Unit Economics

intermediate📖 7 min read

Cohort Analysis

Also known as: Cohort Retention AnalysisCohort TrackingRetention CohortsUser Cohorts

Cohort Retention Rate = (Active Users in Cohort at Month N ÷ Total Users in Cohort at Month 0) × 100

💡

The Concept

Cohort analysis groups customers by their signup date (or another shared attribute) and tracks their behavior over time. Instead of looking at blended metrics that mask trends, you see how each 'class' of customers performs independently. A SaaS company with 5% monthly churn might discover that January cohort churns at 3% while March cohort churns at 9% — the blended 5% hides a deteriorating acquisition quality problem. Amplitude found that companies using cohort analysis identify retention problems 6-8 weeks earlier than those using aggregate metrics.

Real-World Example

Slack's growth team credits cohort analysis for preventing a potential retention crisis in 2015. They noticed that cohorts from Q3 2014 had 15% worse 90-day retention than Q2 cohorts. The blended retention number hadn't moved because Q2's large, healthy cohort was masking Q3's decline. Digging in, they discovered that Q3 signups came disproportionately from a viral growth hack that attracted individuals (not teams). Since Slack's value depends on team adoption, individual signups churned fast. They killed the growth hack and refocused on team onboarding.

⚠️

The Trap

The trap is treating all customers as one pool. Blended metrics create dangerous illusions: your overall retention might look stable at 85%, but if Q1 cohorts retain at 95% and Q4 cohorts retain at 70%, you have a ticking time bomb. By the time blended metrics show the drop, the damage has compounded for months. Another trap: analyzing cohorts too narrowly (daily) creates noise, or too broadly (annually) hides actionable trends. Monthly cohorts are the sweet spot for most SaaS businesses.

🎯

The Action

Build a cohort retention table: rows = signup month, columns = months since signup. Calculate retention rate for each cell. Look for two patterns: (1) Vertical drops — if a specific cohort has abnormally low retention, investigate what changed in acquisition that month. (2) Diagonal patterns — if ALL cohorts drop at month 3, you have an onboarding or value-delivery problem at that stage. Target: Month 1 retention ≥ 80%, Month 12 retention ≥ 50% for healthy SaaS.

⚡

Pro Tips

Overlay cohort retention curves on top of each other. If later cohorts have better curves, your product is improving. If curves are getting worse, your product-market fit may be eroding or your acquisition channels are attracting lower-quality users.

Revenue cohorts matter more than user cohorts. A cohort that retains 70% of USERS but 110% of REVENUE (via upsells) is outperforming a cohort that retains 90% of users but only 85% of revenue.

Run cohort analysis on engagement (not just retention). A cohort whose weekly active usage drops from 5 sessions to 2 sessions is signaling future churn even if they haven't canceled yet.

🚫

Common Myths

✗“Cohort analysis only matters for subscription businesses”

✓E-commerce companies like Amazon and Shopify merchants use purchase-frequency cohort analysis to track repeat purchase rates. Any business with repeat behavior benefits from cohort analysis — including marketplaces, gaming, and media.

✗“You need thousands of customers for meaningful cohort data”

✓Cohort analysis is useful with as few as 50-100 customers per cohort. The patterns (retention curves) are remarkably consistent even at small scale. Slack used cohort analysis when they had just a few hundred teams.

📊

Real-World Case Studies

💬

Slack

2014-2020

success

Slack used cohort analysis to discover that teams sending 2,000+ messages in their first month had 93% retention at month 12, while teams under 500 messages had only 15% retention. This insight drove their entire onboarding strategy: push teams past 2,000 messages ASAP. They built templates, tutorials, and channel suggestions specifically to hit this activation milestone.

2,000+ msgs Month-12 Retention

93%

<500 msgs Month-12 Retention

15%

DAU at IPO

12M+

Paid Customers at IPO

88K+

💡 Lesson: Cohort analysis doesn't just measure retention — it reveals the behaviors that CAUSE retention. Slack discovered the '2,000 messages' activation threshold and rebuilt their entire onboarding around it.

Source →

🎮

Zynga

2012-2015

failure

Zynga ignored deteriorating cohort quality for FarmVille 2 and Words With Friends. New game launch cohorts in 2013 had 40% worse day-30 retention than 2011 cohorts. But total DAU stayed flat because the massive installed base of loyal users masked the decline. By the time aggregate metrics finally showed the drop, they had lost 2 years of intervention time. Stock dropped from $14 to $2.

2011 Day-30 Retention

45%

2013 Day-30 Retention

27%

Stock Price Drop

$14 → $2

DAU Decline

72M → 28M

💡 Lesson: Aggregate metrics are a lagging indicator; cohort metrics are leading. Zynga's total DAU looked stable while each new cohort was dramatically worse. By the time DAU declined, the damage had been compounding for years.

📈

Industry Benchmarks

Month-1 Retention Rate

B2B SaaS (monthly cohorts)

Elite

> 90%

Good

80-90%

Average

65-80%

Needs Work

50-65%

Critical

< 50%

Source: Mixpanel 2024 Product Benchmarks Report

Month-12 Retention Rate

B2B SaaS (monthly cohorts)

Elite

> 60%

Good

45-60%

Average

30-45%

Needs Work

15-30%

Critical

< 15%

Source: Lenny Rachitsky's SaaS Retention Benchmarks, 2024

🛠️

Recommended Tools

📈Mixpanel

Product analytics — track user behavior, funnels, and retention cohorts in real time.

Free up to 20M events/mo

🔬Amplitude

Digital analytics platform — behavioral analytics, experimentation, and CDP in one.

Free up to 10M events/mo

📊ChartMogul

Subscription analytics — track MRR, churn, LTV, and cohorts automatically from your billing system.

Free up to $10K MRR

🎓

Master data analysis for cohort breakdowns

📉Google Data Analytics

Entry

8-course program covering spreadsheets, SQL, R, Tableau, and data-driven decision making.

$49/month (Coursera subscription)

via Coursera

📊Google Analytics

Entry

Proves proficiency in Google Analytics 4 — data collection, reports, attribution, and analysis.

Free

via Google Skillshop

🎮

Decision Scenario: The Cohort Quality Alarm

Your product has 10,000 MAU across all cohorts. Overall month-3 retention is 72%. But your data analyst just showed you this: Jan cohort: 82% month-3 retention, Feb: 78%, Mar: 74%, Apr: 68%, May: 61%. The trend is invisible in the blended metric but crystal clear in cohorts.

Total MAU

10,000

Blended Month-3 Retention

72%

Jan Cohort Retention

82%

May Cohort Retention

61%

Trend

Declining 4-5% per month

Decision 1

Marketing budget doubled in March when you hired a growth marketer who focused on paid social ads. The correlation is suggestive but not conclusive. Your growth marketer argues the 61% retention is 'normal for paid users' and the overall number is still healthy.

Trust the blended metric. 72% is above your 65% threshold. The new marketer is doing their job — growth is up.Click →

By August, the deteriorating cohorts dominate the mix. Blended retention drops to 58%. You've now lost 5 months of intervention time. Worse, the poorly-retained paid users left negative reviews, damaging organic acquisition. Recovery takes 6+ months of product and reputation repair.

Blended Retention (3 months later): 72% → 58%Intervention Window: Lost

Investigate immediately. Split cohorts by acquisition channel. If paid channels show 50% month-3 retention vs 85% organic, adjust targeting or kill underperforming campaigns.Click →

Analysis reveals paid social users have 48% month-3 retention (vs 84% organic). The growth marketer targeted too broadly. You refine targeting to lookalike audiences of your best organic users. June paid cohort improves to 68% retention. Blended metric stabilizes at 73%. Crisis averted through cohort-level diagnosis.

Paid Channel Retention: 48% → 68%Blended Retention: Stabilized at 73%

🧪