Quick answer
Churn analysis measures when and how customers, accounts, subscriptions or revenue leave a defined active base. Retention measures the share or value that remains or returns under a stated rule. A reliable analysis defines the unit, starting population, qualifying activity, churn event, time window and revenue treatment, then compares age-aligned cohorts. Diagnosis combines behavioural, operational and customer evidence, while interventions require experiments or credible comparisons because correlation with retention does not establish cause.
What are churn and retention?
Churn is the loss of a defined customer, account, subscription, contract or revenue unit from an active base during a stated period. Retention describes the share or value that remains or returns according to a specified rule.
The concepts are related but not always simple complements. Customer retention and customer churn can sum to one under a closed starting cohort with no reactivation. Net revenue retention can exceed the starting value when expansion offsets contraction and churn. Product return behaviour may be measured with rolling retention rather than continuous subscription status.
A metric name without a measurement contract is incomplete. Teams should be able to state who enters the denominator, what counts as active or lost, the window, how reactivation is handled and whether the measure uses customers or money.
Customer, logo, user and revenue churn
Customer or logo churn counts economic relationships lost. User churn counts individual users and may move differently in multi-user accounts. Subscription churn counts ended subscriptions, which can exceed customer churn when one customer holds several products.
Gross revenue retention measures starting recurring revenue retained after churn and contraction, excluding expansion. Net revenue retention includes expansion and can therefore be stronger even while some customers leave. Contribution retention can be more decision-relevant when margins differ.
Voluntary churn follows a customer choice; involuntary churn can result from payment failure, expiry or administrative issues. They require different diagnosis and treatment, so report them separately where possible.
A retention measurement contract
Use the five steps below before choosing a dashboard. The contract should be versioned so a trend does not silently change when event logic, identity resolution or billing rules change.
Define the unit
Choose whether retention applies to users, customers, accounts, subscriptions, contracts or revenue.
- What economic relationship is being retained?
- Can one customer own several units?
Define activity and churn
Specify the event that proves continued value and the event or timeout that marks loss.
- What behaviour means active?
- Is churn observed or inferred?
Choose cohort and window
Align customers by start or milestone and observe them at comparable ages.
- Which customers entered together?
- What interval matches natural usage?
Measure value
Track customer count, recurring revenue, contribution and expansion under separate definitions.
- Are all retained customers economically equal?
- How do expansion and contraction affect value?
Diagnose and test
Find causes, prioritize moments and validate interventions with credible evidence.
- What preceded churn?
- Which action changes retention rather than merely predicts it?
Common churn and retention calculations
For a closed starting base, period customer churn is customers lost during the period divided by customers active at the start. Customer retention is eligible starting customers still retained at the end divided by the eligible starting customers. New customers acquired during the period do not belong in that starting denominator.
Revenue calculations use starting recurring revenue as the base. Gross revenue retention subtracts churned and contracted revenue and excludes expansion. Net revenue retention also adds expansion. Publish the exact equation because billing pauses, currency, refunds and one-time revenue can change the result.
Event-based products often use cohort retention: the share of a starting cohort that performs a defined return event in or after a later interval. Return-on, return-on-or-after and unbounded retention answer different questions.
Use cohorts to avoid misleading averages
A blended retention rate mixes customers at different ages and acquisition conditions. Group customers by a meaningful start such as signup, first purchase, activation or contract start, then compare them at equal tenure.
Read across a row to understand how one cohort changes with age and down a column to see whether later cohorts improved at the same age. Mark incomplete cells so recent cohorts are not treated as if they had the full observation window.
Behavioural cohorts can reveal associations between early actions and later retention, but selection effects are strong. Customers who adopt a feature may already be more motivated. Use the pattern to generate an intervention hypothesis, then test it.
Diagnose why customers leave
Begin with timing and segment patterns, then connect them to the journey. Early churn may reflect acquisition mismatch, expectation, onboarding or time-to-value. Later churn may reflect declining need, competitive alternatives, reliability, service or price changes.
Combine cancellation reasons, support contacts, payment failures, product use, delivery and quality records, interviews with churned and retained customers, and competitive evidence. Cancellation-form responses alone are convenient but can be incomplete or socially polite.
Separate prediction from cause. A risk score can identify customers likely to leave without identifying an action that will help. Diagnose the mechanism and the customer's legitimate outcome before designing outreach.
Churn and retention example
The backpack example shows why category context matters. Monthly repurchase would punish a durable product for working. The business needs relationship signals and longer cohort horizons tied to repair, advocacy and later contribution.
A repairable-backpack business wants to measure retention, but customers do not need to buy another bag every month. A subscription-style login metric would misrepresent the relationship.
Use purchaser or household as the primary relationship unit, with product registrations and orders connected under one privacy-respecting identifier.
Define relevant signals such as a repeat purchase, replacement-part order, completed repair, care-service use or verified referral. Keep each outcome separate rather than calling all engagement retention.
Group customers by first-purchase month and compare equal ages. Segment by initial product and acquisition source only when sample size is sufficient.
Combine return reasons, unresolved repair cases, delivery failures, customer interviews and behaviour before inactivity. Distinguish dissatisfaction from simply having no repeat need yet.
Test a clearer ownership and repair onboarding flow against a holdout. Measure repair completion, later contribution and service cost, not email clicks alone.
In a durable-goods category, absence of another purchase may be evidence of product durability rather than churn. The metric must reflect the value cycle.
Design and test retention interventions
Prioritize problems by customer harm, prevalence, value and ability to change. Product quality, expectation setting, onboarding, service recovery, billing reliability and cancellation experience may matter more than promotional reminders.
Match treatment to cause. Payment recovery can address involuntary churn; better qualification can reduce acquisition mismatch; product or service changes may address repeated failure. Do not trap customers with dark patterns or make cancellation intentionally difficult.
Use randomized holdouts where feasible and measure incremental retention, contribution, customer outcome and treatment cost. A campaign can appear successful because it targets customers who would have stayed anyway.
- Unit and identity defined
- Active and churn events documented
- Starting denominator correct
- Natural usage interval chosen
- Reactivation treatment explicit
- Customer and revenue metrics separated
- Cohorts age-aligned
- Causes triangulated
- Interventions tested against credible counterfactuals
- Customer rights and easy cancellation protected
How retention affects lifetime value
Retention is a major input to CLV because continued activity creates opportunities for future contribution. But improving a retention percentage does not automatically create value if the intervention is costly, customers are unprofitable or retained behaviour produces little contribution.
Customer heterogeneity matters. Cohort-level retention often changes with tenure as higher-risk customers leave. Applying one average churn rate forever can understate or overstate residual value depending on the setting.
Connect retention curves to contribution by cohort and product. This supports better CLV forecasts and reveals whether expansion, service cost or margin changes offset customer counts.
Evaluate retention analysis quality
Audit identity stitching, late events, refunds, plan changes and backfilled data. A single customer appearing as several users can inflate churn, while merged households can hide it. Reconcile customer and billing systems.
Track confidence and denominators. Small cohorts create noisy percentages. Report counts, uncertainty and data maturity beside rates. Preserve prior definitions when logic changes so historical comparisons remain honest.
Use qualitative follow-up to interpret changes. A retention curve can show when behaviour changed but rarely explains the full reason by itself.
Limitations and ethical risks
Retention can become a harmful objective when it rewards friction, dependency or unwanted contact. Easy cancellation, informed consent and respectful communication are product quality, not leakage to eliminate.
Prediction models can encode sensitive proxies and direct benefits or pressure unevenly. Use minimal data, test fairness and avoid targeting vulnerable customers with manipulative offers.
Finally, not all churn is preventable or undesirable. Customers complete jobs, businesses change and poor-fit relationships can be expensive for both sides. Diagnose fit before trying to retain everyone.
Retention analysis should help customers continue receiving value, not make leaving harder.
Frequently asked questions
How is churn rate calculated?
A common customer churn formula is customers lost during a period divided by customers active at the start, with exact eligibility and reactivation rules stated.
Are churn and retention always opposites?
Only under matching closed-cohort definitions. Net revenue retention includes expansion, and event-based retention can use different return rules.
What is the difference between customer and revenue churn?
Customer churn counts relationships lost. Revenue churn measures recurring value lost through churn and sometimes contraction.
Why analyze retention by cohort?
Cohorts align customers by start or milestone and reveal age-specific changes that blended averages conceal.
How do you reduce churn?
Identify causal mechanisms such as mismatch, delayed value, failure, service, billing or price, then test a proportionate intervention against a credible comparison.
Sources and further reading
- Stripe: Retention Rate vs Churn Rate ↗Definitions and distinctions for retention and churn in subscription businesses
- Amplitude Docs: Retention Analysis Calculation ↗Official documentation for event-based retention calculation choices
- For Entrepreneurs: SaaS Metrics 2.0 Detailed Definitions ↗Detailed churn, retention, expansion and unit-economics definitions
- Fader and Hardie: How to Project Customer Retention ↗Research on projecting retention patterns for customer tenure and value