Churn is the rate at which customers stop paying a business over a defined period. A streaming service opening a month with 10 million subscribers and losing 200,000 reports monthly churn of 2%. The metric exists because recurring-revenue businesses do not grow on acquisition alone: net growth is the gap between customers gained and customers lost. Acquisition cost is paid once. Churn is charged continuously against the base.

The word comes from butter-making, where agitation keeps the contents of a churn in constant motion. Applied to a customer file, it describes a pool that looks stable in aggregate while turning over underneath.

How churn is counted

The basic calculation divides customers lost during a period by customers present at its start. Five hundred subscribers on March 1, 25 gone by the end, and March churn is 5%. Simple arithmetic conceals a dispute about the denominator. Some operators divide by the average of the opening and closing base, which smooths fast-growing files; others by a total including customers acquired mid-period, which flatters any company adding subscribers quickly. There is no standard, and the choice can move a reported figure by a percentage point or more.

Two quantities carry the same name. Logo churn counts departing customers regardless of what they paid. Revenue churn counts the money that left with them, and the two diverge whenever a base is priced unevenly: one lost enterprise account can outweigh a thousand cancelled trials. Net revenue churn subtracts expansion from upgrades, seat additions and price rises, and can fall below zero when the surviving base grows faster than departures shrink it, a condition reported as net revenue retention above 100% and marketed as negative churn.

Churn also implies a lifetime. The reciprocal of the rate gives average tenure, so 25% annual churn implies four years and 33% implies three. Lifetime value models rest on that inversion, so an error in the churn input propagates straight to the acceptable cost of acquisition.

The split that matters operationally is between voluntary churn, where a subscriber decides to leave, and involuntary churn, where a payment fails. RevenueCat data covering more than 115,000 applications found that 32.2% of subscription cancellations on Google Play are billing failures rather than deliberate cancellations, against 15.2% on the App Store. Among those cancelling deliberately, cost accounted for 25% to 45% of stated reasons and insufficient usage for 26% to 40%, while technical problems explained 3% to 7%.

Timing is the other axis. Measured by cohort, cancellations cluster. Thirty-five percent of annual app cancellations fall in the first month, decay to roughly 5% a month mid-year, then return at 9% to 14% in month 12 as renewal approaches. An aggregate monthly figure averages those spikes into a number describing no actual subscriber.

Contractual and non-contractual bases

Where a contract exists, churn is observed: a cancellation is a timestamped event, and the calculation is bookkeeping. Where no contract exists, as in retail and most e-commerce, nobody announces a departure. A lapsed buyer and a slow buyer look identical, so churn must be inferred from purchase timing.

Academic work solved that before subscription software existed. The Pareto/NBD model published by David Schmittlein, Donald Morrison and Richard Colombo in 1987 treated purchasing and dropout as separate stochastic processes, estimating the probability a customer was still active. Peter Fader, Bruce Hardie and Ka Lok Lee simplified it in 2005 with the beta-geometric variant, still the standard tool in non-contractual settings.

Operationally, churn sits with lifecycle and customer relationship management teams rather than media buyers. Klaviyo maintains a predictive layer estimating each profile's lifetime value, churn risk and expected next order date. Billing systems handle the involuntary side through retry schedules and grace periods, and Apple exposed the underlying states when it added more than 50 subscription metrics to App Analytics, including plans facing billing issues and plans already churned.

From telecoms to app stores

Churn became an operating metric in telecommunications, where switching costs collapsed. Deregulation, then number portability, removed the friction holding subscribers in place, and mature mobile markets settled into monthly churn of roughly 1% to 2%.

The commercial argument for tracking it appeared in the September 1990 Harvard Business Review, where Frederick Reichheld and W. Earl Sasser published "Zero Defections: Quality Comes to Services". Cutting the defection rate by five percentage points, they reported, produced 85% more profit in one bank's branch system, 50% more in an insurance brokerage and 30% more in a car servicing chain. MBNA America halved a 10% defection rate and saw profits rise 125%.

Prediction followed measurement. The KDD Cup 2009, run between March 10 and May 11 of that year on masked records from the French operator Orange, asked more than 450 participants from 46 countries to rank customers by propensity to switch provider, against an in-house benchmark score of 0.8311. Software firms then rebuilt the vocabulary around monthly recurring revenue, and mobile platforms rebuilt it again for app stores, where cancellation is a system event rather than a phone call.

Why churn matters to marketers

Churn sets the ceiling on lifetime value, and lifetime value sets the maximum defensible acquisition cost. Every bid in a performance campaign is a claim about how long the acquired customer will stay, and that link is now mechanical. Meta released LTVision, an open-source library for predicted lifetime value modelling in January 2025, and value-based bidding optimises against modelled long-run revenue. The failure mode shows in RevenueCat's data: artificial intelligence applications generate 41% more realised lifetime value per payer in year one, yet their 12-month monthly retention is 6.1% against 9.5%, so algorithms anchored to 30-day or 60-day windows systematically overvalue them.

Retention economics have moved budget. The Interactive Advertising Bureau's January 2026 forecast of 9.5% advertising growth attributed a shift toward retention strategies to rising acquisition costs and maturing first-party data, treating loyalty programmes as growth engines rather than support tactics.

Packaging is itself a churn instrument. Publisher data shows bundled subscriptions churning at 0.7% against 16.4% for single titles, a gap translating into 26 times higher lifetime value. Streaming moved the other way, with monthly churn reaching 5.5%, up from 2% five years earlier, as households cycle through five or more services and accept advertising for a lower price. Win-back spending faces a ceiling: annual reactivation is capped at about 5% regardless of price tier or geography, which offer design does not appear to lift.

Limitations and disputes

Comparability is the first problem. Because denominators, periods and inclusion rules vary, two companies reporting the same churn rate may not be measuring the same thing, and a figure quoted without its formula says little.

Provenance is the second. Churn is frequently self-reported by parties with an interest in the result. AppLovin disclosed almost no customer churn until a short-seller report claimed roughly 23% of its e-commerce customers appeared to have removed its pixel in the first quarter of 2025, a contradiction the company disputed.

The canonical retention statistic has also drifted. A widely repeated claim that a five-point retention improvement lifts profits by 25% to 95% is routinely attributed to the 1990 Harvard Business Review article, which reported a different set of figures across named industries and never stated that range.

Involuntary churn, meanwhile, is read as dissatisfaction. Payment failure on Android is a billing problem, not a verdict on the product, yet it lands in the same number and triggers retention campaigns aimed at customers who never chose to leave.

A low rate can also measure friction rather than satisfaction. Regulators treat difficult cancellation as a consumer harm in its own right, which makes any figure produced under obstructive cancellation flows partly an artefact of design.

Not the same as

Retention rate is the complement of churn only when both use the same base and window. Retention is usually reported by cohort and churn in aggregate, so subtracting one from 100% often produces a number matching nothing.

Attrition is interchangeable with churn for customers but refers to staff when applied to organisations. Marketing leadership turnover is a separate question, and Spencer Stuart data on 218 chief marketing officer exits between 2021 and 2025 complicates the popular reading of it as instability.

Unsubscribe rate measures opt-outs from a marketing list. It costs reach, not revenue: a paying customer can unsubscribe from email without churning.

Cookie churn describes turnover of browser identifiers through deletion or expiry. It concerns identity persistence, not customer loss.

Recent developments

Regulation of cancellation is unsettled. The Federal Trade Commission finalised its click-to-cancel rule on October 16, 2024, requiring cancellation to be as simple as sign-up. Trade bodies including the Interactive Advertising Bureau challenged it, and the Eighth Circuit vacated the rule entirely on July 8, 2025, holding that the agency had skipped a required preliminary regulatory analysis. A draft advance notice of proposed rulemaking followed on January 30, 2026. Enforcement under the Restore Online Shoppers' Confidence Act continued throughout, producing a $60 million Instacart settlement in December 2025 and a complaint against JustAnswer in January 2026.

Other jurisdictions moved further. The United Kingdom confirmed a subscription contracts regime in an April 2, 2026 government response, expected to commence in spring 2027. Australia's Federal Court found in August 2026 that eHarmony breached consumer law across five categories, including automatic renewals at prices reaching five times the initial subscription.

Modelling has kept pace. IAB Italia's April 2026 white paper placed churn probability alongside purchase propensity and lifetime value as standard scoring outputs, and Adobe research from 2026 read uneven feelings of being heard across generations and devices as distributed churn risk rather than as a satisfaction score.

Timeline

  • 1987: Schmittlein, Morrison and Colombo publish the Pareto/NBD model for estimating dropout in non-contractual customer bases
  • September 1990: Reichheld and Sasser publish "Zero Defections: Quality Comes to Services" in Harvard Business Review
  • 1990s: telecommunications operators adopt churn as a core operating metric as switching barriers fall
  • 2005: Fader, Hardie and Lee publish the beta-geometric variant, simplifying non-contractual churn estimation
  • March 10 to May 11, 2009: the KDD Cup 2009 runs a churn prediction challenge on Orange customer records, drawing more than 450 participants from 46 countries
  • October 16, 2024: the Federal Trade Commission finalises the click-to-cancel rule
  • June 2025: Apple adds more than 50 subscription metrics, including churn states, to App Analytics
  • July 8, 2025: the Eighth Circuit vacates the click-to-cancel rule in its entirety
  • November 18, 2025: the Competition and Markets Authority opens its first consumer protection investigations under the Digital Markets, Competition and Consumers Act
  • December 2025: Instacart settles with the Federal Trade Commission for $60 million over subscription and fee practices
  • January 13, 2026: the Commission files its complaint against JustAnswer
  • January 30, 2026: the Commission submits a draft advance notice of proposed rulemaking on negative option plans
  • March 2026: RevenueCat publishes part one of its 2026 subscription benchmark; the Commission reopens negative option rulemaking
  • April 2, 2026: the United Kingdom publishes its government response confirming a new subscription contracts regime
  • May 28, 2026: RevenueCat publishes part two, covering cancellation distribution and reactivation ceilings
  • August 2026: Australia's Federal Court finds eHarmony breached the Australian Consumer Law across five categories
  • Spring 2027 (anticipated): the United Kingdom subscription contracts regime commences

Summary

Who. Lifecycle, retention and customer relationship management teams own churn on the brand side, with billing platforms and app store infrastructure supplying the raw cancellation events. Finance teams report it to investors. Regulators including the Federal Trade Commission, the United Kingdom Department for Business and Trade and the Australian Competition and Consumer Commission govern how easily it can occur.

What. The rate at which customers stop paying over a defined period, reported either by customer count or by revenue, split into voluntary and involuntary causes, and inverted to produce the average customer lifetime that underpins lifetime value models.

When. Established as a telecommunications operating metric through the 1990s after Reichheld and Sasser's 1990 article made the profit case, formalised for non-contractual settings in academic work from 1987 and 2005, and industrialised through prediction contests from 2009 and subscription benchmarking platforms during the 2020s.

Where. Any recurring-revenue category: telecommunications, streaming, publishing, software, mobile applications, insurance and utilities, plus inferred churn in retail and e-commerce where no cancellation event exists.

Why. Churn determines how long an acquired customer pays, which fixes the maximum defensible acquisition cost and therefore the bid in every performance campaign. It also determines whether retention spending or acquisition spending produces more revenue, a calculation that has moved budget toward the former as acquisition costs rise.