Retention calculator and cohort analysis

Retention is the share of a cohort still active after some time. A cohort is a group of users who joined at the same time: everyone who signed up in the same week or month. The classic formula for period N:

For example, of 1,200 users in the January cohort, 348 used the product in the third month after sign-up: month 3 retention = 348 / 1,200 = 29%.

N-day, unbounded and rolling retention

The single word “retention” hides several different metrics, and mixing them up is the main reason numbers do not compare. The definitions below match the documentation of the Amplitude and Mixpanel product analytics tools.

• N-day retention (classic, “exactly on day N”) — the share of the cohort active precisely on day (week, month) N. This is what the calculator computes. It is honest but noisy at daily granularity: a user who skipped day 7 is not counted even if they came on days 6 and 8.
• Unbounded retention (often called rolling retention) — the share of the cohort active on day N or later. The curve never rises and sits above the classic one, and recent values grow retroactively: a user who returns tomorrow raises the retention of all past days.
• Bracket retention — activity within chosen intervals such as “days 1–3”, “days 4–7”, “days 8–30”. Useful when the product’s natural usage rhythm is irregular.

If you paste unbounded retention data, the calculation is still correct, but you cannot compare it with other products’ classic retention.

A simple average of cohort percentages gives a small cohort the same weight as a large one. If a holiday month brought 200 random users with 10% retention and a normal month brought 5,000 with 40%, the simple average says 25%, although 2,020 of 5,200 users stayed — 38.8%. So the calculator uses a weighted curve:

“That have reached period N” matters: young cohorts could not have lived five months yet, and treating their empty cells as zeros would make the right side of the curve collapse artificially. The flip side: late points of the average rely on fewer cohorts and are noisier; the chart tooltip and the “Cohorts in average” row show how many cohorts stand behind each point.

Churn per period uses the same cohorts that reached period N — otherwise the changing cohort mix between neighboring periods would create false jumps.

Plateau

For a product that delivers value, the retention curve levels off after the initial drop: casual users are gone and the people who need the product remain. A flattening curve is widely seen as a sign the product has found its audience; if the curve keeps heading to zero, no amount of acquisition builds a lasting base.

The calculator reports as the plateau the first period after which the average curve loses no more than 2 pp and no more than 10% of its value per period, provided at least two more periods follow. It is a heuristic, not a standard definition: the thresholds suit monthly and weekly data, and on daily curves the plateau may be found later.

• Activity = login. If any visit counts as active, retention includes users who opened a reminder email and left. Count as active those who performed the product’s key action.
• Incomplete periods. The current month is not over yet, so its values are understated. Leave such cells empty.
• Mixing classic and unbounded retention in one report or when comparing with competitors.
• Comparing with an “industry benchmark”. A messenger, an accounting tool and a travel app cannot be measured with one ruler: their natural usage rhythms differ. Compare your own cohorts with each other — is the plateau rising for newer cohorts?
• Conclusions from one small cohort. A cohort of 50 people swings by several percentage points by pure chance.

Retention is decided in the first days: users who never reach the first key action almost always leave. That is why better onboarding often lifts not only early retention but also the plateau level — see Product onboarding: what it is and how to set it up and How to increase user activation in SaaS.

1. Amplitude Docs — Retention Analysis: N-day, unbounded and bracket retention (amplitude.com/docs).
2. Mixpanel Docs — Retention: first-time and recurring retention, cohort calculation (docs.mixpanel.com).
3. Croll A., Yoskovitz B. Lean Analytics: Use Data to Build a Better Startup Faster. O'Reilly Media, 2013 — cohort analysis and retention as a key product metric.
4. Fader P. S., Hardie B. G. S. How to Project Customer Retention. Journal of Interactive Marketing, 21(1), 2007 — cohort heterogeneity and why average retention rises over time.