oct?'), use segments to measure current state ('which plans churn fastest?').

Why do older cohorts always look better?

Survivorship bias. Older cohorts have already shed their worst users, the 30% who churn in week 1 are gone. Newer cohorts still have those early churners in the data. Fix: (1) track both absolute retention (% still active) and cohort size (how many dropped); (2) compare only cohorts mature enough to reach their natural churn floor (usually 12+ weeks); (3) segment by acquisition channel/quality within cohorts (paid users vs organic behave differently).

How do I account for seasonality in cohorts?

Year-over-year cohorts. Instead of comparing Jan cohort to Feb cohort (affected by New Year's resolution bias), compare Jan 2024 to Jan 2025. Also use rolling 4-week cohorts to smooth weekly noise. Be careful: a major product launch in March will create an obvious spike in retention for all March+ cohorts. Document the launch date and call it out in presentations.

Revenue cohorts vs user retention, which matters more?

Revenue cohorts matter more for SaaS. A cohort with 50% user retention but 80% revenue retention (high-paying users stick, low-payers churn) is far healthier than the opposite. Track both; prioritize revenue retention in your dashboards. LTV = driven by revenue cohorts, not user count.

My cohort table shows flat retention, does that mean my product isn't improving?

Not necessarily. Flat retention despite a major feature launch might mean: (1) the feature didn't matter for retention (focus on the right metrics), (2) it's offset by seasonality or market conditions, (3) early cohorts benefited, but newer cohorts haven't had time to mature yet. Compare pre/post feature cohort lifecycles at the same age (e.g., day 30 retention for Jan cohort vs day 30 for Feb cohort).

What tools should I use, Mixpanel, Amplitude, or SQL?

For SaaS: Amplitude (free tier, best-in-class cohort analysis, visualizations). For data teams: dbt + SQL (most flexible, version-controlled). For marketing: Mixpanel (easiest for non-technical users, good for event cohorts). For BI: Tableau/Looker dashboards on top of a data warehouse. Most mature companies use SQL + a BI tool, and Amplitude/Mixpanel for quick ad-hoc analysis.