Group users by a shared characteristic and track them over time. Beyond signup-date cohorts, five segmentation strategies turn cohort data into strategic decisions — not just retention curves.
▸ Try the interactive toolCohort analysis groups users by a shared characteristic and tracks their behaviour over time. Module 3 introduced it as the way to separate real change from composition change and to find the activation moment. This advanced treatment covers the different cohort dimensions that turn cohort data into strategic decisions — because which characteristic you group by determines what you can learn.
The default cohort is by acquisition date (signup month), which reveals whether the product is improving over time. But you can cohort by many other dimensions, each answering a different question: by acquisition channel (which sources bring users who retain?), by first action (does the activation path predict retention?), by plan or segment (do enterprise users behave differently from self-serve?), by behaviour (do power users retain differently?). The skill is choosing the cohort dimension that answers the strategic question you actually have.
By acquisition date — is the product improving over time?
By channel — which sources bring users who stay?
By first action — does the activation path predict retention?
By plan / segment — do segments behave differently?
By behaviour — how do power users differ?
Defaulting always to date-based cohorts answers only one question (is the product improving?) and leaves powerful strategic questions — about channels, activation, and segments — unasked.
| The shortcut | What it costs | What it gives you instead |
|---|---|---|
| Only date-based cohorts | Answers 'improving over time?' but misses channel, segment, behaviour insights. | Different cohort dimensions unlock different strategic questions. |
| Blended analysis | Mixing all users hides that segments behave very differently. | Cohorting by segment reveals the real per-group behaviour. |
| Not linking activation to retention | Missing which first actions predict who stays. | First-action cohorts reveal the activation paths that retain. |
| Channel blindness | Treating all acquisition sources as equal. | Channel cohorts show which sources bring users who actually stay. |
Decide what you actually want to learn — 'is the product improving?', 'which channel retains best?', 'does this first action predict retention?' The question dictates the cohort dimension.
Map the question to the right grouping: improvement → date; channel quality → acquisition source; activation → first action; segment behaviour → plan/segment; power-user patterns → behaviour.
Plot retention or the relevant metric for each cohort across periods. The comparison between cohorts is where the insight lives, just as with date-based analysis.
Read the differences: a channel cohort that retains far worse, a first-action cohort that retains far better. These differences are strategic signals — invest in what retains, fix or cut what doesn't.
Shift acquisition toward retaining channels; guide users toward the activating first action; tailor the experience for segments that behave differently. The cohort dimension turns data into a decision.
A team ran only date-based cohorts — useful for seeing the product improve over time, but blind to a question that mattered more: their acquisition was split across several channels, all treated as equally good because they all delivered sign-ups.
Cohorting by acquisition channel exposed a sharp difference: one channel's users retained well and another's churned almost immediately, dragging down the blended numbers. The blended view had hidden it entirely. Shifting spend toward the retaining channel — and away from the one bringing users who never stayed — improved overall retention without any product change, simply by acquiring better-fitting users.
The deliverable is cohorts grouped by the dimension matching your question — revealing channel, activation, or segment differences a blended view hides.
| Cohort by… | Answers | Decision it informs |
|---|---|---|
| Acquisition date | Is the product improving? | Whether changes are working |
| Channel | Which sources retain? | Where to spend acquisition |
| First action | Does activation predict retention? | What to guide users toward |
| Plan / segment | Do segments differ? | How to tailor the experience |
Cohort analysis is only as insightful as its grouping dimension. The advanced skill isn't running cohorts — it's choosing the dimension that answers the strategic question you actually have, rather than defaulting to signup date every time.
The trap is that date-based cohorts are the default everyone reaches for, and they answer a genuinely useful question (is the product getting better over time?) — which makes it easy to stop there and never ask the others. But the most actionable insights often live in other dimensions: a channel cohort that reveals you're paying to acquire users who never stay; a first-action cohort that reveals the single activation step that predicts long-term retention; a segment cohort that reveals enterprise and self-serve users need completely different experiences. Each is invisible in a blended or date-only view. The discipline is starting from the strategic question and letting it choose the cohort dimension — which is what turns cohort analysis from a retention chart into a decision engine.
One dimension, one question. Match the cohort dimension to what you want to learn.
Different segments behave differently; blending hides it. Cohort by segment.
Sign-ups aren't equal across sources. Cohort by channel to see who retains.
The activating first action is gold. First-action cohorts reveal it.
Cohort analysis is how retention curves (Tool 08) are segmented and compared.
This is the strategic-dimensions treatment of the cohort method introduced in discovery (Module 3, Tool 11).
Channel cohorts inform where to spend, connecting to LTV:CAC (Tool 15).
First-action cohorts identify the activating behaviour that drives retention.
Take three questions: 'is our product improving?', 'which channel brings users who stay?', 'does completing onboarding predict retention?' Match each to the cohort dimension that answers it.
Notice that only the first is answered by the default signup-date cohort.
If two of the three questions need a non-date cohort dimension, you've seen why defaulting to signup-date cohorts leaves most strategic questions unasked.
New tools and AI deep-dives, occasionally.
No spam. Unsubscribe anytime.