Churn cohort analysis: the one chart that tells you if retention is actually improving

Group your users by the week (or month) they signed up. That's the cohort. Plot, for each cohort, the percentage of users still active 1 week, 2 weeks, 4 weeks, 12 weeks, 26 weeks, and 52 weeks after signup. Each cohort becomes a line on the chart; time-since-signup is the x-axis. Simple shape, enormous information density.

You end up with a set of overlapping curves that drop off sharply in the first weeks and flatten over time. Each cohort can be compared to others at the same cohort age — "at week 4, the January cohort was at 42% retained; the April cohort was at 48%." That comparison is the entire point. Everything else on the chart is detail.

A note on granularity: weekly cohorts for fast-moving consumer products, monthly for slower-moving or B2B. Weekly is noisier but catches changes faster. Monthly is smoother but can hide a regression that happened three weeks ago. Most programs settle on monthly for executive review and weekly for the team's working view.

The first-week drop. Where most attrition happens. A cohort at 100% on day 0 and 40% by day 7 has lost 60% in the first week — that's the activation window. Improvements to onboarding and welcome flows show up here first, usually within four weeks of launch.

The 30-day curve shape. Continued steep decline through 30 days means activation stuck but engagement didn't. Flattening by day 14 means users who survived the first week are largely sticking. Two cohorts with the same 30-day retention can have very different shapes, and the shape is what tells you where the program is working and where it isn't.

The asymptote. The flat line cohort retention approaches over months. Cohorts flattening at 15% means you have a 15% "committed user" base that rarely churns. Cohorts that keep declining past month 6 mean ongoing attrition even among established users — which points at product value or re-engagement, not onboarding.

Cohort-to-cohort comparison. Stack recent cohorts against older ones at the same age. Newer above older means improvement. Newer below older means regression. This is the single most valuable comparison on the chart and the one most programs forget to make explicit.

Worth saying plainly: industry benchmarks for retention are nearly useless because they vary wildly by category, and comparing your product to a public company with a totally different business is a recipe for bad decisions. The honest comparison is always cohort-to-cohort within your own program.

A cohort chart by week of signup is the default. Richer analysis stratifies the cohorts by:

Acquisition channel. Paid social cohorts typically retain worse than organic. Referred users typically retain best. If blended retention is stable but channel mix shifted toward paid, your real retention is declining and the blend is politely hiding it. That's the quietest way a lifecycle program can lose ground without anyone noticing.

First-action experience. Users whose first meaningful action was X versus Y often show wildly different retention curves. This is where product-led retention wins come from — find the "aha moment" action and shape the first-week flow around reaching it.

Geography or plan tier. If your product varies meaningfully by region or plan, cohorts along those dimensions show where to invest. Blended "just fine" retention often hides one high-retention segment funding a low-retention one.

The aha moment guide covers how to identify the first-action that drives retention. Once you find it, that's usually the primary cohort stratification for the rest of the program's life.

It also misses resurrections. A user who signed up in January, went dormant in February, re-engaged in April is absent from the January cohort's week-12 retention but present in the quarterly active-user count. If your win-back or sunset work is producing meaningful resurrection, a separate "resurrection cohort" chart captures it cleanly; stacking it alongside the retention curve is the honest picture.

Common anxiety worth pre-empting: "why are newer cohorts sometimes worse than older ones?" Three suspects in rough priority order — acquisition-channel shift toward lower-quality traffic, a product regression between cohorts, and seasonal effects. Stratify by channel to rule out the first, annotate the timeline with product changes to identify the second, and compare year-over-year to isolate the third.

Most modern analytics stacks support cohort analysis natively — Amplitude, Mixpanel, Heap all have the view built in. For SQL-native teams, the query is about 30 lines: one CTE for signups grouped by cohort week, another for activity events, a JOIN on user_id with a time-since-cohort calculation. Not complicated, just finicky to get right the first time.

Pick one "active" event definition and stick with it across every cohort. "Active" should mean one specific thing — e.g., "user performed [key product action] in the week ending [date]". Something meaningful, not a login. For a marketplace that might be "viewed a product". For SaaS, "performed core action X". For content, "read an article". Logged-in is too loose; purchased is usually too conservative for early-stage retention. And shifting the definition between runs produces curves that cannot be compared, which defeats the chart's only purpose.

Worth noting a useful variant: cumulative revenue per cohort. Instead of "percent still active at week N", plot "cumulative revenue per user at week N." Flattens tell you revenue saturates. Continued steep climb tells you users keep spending. Pair with user retention to separate a stable-revenue base from a growing-spend-per-user dynamic. For subscription businesses, these two charts together are basically the whole board-report.

The quarterly business review is where cohort curves earn their keep. The structure that works:

1. Current quarter cohorts stacked against the prior four quarters, at the same cohort ages.

2. Highlight any cohort meaningfully above or below trend. Discuss what changed.

3. For cohorts above trend: what did we do that we should keep doing?

4. For cohorts below trend: what happened, and what's the remediation?

5. Decide next quarter's priority lifecycle work based on where the curves say leverage lives.

One more timing consideration worth calling out because it catches people: different levers show up at different cohort ages. Onboarding improvements read in week-1 retention within four weeks. Win-back changes read in 90–180-day retention within 3–6 months. Systemic retention work — product value, re-engagement programs — needs 6–12 months of cohort data before you can confidently say the curve moved. Build the review cadence around that reality, not around the quarterly calendar.

uses cohort curves as the foundation of quarterly roadmap decisions. Programs that don't look at cohorts drift into reactive, campaign-level thinking within two quarters. Every time.