Cohort analysis and funnel analysis at a glance
A cohort analysis puts customers who share an acquisition event in the same period into one group, usually everyone whose first order or first subscription fell in a given month, and records what each group does at each age: the share still active, the revenue per original customer, the orders per original customer. Laid out with cohorts as rows and age as columns, the table shows whether newer customers behave better or worse than older ones at the same age. The cohort analysis tutorial builds that table from order or event data, plots the curves and reads the triangle; this page does not repeat the build.
A funnel analysis writes a journey as the steps a person passes through, arrive, view a product, add to basket, start checkout, pay, maps each step to the event that marks it, and reads the share who move from each step to the next. Splitting the funnel by device or source shows where one group loses more than another. The funnel analysis tutorial builds that funnel as an exploration in Google Analytics, including the choice between open and closed funnels, and turns the biggest leak into a hypothesis.
Google's help page on Explorations lists funnel and cohort explorations among the techniques available in Google Analytics, and the GA4 explorations tutorial covers the mechanics of both. The tools are next to each other in one menu; the questions they answer are not. The rest of this page is about matching the question to the analysis, and about the point where one is used to check the other.
Compare table
Cohort analysis vs Funnel analysis
| Row label | Cohort analysis | Funnel analysis |
|---|---|---|
| Unit grouped | Customers who started in the same period | People who reached each step of one journey |
| Time axis | Age of the group in days, weeks or months | None as such; the axis is the sequence of steps |
| Question answered | Do the people we acquire stay, return and pay, and is that changing? | Where in this journey do people drop out, and for whom? |
| Data needed | A stable customer identifier, an acquisition date and a record of activity per period | The events that mark each step, firing once and in order |
| Where it is built | A warehouse or spreadsheet on orders, or a cohort exploration for event-based groups | A funnel exploration, or any tool that can sequence events |
| How fast it reads | Slowly: a group has to age before its curve says much | Quickly: a few weeks of traffic show the step rates |
| Typical output | A retention or revenue triangle and one curve per cohort | Step-to-step rates, with the biggest leak highlighted |
| What it hides | Where inside a journey people were lost | What happened to people after the last step |
| How it misleads | A bigger or cheaper cohort changes the blended figure without any change in behaviour | A step can improve by letting through people who will never pay |
| Who uses it most | Growth, finance and retention teams setting lifetime value and payback | Conversion and product teams choosing what to fix on a page or flow |
The second and eighth rows carry the difference. A funnel has no clock: it reads one journey and stops at the last step. A cohort is nothing but a clock: it follows the same people for as long as there is data. So the funnel is the faster diagnostic and the cohort is the slower verdict, and each one's blind spot is the other's subject.
When to choose cohort analysis
Choose cohorts when the question is about people over time. Is month-three retention better for customers acquired this spring than last? Does a channel that looks cheap bring customers who lapse after one order? Has a pricing change altered what customers of every age do in the month it happened? Each of those needs groups held at the same age, which only a cohort table gives. The repeat purchase rate and the churn rate both make sense only when they are read for customers of the same age.
Cohorts are also the honest replacement for a blended retention figure. A single retention rate mixes customers of every age, so it rises whenever acquisition grows; the cohort tutorial shows why and how to define the measure as a specific cell instead. The lifetime value and payback inputs in the retention and payback guide come from the same curves, so a team setting what it can afford to pay for a customer needs the cohorts before it needs the funnel. When the question is where this month's change in the whole active base came from, rather than how each starting month ages, the growth accounting tutorial reads the same activity by calendar period.
When to choose funnel analysis
Choose the funnel when the question is about a journey that people go through in one visit or a short span: sign-up, onboarding, checkout, a lead form. The funnel shows which step loses the most people who should have continued, splits that by device and source, and compares one period with another, all within a few weeks of data. That speed is why conversion teams start there. The conversion rate calculator does the step arithmetic, and the funnel tutorial's method for estimating which leak is worth the most is the way to choose between steps.
A funnel is also the right tool when the cohorts show a problem and the team needs to know where it starts. If recent cohorts retain worse at the first month, a funnel of the first-use journey for those customers often shows the step that changed. The funnel locates; it does not explain, so the finding goes to recordings, surveys and a written hypothesis before anything is rebuilt, as the funnel tutorial sets out.
Decision table
| Situation | Use | Why |
|---|---|---|
| Checkout conversion fell last week | Funnel analysis | The step rates show where, and by device or source |
| A promotion brought a surge of first orders | Cohort analysis | Only a cohort shows whether those customers come back |
| Choosing which step of sign-up to test first | Funnel analysis | The leak arithmetic ranks the steps |
| Setting lifetime value and payback inputs | Cohort analysis | The curves are the inputs |
| A channel looks cheap per sign-up | Both | The funnel shows how its visitors convert; the cohorts show whether they stay |
| Retention is falling and nobody knows why | Cohort analysis, then a funnel | The cohorts date the change; the funnel finds the step |
| A redesign shipped and the step rate rose | Funnel, then the following cohorts | A better step is only a gain if those customers stay and pay |
| A new product with little history | Funnel analysis, while recording cohort data | Cohorts need time to age before they read |
Using both: checking a funnel fix in the cohorts it produced
The commonest trap in conversion work is a funnel fix that wins the step and loses the customer. Removing a question from sign-up, adding a discount at checkout or skipping a verification step can each lift a step rate by letting through people who would have dropped out for a reason. The funnel cannot see the reason; it stops at the last step. The cohorts can, because they follow the people the fix let through.
The way to check is to treat the change as a dividing line between cohorts. Cohorts that started after the change are compared with those that started before, at the same ages, on the measure the business cares about: the share still active, the repeat purchase rate, the revenue per original customer. If the later cohorts are larger and retain as well, the fix added customers. If they are larger but retain worse, the fix added people who were never going to stay, and the gain in the funnel may be a loss in the bank. Other things change between cohorts too, the channel mix most of all, so split the comparison by channel before crediting or blaming the change, as the cohort tutorial advises.
The AARRR framework gives the two analyses a common map. The funnel reads the steps inside acquisition and activation; the cohorts read retention and revenue for the same people. A team that sets one honest measure for each stage ends up using both analyses without having to argue about which is right. The comparison of acquisition and retention covers the budget question that follows once both are read.
A recipe app reads a sign-up change both ways
Pelvorne Recipes, an invented meal-planning app, cut its sign-up flow from five screens to two by dropping the questions about diet and household size. The funnel moved at once: the share of visitors who started sign-up and finished it rose from 38% to 61%, and monthly sign-ups rose from 4,200 to 6,700. The team then read the monthly cohorts of paying subscribers. Before the change, 22% of a month's sign-ups had started paying by the end of their first month and 64% of those payers were still subscribed at month three. For the first two cohorts after the change, 14% had started paying, and 51% of those were still subscribed at month three. Paying subscribers per month had barely moved, from 924 to 938, while support contacts asking why the plans did not fit their diet had doubled. The team restored the diet question as an optional screen after the first plan was shown, not before sign-up. The funnel's finish rate settled at 55%, and the next cohort's first-month paid share recovered to 20%.
Common mistakes
- Reading a funnel improvement as growth. Check the cohorts the change produced before calling it a gain.
- Using cohorts to find a broken page. A cohort can date a problem but cannot say which step caused it.
- Comparing cohorts at different ages. A young cohort's first month against an old cohort's sixth says nothing.
- Reading an open funnel as a sequence. People who entered mid-way make later steps look larger than earlier ones.
- Cohorts of sign-ups when the business sells to paying customers. The curve then measures activation, not retention.
- Ignoring the mix. A shift between channels changes both the funnel's step rates and the cohorts' curves; split by channel before drawing conclusions.
- Treating the two as rival reports. They answer different questions; the comparison of AOV and revenue per visitor shows the same pattern of a local gain that a wider measure can undo, and the comparison library lists other pairs that are easy to confuse.
