Cohort Analysis & Retention
A cohort is a group of customers who started in the same month — tracking each cohort over time shows whether people actually stick around or quietly leave.
What you will learn
- Define a cohort and explain why grouping by start date matters
- Read a retention cohort table
- Tell a healthy retention curve from a leaky one
The number totals hide a leak
Imagine an app that proudly reports "10,000 active users" every month, a flat, steady line. It looks stable. But underneath, it might be losing 4,000 old users and replacing them with 4,000 new ones each month — a bucket with a hole that you keep refilling. The total hides the leak. Cohort analysis is how you find the hole.
A cohort is a group of customers who share the same starting point — usually the month they first signed up or first bought. The "January cohort" is everyone who joined in January. You then follow that exact group month after month and watch how many stay.
What "retention" means
Retention is the percentage of a cohort that is still active after some time. If 1,000 people joined in January and 400 are still active three months later, three-month retention for that cohort is 40%. The opposite is churn — the percentage who left (60% here).
Reading a cohort table
A cohort table puts each starting group on its own row and time across the columns. Each cell is the percentage of that cohort still active. Here is a meal-kit service:
| Cohort (joined) | Month 0 | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| January (1,000) | 100% | 55% | 42% | 38% |
| February (1,200) | 100% | 58% | 47% | 44% |
| March (1,500) | 100% | 64% | 55% | — |
Read it two ways. Across a row: the January cohort fell from 100% to 38% over three months — most of the drop happened in the first month (100% to 55%). Down a column: compare Month 1 across cohorts — 55%, then 58%, then 64%. Newer cohorts are retaining better, which means recent changes (maybe a better onboarding email) are working.
January cohort journey:
Month 0: 1,000 active (100%)
Month 1: 550 active (55%) <- biggest drop is here
Month 2: 420 active (42%)
Month 3: 380 active (38%)
The fall flattens after Month 1 -> the people who stay past
month 1 mostly keep staying.Note: The steepest loss is between Month 0 and Month 1 — over half leave almost immediately. That points your attention firmly at the first month experience (the first delivery, the welcome flow). Fix that one leak and the whole curve lifts.
Healthy curve vs leaky curve
Every retention curve falls at first — some people always leave. What matters is whether it flattens (good) or keeps sliding to zero (bad):
| Shape | What it means | Health |
|---|---|---|
| Drops, then flattens into a plateau | A loyal core sticks around long-term | Healthy |
| Keeps falling toward 0% | No one stays; you must constantly refill | Leaky — fix retention first |
| Newer cohorts retain better than older | Your recent product/onboarding changes worked | Improving |
This is why retention beats raw growth for a long-term business: pouring more money into acquisition to fill a leaky bucket just loses money faster. A flat retention plateau means every new customer adds to a growing base instead of replacing a lost one.
Where to find it
You do not have to build cohort tables by hand. GA4 Explorations has a built-in Cohort exploration that groups users by the week or month they first visited and shows retention automatically. Subscription tools and spreadsheets can do the same. The skill is not making the table — it is reading it and acting on the leak it reveals.
Tip: Always look at the first period drop most closely. For most businesses the largest churn happens immediately after signup, so the biggest, cheapest wins come from improving the very first experience — the first email, first order, first login.
Watch out: Do not compare a brand-new cohort’s short history against an old cohort’s long one and panic. March only has data up to Month 2; judging it against January’s full Month 3 figure is unfair. Compare cohorts at the same age (Month 1 vs Month 1).
Q. A retention curve drops sharply for the first month and then stays roughly flat for many months. What does that flat part tell you?
✍️ Practice
- From the cohort table, calculate the Month-1 churn (percent who left) for the February cohort.
- A cohort goes 100% -> 30% -> 12% -> 4%. In one sentence, describe the health of this retention and what you would fix.
🏠 Homework
- For an app or subscription you use, sketch a 3-month retention cohort table with made-up but realistic percentages. Mark the biggest drop and write one idea to reduce it.