New format
Cohort analysis in marketing: what it is and why it matters
Cohort analysis compares groups of people who did the same action in the same period — first session, first purchase, install — and watches how a metric changes over time.
Below: definitions, how a cohort differs from a segment, and why cohorts matter for channels, LTV/ROI, and tests. Google Analytics Universal had a dedicated report; in GA4 cohorts live in Explorations and related reports — check tab names in the current UI.
Cohort vs segment
A cohort is people with one start event in a time window. Then you track one metric (retention, revenue, sessions) by cohort “age”: week 0, 1, 2…
A segment answers “who are they now by a set of properties.” A cohort answers “how do people who started then behave.” Both tools matter; mixing them up is a common brief mistake.
Analyst parameters: start event, cohort window size, observation horizon, comparison metric.
Example:
- segment: spent >$100 in January and live in London
- cohort: everyone with a first purchase in January (then watch repeat purchases by month)
Acquisition channels and retention
Build a “first visits in a month” cohort, slice by channel (paid search, social, organic, email). Compare the share who returned and bought in 1–4 weeks.
A channel with a cheap click but zero retention is often costlier than a slightly more expensive channel with repeat purchases. Cohorts show that better than yesterday’s report.
Weak channels get fixed or cut; strong ones scale carefully — without forgetting lead quality.
LTV, ROI, and payback over time
From first-buyer cohorts you compute cumulative revenue and group ARPU. Comparing cohorts from different start months estimates median LTV and room for acquisition cost.
Channel ROI in month one is often understated: the customer is still in the funnel. A monthly cohort view shows when payback catches up — important for long deal cycles.
The formulas stay the same — (revenue − cost) / cost; the value is the time and channel cut — not one “launch = success” cell.
Apps, A/B, and seasonality
For apps, an install cohort plus a source cut helps you not confuse “many installs” with living retention.
A/B gives instant variant conversion; cohorts of A and B users show whether the effect holds weeks later. Otherwise you may pick a design that flared and burned out.
Compare “arrived at New Year” vs “arrived in spring” cohorts: seasonality and promo promises change behavior.
Where to look and how not to overcomplicate
In GA4 — Explorations (cohort template, funnels, path), plus exports to Sheets or BI when needed. Don’t expect “Google Sheets will calculate everything” without your model.
Start with one retention metric and one channel cut. When the team can read the table — add LTV and tests.
Cohorts don’t replace daily CPA control: they’re a mid- and long-horizon view next to operational paid search.
Remember:
- one start event plus time
- a metric over time — not only “yesterday”
- compare channels by retention and LTV
- GA4 is not a Universal Analytics 2019 screenshot
- A/B plus cohorts beat either alone
FAQ
How is a cohort different from a segment?
A cohort is one unifying trait plus a start time (everyone who bought in January). A segment can combine many traits at once (spent >$100 and live in London).
Which action should start the cohort?
Whatever matters to the business: first visit, signup, first purchase, install. That choice defines what retention means.
Why use it for ads?
To see which channel brings people who return and pay later — not only a cheap first click.
Can I run it in GA4?
Yes, via Explorations and related reports. Don’t expect a one-to-one Universal Analytics screen from 2019 guides.
Do cohorts replace A/B tests?
No. A/B compares variants now; cohorts show how a chosen group’s behavior stretches across weeks.
Which cohort period should I pick?
Day, week, or month — by purchase frequency. Little data — go wider; lots of data — you can go finer.
Is LTV from cohorts 100% accurate?
It’s an estimate from history. New products and season shift the picture — refresh the calc.
Cheap clicks that never come back — and you only see “yesterday”?
We’ll build one first-visit cohort by channel and read retention/LTV over weeks — not just CPC.
Discuss the task
