What is Retention Rate?

Retention rate is the share of a cohort of users or customers who, at a chosen moment, are still active or have done the target action again. It answers “who stayed.” The pair is churn rate (“who left”). With one activity definition and one base:

retention rate = 1 − churn rate

If the denominators differ (calendar base vs install cohort, “paid” vs “opened the app”), the identity fails. Then report both metrics; do not derive one from the other.

Formula on a customer base for a period:

Retention rate = Still active at period end ÷ Customers at period start

Do not add people acquired inside the period to the starting denominator — retention then “rises” from inflow. For one-off purchase products, “still active” is a rule: a repeat order in N days, a live subscription, a visit. Without a rule, retention is undefined.

Cohort, not “everyone alive this month”

Correct retention is almost always cohort-based: fix a start date (install, first order, FTD, trial start) and read that group’s share on day N or in calendar month M. That is cohort analysis: a curve, not one number on the whole base.

Calendar retention (“of January actives, who came back in February”) mixes veterans and new users and is a poor buying input. A buyer needs: cohort from channel A at D7 / day 30 vs channel B, at comparable CAC.

Link to LTV:

LTV ≈ Σ (ARPU in period t × cohort retention to t)

(plus margin, minus refunds). Without a retention curve, LTV is either the day-0 receipt or fiction. Early-day retention usually moves LTV more than the same percent in the tail: few people are left there.

D1, D7, D30 — slices of one curve

In apps and some web products, retention is read by cohort age. These are report sections, not separate entities and not reasons for extra slugs.

D1 (Day 1) — the share of the day-0 cohort that returns on the next calendar day (or within 24 hours — lock the rule). A signal of onboarding, creative–product mismatch, incent, and low-quality installs. Low D1 is rarely fixed by media mix if the store listing and first session stay the same.
D7 — a week: habit, or mobile junk after session one. For games and utilities, D7 is a working quality proxy before revenue matures. Compare D7 under one definition (“opened / event X”), not “opened” vs “paid.”
D30 — a month: a coarse product-market-fit and GEO filter. For subscriptions, trial-to-paid and billing retention matter more than D30 sessions.

Between the points: D3, D14, W1, M1. Same formula: active at the mark ÷ cohort size at start. Uninstall by D7 is not the same as “did not open on D7”: the second is usually lower and closer to session analytics.

Do not confuse D1 with bounce rate: bounce is a one-page visit; D1 is a return the next day.

Retention and churn — keep them aligned

Churn rate on the same cohort and the same “active” rule:

Churn rate = 1 − retention rate

Example: cohort of 2,000 installs, 280 active on D7 → 14% retention, 86% churn as a label on that point. That is not 86% cancel on a paying SaaS base in a week. Subscriptions are counted separately: billing retention = share whose charge succeeded / who reached rebill.

Logo retention and revenue retention diverge the same way as logo and revenue churn: whales stay — revenue retention is higher; everyone stays on the cheap plan — the opposite. Net revenue retention can exceed 100% with non-zero logo churn because of expansion.

In e-commerce, “D30 retention” with no order-in-window definition is empty: use repeat purchase rate of the first-order cohort. That is buyer retention, not session retention.

Practice in UA and affiliate

CPI and in-app: networks and MMPs report D1/D7 by campaign. Scale not raw IPM but cohorts whose D7 and event retention still support target LTV. Incent and rewarded often buy the install and kill D1 — visible in the cohort, not in yesterday’s installs.

Subscriptions and rebill: retention is the share that reaches the 2nd and 3rd charge. Day-0 CPA with falling billing retention inflates CAC per customer who actually lives. Hybrid CPA + revshare aligns incentives only in part; the curve is still required.

Ads can raise inflow and damage retention at once (broad interest, other GEO, creative promises what the landing lacks). Then start-of-funnel incrementality does not cancel a leaky LTV. The reverse: brand reach barely moves D1 but later lifts repeat — invisible in windowed CPA.

Limits

Attribution windows and privacy (ATT, SKAN) cut visibility of returns on the same finger that paid for the install. Comparing iOS and Android D7 without a caveat is a mistake. Analytics tools split days differently (calendar vs 24h) — a few points of D1 gap is normal, less so at D30.

Benchmarks like “D1 = 40%” without genre, GEO, and event definition are useless. Retention does not replace margin: retaining an unprofitable plan grows negative LTV.

Retention describes cohort life. The scale decision still closes on LTV > CAC with a buffer and, when the fight is “these people would have returned anyway,” incrementality — not a single D7 on a slide.