Cohort: definition
A cohort is a group of users who share the same starting event in the same time window, most often signing up on the same day, in the same week or in the same month, and who are then measured together from that shared start. Comparing cohorts shows whether the users you acquire today do better or worse than the ones you acquired before.
What is a Cohort?
In product analytics, a cohort is a group of users who share a common characteristic, most commonly their sign-up date. A "March 2026 cohort" is all users who created their account in March 2026. Cohort analysis tracks this specific group's behaviour over subsequent time periods.
The power of cohort analysis isn't in looking at a single cohort in isolation; it's in comparing cohorts. If your March cohort retains better at 90 days than your January cohort, something you did between January and March improved retention. That's actionable information.
The word is much older than software. A cohort was a unit of a Roman legion, a body of soldiers who served together. Demographers borrowed the term for people born in the same period, and analytics borrowed it again for users who started at the same time. That is also what cohort means in business more broadly: a class of new hires, the customers you won in the same quarter, a group of trainees who began on the same day.
Cohort Date, Cohort Size and Other Cohort Terms
Cohort reports use a small vocabulary that is easy to mix up. These are the terms you will meet on almost every cohort chart:
| Term | What it means | Example |
|---|---|---|
| Cohort date | The date of the event that put a user in the cohort, usually the sign-up date. It labels each row of a cohort report. | Everyone who signed up in the week of 2 March 2026 is in the "2 Mar" cohort. |
| Cohort size (total users) | How many users the cohort had on day zero. It never changes, and it is the denominator for every percentage in that row. | 1,200 people signed up that week, so the cohort size is 1,200. |
| Period | The interval you measure in after day zero: day 1, week 2, month 3. Period 0 is the start itself. | "Week 4" means four weeks after sign-up, whatever the calendar date. |
| Return event | The action that counts a user as still active in a period. A login is the weakest choice; an action that delivers value is a better one. | Created a report, sent an invoice, completed a project task. |
| Cohort active users | How many users from the original cohort performed the return event in a given period. | 420 of the 1,200 created a report in week 4. |
| Cohort retention rate | Cohort active users in a period divided by the cohort size. | 420 / 1,200 = 35% retained in week 4. |
All numbers in this table are illustrative.
Daily, Weekly and Monthly Cohorts
The cohort window decides how many sign-ups are grouped into one row. A daily cohort is everyone who signed up on the same calendar day, a weekly cohort everyone who signed up in the same week, and a monthly cohort everyone who signed up in the same month.
Two questions settle the choice. First, how many sign-ups land in one window? A cohort needs enough users that one person returning or leaving cannot move its percentage by more than the change you are trying to detect. Second, how quickly do you need an answer? Daily cohorts show the effect of a launch within days but are noisy unless you have a high sign-up volume; monthly cohorts are steady but make you wait a month for every new data point. Weekly cohorts are the usual compromise for B2B SaaS products.
The window that groups users and the period you measure them in are separate choices, even though most reports use the same unit for both. The period should match how often a healthy user naturally comes back; our guide to building and reading a cohort table walks through both decisions.
Types of Cohort Analysis
Acquisition cohorts (most common): Group users by when they signed up. Used to track retention, activation rate, and lifetime value across sign-up periods.
Behavioural cohorts: Group users by what they did: users who completed onboarding vs. those who didn't, users who used Feature X in their first week vs. those who didn't. Behavioural cohorts are powerful for understanding which actions predict long-term retention.
Segment cohorts: Group users by an attribute such as plan type, acquisition channel or company size. Used to understand whether different customer types behave differently and deserve different onboarding or success motions.
Retention Cohorts: From a Cohort to a Retention Curve
A retention cohort is an acquisition cohort measured on a return event. Lay several retention cohorts out as rows, one per cohort date, with the periods as columns, and you have a cohort retention table. Each cell answers one question: what share of this cohort was still active this many periods after it started?
Take a single row and draw it as a line and you get that cohort's retention curve. Draw several rows on the same axes and the comparison becomes visual: if each newer line sits above the older ones at the same age, the product or the onboarding is getting better.
Each row of a cohort retention table is one cohort. Plot a row and you get that cohort's retention curve; newer cohorts have fewer cells because they have not aged yet.
Reading a Cohort Retention Table
A cohort retention table shows each cohort (rows) at different time intervals after acquisition (columns). Each cell shows what percentage of the original cohort was still active at that interval.
What to look for:
- Improving new cohorts: If recent cohorts show higher retention at day 30 than older cohorts, your product or onboarding improvements are working.
- The flatline: Where does the retention curve stop declining? This is your "core retained" percentage: the users who become genuinely dependent on the product.
- Early drop-off: A steep decline in week 1 almost always indicates an onboarding problem, not a product problem.
Cohort vs. Segment
The two words are often used as if they meant the same thing, but they answer different questions:
| Cohort | Segment | |
|---|---|---|
| Defined by | When users started (a cohort date or starting event) | Who users are, or what they are doing now |
| Membership | Fixed for good: nobody joins the March cohort in April | Changes as users change: an account enters "power users" the week it crosses the threshold |
| Question it answers | Is the product getting better for each new group of users? | How do different kinds of users behave? |
| Typical chart | Cohort retention table, retention curves | Breakdowns and funnels by segment |
The useful analysis usually combines them. Split a cohort by segment (the March cohort, self-serve accounts only) to see whether a change helped everyone or only one kind of user, and follow one segment across cohorts to see whether it is improving. Behavioural cohorts sit in between: "users who finished onboarding in their first week" is defined by behaviour, but the group is frozen once that first week is over. For defining segments in the first place, see our user segmentation guide.
Cohorts and Onboarding Changes
Cohorts are the cleanest way to tell whether an onboarding change worked, because a change only reaches the users who sign up after it ships. Compare the first cohort after the change with the last cohort before it, at the same age and on the same return event, and you get a before-and-after reading that an overall average would have blurred.
Two habits keep that reading honest: ship the change on a known date so the cohort boundary is clean, and wait until the new cohort has aged into the column you care about before calling the result. With Kompassify you can publish a product tour or an onboarding checklist to the segment you choose, on the day you choose, without an engineering release, then follow how each step performs. Week-one retention by cohort is also one of the twelve numbers in our list of user onboarding metrics.
Go deeper on the Kompassify blog
Longer guides on the ideas in this entry.
Frequently Asked Questions
See how each cohort moves through your onboarding
Kompassify shows how every product tour, checklist and survey performs step by step, and lets you publish a fix to the users who need it without an engineering release. No code required.
Start for free →Free up to 100 monthly active users · plans from $129/month · GDPR-compliant and EU-hosted