Retention

Retention measures how much of a defined customer relationship continues over a stated interval. The relationship might be an active subscription, a repeat purchase, or a meaningful product action. It complements churn only when both metrics use the same population, time interval, and rules for leaving or returning. Customer counts, product activity, and retained revenue answer different questions.

Customer retention rate follows customers who were present at the start of a period. A common account-count calculation is:

\[ \text{CRR} = \frac{E - N}{S} \times 100 \]

with \(S\) customers at the start, \(E\) at the end and \(N\) acquired during the period and still present at its end — start with 100, lose 10 of that cohort and retain 20 newcomers at period end, and CRR is 90%. Counting every acquisition, including newcomers who already left, would understate retention of the starting cohort. Cohort retention follows one group over time: 100 users sign up, 40 are still active in week 4, so week-4 retention is 40%. Revenue retention measures money rather than people, as gross revenue retention, which ignores expansion, or net revenue retention, which includes upsells and cross-sells.

These three are often quoted interchangeably. They are not interchangeable, and the gaps between them are large enough to change what a chart says.

Why a flat rate can hide a shrinking cohort

Suppose each monthly starting cohort has a 90% probability of remaining subscribed for the next month, with no reactivation. Monthly retention stays at 90%. The fraction of the original cohort expected to remain after six months is 0.9⁶, approximately 53.1%. The two rates use different horizons and denominators. New subscribers can replenish the total customer count without restoring customers lost from that original cohort.

Now consider 100 subscribers who initially pay 100 units each per month. If 65 remain and each now pays 170, customer retention is 65%, while revenue from the starting cohort is 65 × 170 = 11,050 compared with 10,000 initially. Net revenue retention is therefore 110.5%. With no downgrades among the survivors, gross revenue retention is 65%. Expansion has offset lost revenue, but it has not reversed the loss of customers.

Choose the event and observation window

For product usage, define the starting event and the return event. Signup followed by any page view measures something different from completing a purchase followed by another purchase. Choose the event that represents the relationship of interest, and decide whether the unit is a person, account, or subscription. Multiple devices and duplicate identities can change the denominator if identity rules are inconsistent.

An exact-period metric counts a return within a particular interval after entry. A return-on-or-after metric counts a return in that interval or later within the observed window. A person returning in week six can count for on-or-after week-four retention while not counting for exact week-four retention. Amplitude documents these distinct modes in Interpret your retention analysis. Record which convention a chart uses before comparing it with another chart.

Customers who joined two weeks ago have not had an opportunity to reach week four. Do not count them as week-four failures. Compare cohorts at the same age, state the eligible denominator at each age, and define calendar versus elapsed-time boundaries. For subscriptions, separately state how pauses, payment failures, cancellations, and reactivations affect eligibility and retained status.

Keep revenue retention tied to the starting base

Let starting recurring revenue be R, lost revenue from cancellations be C, revenue lost through downgrades be D, and expansion from the starting base be X. Under a convention that excludes reactivation, gross revenue retention is (R − C − D)/R and net revenue retention is (R − C − D + X)/R. New-customer revenue is excluded. Specify recurring-revenue units and whether reactivation is included; ChartMogul’s Net MRR Retention documentation illustrates why configuration changes can alter the result.

Show customer retention alongside gross and net revenue retention when both loss and expansion matter. Net revenue growth may reflect upgrades, price changes, or increased usage among survivors. It cannot establish that more customers stayed or that their experience improved.

Why it matters, and how to report it

Retention can reveal whether a relationship persists, but it does not identify the cause. Contracts, switching costs, seasonality, and the selected activity definition can affect a curve. A flat observed tail is not proof that customers will stay forever or that a product meets every customer’s needs. Compare relevant cohorts and inspect the underlying behavior before attributing a change to product quality.

Use period customer retention to examine losses from the starting base, and cohort curves to follow comparable customers as they age. When estimating LTV, distinguish observed retention from the future survival assumption. Similar early curves can diverge later; cohort-based LTV still needs a justified forecast beyond the observed period. Report sensitivity to that forecast instead of treating a guessed retention floor as measured value.

Models that flag likely churners and experiments aimed at engagement are common companions to retention analysis. Predicting who will leave is not the same as knowing whom an intervention will keep, so the retention curve, rather than prediction accuracy, is what an intervention should be measured against.

References

• Amplitude, Interpret your retention analysis.
• ChartMogul, Report: Net MRR Retention.