Cohort Retention Analysis

What is Cohort Retention Analysis?

Cohort retention analysis groups your users by when they started, most commonly the week they signed up, and tracks what share of each group comes back to use the product in the weeks after. A cohort is simply that group: everyone who signed up in the week of 1 September, everyone from the week of 8 September, and so on. Retention is the share of a cohort still doing the thing that matters in week 1, week 2, week 6.

Total user counts hide the truth. You can add 200 people a week and still have a product nobody keeps, because the growth at the top masks the leak at the bottom. Cohorts strip that away. They also let you compare: did the people who joined AFTER your big change stick around longer than the people who joined before?

When should you use it?

Use it once you have at least 4 weekly cohorts of 30 or more people each. Smaller groups swing too much for percentages to mean anything; with 10 people, one person is 10 percentage points.

• Fits: apps and SaaS, subscriptions, newsletters, communities, marketplaces (track buyers and sellers separately), online shops selling things people reorder.
• Use monthly cohorts: for products people need once a month, like invoicing or bookkeeping. Weekly cohorts would look like a disaster for a product nobody should open weekly.
• Skip: one-off purchases with no reason to come back (a wedding dress, a single course). Measure referrals and reviews instead.

How do you do it, step by step?

1. Choose the return event that equals value. Logging in is weak; people log in, look around and leave. Pick the action that shows the product did its job: “generated a meal plan”, “sent an invoice”, “booked a class”. This one choice decides whether your curve tells the truth.

2. Match the period to natural use. Ask how frequently a happy customer would use the product. Daily habit app: daily or weekly cohorts. Weekly planning tool: weekly. Monthly billing: monthly.

3. Build the table. In PostHog, create a Retention insight, set the start event (sign-up) and the return event (your value action), and pick weeks; the retention documentation shows how results appear as a cohort table plus a curve. In GA4, start a new exploration and choose Cohort exploration. In Google Sheets, export a list of user_id, signup_date, activity_date, add a column with =INT((C2-B2)/7) for weeks since sign-up, and build a pivot table: rows are sign-up weeks, columns are weeks since, values are COUNTUNIQUE of user_id.

4. Read the table in two directions. Read ACROSS a row to see one cohort fading over time. Read DOWN a column to compare cohorts at the same age: is week 4 retention better for September cohorts than for August ones? Down the column is where you see if your changes work.

5. Plot the curve and look for the flattening. Draw one line per cohort, with weeks since sign-up on the bottom and retention on the side. Every curve falls at first. The question is whether it levels out above zero. A curve that flattens at 25% means a quarter of people found a lasting reason to stay. A curve that keeps sliding toward zero means nobody did yet.

6. Mark your changes on the timeline. Note the week of each big release, price change or new channel next to the cohort it first touched. Without these notes, you will forget why the November cohort behaves differently.

7. Break it down by source. Compare cohorts from different channels. People from a niche newsletter may retain at twice the rate of people from a viral post, which tells you where to spend your next month.

What does a worked example look like?

Say Ines runs a web app that builds a weekly meal plan for parents of toddlers. Her return event is “generated a weekly plan”. After two months she builds weekly cohorts in PostHog:

The first two rows slide steadily toward zero. In mid-September Ines added an automatic shopping list sorted by supermarket aisle, and the 15 September cohort behaves differently: retention drops to 33% by week 3 and then barely moves. Reading down the week 3 column, 19% and 20% became 33% and 35%.

Her conclusion: the meal plan alone was a nice idea, and the shopping list is the reason to come back. She rewrites her landing page to lead with the shopping list and plans to ask her retained users the Sean Ellis product-market fit question next month.

Which free tools help?

What mistakes should you avoid?

• Using “logged in” as the return event. It flatters you. Pick the action that proves value.
• Picking a period that doesn’t match use. Weekly cohorts for a monthly product look like failure; monthly cohorts for a daily app hide a fast leak.
• Reading tiny cohorts as trends. A cohort of 12 that jumps from 30% to 50% is two people. Merge weeks into months until each cohort has 30 or more.
• Only reading across. The real signal for your product decisions is reading DOWN the columns, comparing cohorts at the same age.
• Counting free and paid users together. Paying users behave differently. Run a second table for paid retention once you have enough of them.

How do you know it worked?

The analysis worked when you can point at your curve and say whether it flattens, at what level, and which change moved it.

Our rule of thumb for go / no-go: if your best recent cohorts flatten at any level above zero for 3 or more periods in a row, you have a group of people who need the product. Find out who they are and get more of them. If every cohort keeps sliding toward zero after 8 weeks, more marketing will only fill a leaky bucket faster; go back to the people who left.

What should you do next?

Talk to the people who fell off with churn interviews, and ask the ones who stayed what they would miss with the Sean Ellis survey. Once retention flattens, plug the numbers into unit economics to see what each customer is worth. Founders in the Fe/male Switch startup game work through these numbers with an AI co-founder that asks what they mean for the business.

FAQ

What is a good retention rate for a startup?

It varies so much by product type that a single number misleads. A daily social app and a monthly invoicing tool live in different worlds. The useful test is the shape: a curve that flattens above zero beats a higher curve that keeps falling.

What is the difference between retention and churn?

Retention is the share of a cohort still active after a given time; churn is the share that left. For paying subscribers, monthly churn is the share of customers who cancelled that month. They describe the same leak from opposite sides.

Can I do cohort analysis in Google Sheets?

Yes. Export one row per user activity with the sign-up date and the activity date, calculate weeks since sign-up, and use a pivot table that counts unique users per cohort and week. Divide each cell by the cohort size to get percentages.

Updated 26 October 2026 · Written by the Fe/male Switch team

Mark this method as done