The cohort analysis every product team should be running — but isn't

The retention analysis most teams run — sign-ups grouped by month, tracked over time — is not wrong. It's just so aggregated that it's nearly useless for making decisions. Here's the cohort analysis that actually tells you what to fix.

Why the standard cohort chart is a category error.

The retention cohort chart is one of the most widely used visualisations in product analytics. It groups users by the period they signed up, then tracks what percentage are still active at defined intervals — day 7, day 30, day 90. Most product teams have one. Most of them look at it regularly. Most of them don't know what to do with it beyond "retention is declining" or "retention is improving."

The problem is not the chart. The problem is what's being grouped. Grouping users by sign-up month and calling it retention analysis is roughly equivalent to a hospital grouping patients by the month they were admitted and tracking mortality rates. The month of admission has almost nothing to do with the outcome. What matters is the diagnosis, the treatment, and the specific path through the system — not the calendar month.

When you group January sign-ups together, you are blending users who came from organic search (high intent, self-selected), paid social (low intent, interrupt-driven), a referral from an existing customer (contextualised, trusted), and a partnership integration (unknown job to be done). These four users have fundamentally different starting states, different expectations, and different probabilities of retention. Averaging them produces a number that is technically correct and practically useless for any intervention you might design.

The chart goes down over time. You know that. What you need to know is which users it goes down for, and why. Standard acquisition cohorts don't tell you that. Behavioural cohorts do.

The first-action cohort: where the insight lives.

A first-action cohort groups users not by when they signed up but by what they did first — specifically, the first meaningful product action they took in their first session. It is often the single most revealing cohort dimension in product analytics.

Here's why it matters. Every product has multiple entry paths into its core value. In a tool like Loom, a new user might record a video first, watch a video they received first, or browse templates first. In a tool like Notion, they might start from a blank page, import from another tool, or duplicate a template. In a tool like Webflow, they might start with a tutorial project, a blank canvas, or by cloning a community site. These are not equivalent experiences. They produce radically different first-session outcomes, and radically different long-term retention curves.

When you run a first-action cohort analysis on a mature product, you often find that the highest-retention first-action cohort retains at a multiple of the lowest. Users who took action A in their first session might retain at twice the rate of users who took action B. This is the most actionable insight in product analytics because it gives you a direct design target: route more new users toward action A in their first session, and your overall retention improves without changing anything else.

Take a file-sync product. You would expect users who install the desktop sync client in their first session to retain far better than users who only use the web interface — the desktop client embeds the product into their daily workflow, while the web interface keeps it an occasional destination. If the first-action cohort confirms it, the action is clear: redesign onboarding to make desktop client installation the primary first-session goal.

Twitter's growth team, under Josh Elman, found that users who followed 30 or more accounts in their first session retained at dramatically higher rates than those who followed fewer. This discovery drove the "suggested users" feature and its prominence in the onboarding flow. The insight came from a first-action cohort analysis — specifically, from correlating depth of onboarding actions with long-term retention curves.

The acquisition channel cohort: your growth budget's most important signal.

The second cohort analysis most teams aren't running is an acquisition channel cohort — grouping users by the channel that brought them in and tracking their retention curves separately. This analysis has enormous financial implications that most growth teams are ignoring.

When you run acquisition channel cohorts for a typical B2B SaaS product, you find a retention spread that makes the blended average almost meaningless. Organic search users — who arrived because they were actively searching for a solution to a specific problem — often retain far better than paid social users. Referral users (from other customers) typically retain well too. Content-attributed users (came through a blog post or guide) typically retain somewhere between organic search and referral. Paid search users sit closer to organic search than paid social. Display and retargeting users typically have the lowest retention of all.

This has direct implications for customer acquisition cost calculations that most finance teams are getting wrong. If you're using a blended average LTV in your CAC payback calculations, you are almost certainly overstating LTV for paid social and display, and potentially understating it for organic. This means you are:

Overspending on low-retention paid channels because they look profitable on a blended CAC/LTV basis, and underspending on organic content and SEO because the ROI feels slow even though the retained LTV is superior. When you disaggregate by channel and apply channel-specific LTVs to your acquisition spend, your channel allocation decisions change materially — often dramatically.

The plan-tier cohort: where pricing and retention intersect.

The third cohort most product teams aren't running is a plan-tier cohort — tracking retention separately for each pricing tier. This analysis surfaces something that almost every product with a freemium or multi-tier model is misunderstanding about its own retention economics.

In most freemium products, free tier retention curves are catastrophically bad and paid tier retention curves are materially better. This is expected. What's unexpected is what you find when you run retention cohorts by entry tier and track upgrade paths over time. Specifically: the free users who upgrade to paid in months two through four tend to have significantly higher long-term retention than users who signed up directly to paid at acquisition.

This seems counterintuitive. Surely direct-to-paid users are higher intent? They are — at the moment of sign-up. But the users who converted from free to paid did something more valuable: they used the product enough in the free tier to decide it was worth paying for. They've completed a self-selection process that filtered out users who were curious but not committed. Their retention is better because their commitment is based on demonstrated value, not on intent at sign-up.

Large freemium products like Canva illustrate the logic. Only a small fraction of free users ever convert, but a generous free tier functions as a long, low-friction retention filter: it lets users accumulate enough value experience to make a high-confidence paid commitment. Investing heavily in the free experience rather than aggressively gating features is a counterintuitive choice — and exactly the kind that plan-tier cohort analysis can support or refute.

Annual versus monthly plan retention cohorts are equally revealing and rarely tracked separately. Annual plan users churn at materially lower rates than monthly plan users in most subscription products, even allowing for selection effects. The reasons are partly psychological (sunk cost and commitment) and partly practical (the renewal decision happens once instead of twelve times). But the magnitude of the retention differential is consistently larger than teams expect when they first run this analysis.

How to build these cohorts if your analytics stack is basic.

None of the analyses I've described require enterprise analytics infrastructure. The inputs are three things: a user event log (what each user did and when), a user properties table (channel, plan tier, sign-up date), and a way to join them. If you have Mixpanel, Amplitude, or PostHog with basic event tracking, you can build first-action cohorts and plan-tier cohorts natively. If you're on Segment or have raw event logs in a data warehouse, a few SQL queries get you there.

The first-action cohort query is essentially: for each user, find the first event in the product with a meaningful action type (exclude page views, focus on feature-specific events), group users by that event type, then calculate retention at defined intervals for each group. The comparison between groups is your insight.

The acquisition channel cohort requires accurate channel attribution — which is where many teams hit a wall. UTM parameter hygiene is essential, and most products have gaps. If your attribution is imperfect, a useful proxy is to run the cohort analysis on users whose first session referrer is clean and clearly attributable. Even a partial dataset often produces clear enough directional signal to inform channel allocation decisions.

• 01
  Run first-action cohorts before you optimise anything else. Pull the first meaningful product event for your first 1,000 retained users and first 1,000 churned users. Compare the distributions. The action that appears most in retained users is your activation target — redesign onboarding to route everyone there.
• 02
  Disaggregate retention by acquisition channel immediately. If you're making growth budget decisions on blended LTV, you are making them incorrectly. Calculate channel-specific LTVs and rebuild your CAC payback models. This single change often reallocates a meaningful share of growth spend.
• 03
  Track annual versus monthly plan retention separately. Most teams know annual churns less — few have quantified the magnitude. Run this cohort and use the differential to inform how aggressively you discount annual plans at the point of conversion.
• 04
  Track free-to-paid conversion cohorts by the month they upgraded, not the month they signed up. Users who upgrade in month two have a different retention profile than those who upgrade in month six. The timing of the upgrade decision is itself a retention signal.
• 05
  Run your cohort analysis on revenue, not just users. A 60-day retained user on the growth plan is worth more than a 90-day retained user on the starter plan. Weight your cohort analysis by revenue to identify which user profiles to prioritise in both product investment and acquisition spend.
• 06
  Refresh cohort analyses quarterly, not annually. Product changes, channel mix changes, and market positioning changes all shift cohort behaviour. A first-action cohort analysis from eighteen months ago may be measuring actions that no longer exist in your onboarding flow. This analysis is perishable.

Frequently asked questions.