Behaviour beats opinion
Surveys capture what customers think; behavioural metrics capture what they decide. A customer who rates the business highly and then leaves has told you which of the two matters. That is why retention measures sit at the top of any serious experience measurement plan, with perception and operational measures beneath them as explanations.
Behavioural metrics have their own weaknesses. They lag: by the time a customer has churned, the cause is weeks or months old. They are silent about reasons: a churn figureShare of customers at the start of a period who left during it. says how many left, not why. And they depend completely on definitions that are easy to get wrong and easy to change quietly. The voice of the customer guide covers the perception side that supplies the reasons; this guide covers the behaviour.
Definitions deserve more attention than they usually get. What counts as a customer? When does a paused subscription become a cancelled one? Is a customer who buys once a year active in a month when they did not buy? Each choice changes the number, and different teams often make different choices. Write the definitions down once, agree them across finance, marketing and product, and hold them in a shared metrics layer as the metrics layer and definitions guide describes. Without that, every review spends its time reconciling figures rather than acting on them.
Churn: customer vs revenue definitions
Churn comes in two main forms, and a business with recurring revenue needs both.
| Measure | What it counts | Basic definition | Tells you | Watch out for |
|---|---|---|---|---|
| Customer churn rate | Customers lost | Customers lost in the period divided by customers at the start of the period | How many relationships are ending | Counting new customers who joined and left within the period |
| Gross revenue churn | Recurring revenue lost from existing customers | Revenue lost to cancellations and downgrades divided by starting recurring revenue | How much income the base is shedding | Excluding downgrades makes it look better than it is |
| Net revenue retention | Revenue kept and grown from existing customers | Starting revenue minus losses plus expansion from the same customers, divided by starting revenue | Whether the existing base grows or shrinks on its own | Can look healthy while many small customers leave |
| Voluntary churn | Customers who chose to leave | Cancellations initiated by the customer | Experience and value problems | Mixing with failed payments |
| Involuntary churn | Customers lost to failed payments or expired cards | Accounts closed after payment failure | Billing and recovery problems | Usually cheap to fix; often unmeasured |
The basic customer churn calculation is:
Customer churn rate = customers lost during the period ÷ customers active at the start of the period(ratio)
Three rules keep it honest. Use the customers at the start of the period as the base, so new customers acquired during the period do not dilute the figure. Report the period alongside the rate, because a monthly and an annual rate are not comparable. And separate voluntary from involuntary churn, because they have different causes and different owners.
Customer and revenue churn diverge when the customers who leave differ in size from the average. A business losing many small accounts and few large ones has high customer churn and modest revenue churn; one losing a few large accounts has the reverse. Reading both shows which problem it has. The financial consequences, including how churn sets the payback period, belong to the retention and payback guide.
Revenue definitions matter for retention offers too: the cancellation save offer test reads net retained revenue per cancellation starter, because saves counted at the moment of the offer ignore subscribers who leave when it ends.
In business-to-business accounts the expansion part of net revenue retention usually comes from teams, sites and products the first contract did not cover, and the guide to expansion and renewal marketing covers how marketing maps and works that part of the customer base.
Retention, repeat and active-customer rates
Retention is churn seen from the other side: the share of customers at the start of a period still active at the end. For subscription businesses the two mirror each other. For businesses without subscriptions, such as most retailers, there is no cancellation event, and retention has to be defined through activity.
Retention rate is the share of a group of customers still active after a given time. It is most useful when tied to a cohort, described below, because the rate for a fixed group over time shows how quickly customers fall away.
Repeat purchase rate is the share of customers who have bought more than once within a defined window. It is the natural headline for transactional businesses, where the hardest step is usually the second purchase. It reads best alongside purchase frequencyAverage number of orders each customer placed in the period. and AOVAverage revenue per order., because a business can grow revenue per customer by more frequent smaller orders or less frequent larger ones, and the experience behind each is different.
Active-customer rate is the share of known customers who have bought or used the product within a recent window. The window should reflect the natural purchase cycle: a grocery business and a furniture retailer cannot use the same one. Set it from the business's own data on the gaps between purchases, so that a customer marked inactive is genuinely overdue rather than simply between purchases. For an app, where the gaps are between core actions rather than purchases, the tutorial on setting an app's lapse threshold builds the table that finds the window.
Each of these feeds customer lifetime value. A customer retained longer, or buying more often, is worth more over their lifetime, and LTVTotal value a customer brings over time. rises accordingly; that in turn sets how much the business can afford in CACCost to acquire one new customer.. The maths is in the growth economics guides; the point here is that the retention measure chosen should be the one that best predicts lifetime value for this particular business.
Cohort views in brief
A blended retention rate, calculated across all customers, is dominated by the long-standing loyal customers who make up much of any mature base. It can stay flat while newer customers are retained worse and worse, and the problem only shows when those newer customers become a large share of the base.
A cohort view fixes this. Customers are grouped by when they were acquired, and each group is followed over time, so that the retention of customers who joined this quarter can be compared directly with those who joined a year earlier at the same age. Cohorts can also be grouped by acquisition channel, first product or campaign, which shows whether some sources bring customers who stay and others bring customers who leave.
A pet food subscription reads its cohorts
Kibblewick, a pet food subscription, reports a steady monthly churn of about 4% for a year. A cohort view tells a different story. Of customers who joined in January, 82% are still subscribed after three months; of those who joined in June, only 64% are. The difference is concentrated in customers acquired through a heavily discounted first-box offer that began in May: 51% of them remain at three months, against 79% of customers from other channels. The blended rate has stayed flat because the large base of older subscribers is very stable. The team caps the first-box discount and adds a feeding-guide email in week two, and the September cohort's three-month retention recovers to 76%.
Building cohort tables, from grouping customers to handling partial periods, is covered step by step in the cohort analysis tutorial. For experience work, the key reading is the shape of the curve: a steep early drop points to onboarding, a gradual slope to habit and value, and a step at a particular age to a specific event such as the end of a promotion or the first renewal. Reading cohorts by acquisition channel needs billing and analytics joined on a shared customer identifier, which the subscription measurement guide covers along with judging channels on revenue that arrives after the first payment. Whether retention is read by cohort at all is one of the two gates in the customer retention audit scorecard, which rates the measurement here alongside the experience, service and exit it is meant to steer. When a churn figure has risen for several periods, the diagnosis for a customer churn rate that is increasing applies these definitions and cohort readings in order, starting with whether the figure is real.
Leading indicators and service metrics
Churn is a lagging measure. By the time it moves, the customers have gone. Leading indicators move earlier and give the business time to act.
Behavioural indicators are usually the strongest. Falling usage or purchase frequency, a lengthening gap since the last purchase, features no longer used, logins that stop, a downgrade, a removed payment method or a visit to the cancellation page all tend to precede churn. Which indicators matter depends on the business; find them by comparing what customers who left did in the weeks before leaving with what similar customers who stayed did.
Service indicators are the second group. Repeat contacts about the same issue, unresolved complaints, long waits for a reply and failed deliveries are all warning signs, and support data records them as they happen. Contact rate per customer, first-contact resolution and complaint volume by cause belong on the retention dashboard rather than only on the support team's.
Perception indicators are the third group: low transactional survey scores and negative comments. They are useful when they have been shown to predict behaviour in the business's own data.
Build the indicators into a simple risk view: customers showing several warning signs at once are flagged for a response, whether a lifecycle message, a support call or an account review. Check the flags against what actually happens, and drop indicators that do not predict churn.
A retention KPI tree
A KPI tree connects the retention outcome to the drivers the business can influence, so that a movement in the outcome can be traced to its cause and each driver has an owner. The KPI tree framework describes the general method; applied to retention, the tree usually looks like this.
At the top sits the outcome: revenue retained from existing customers, or the retention KPIA metric tied directly to an objective. the business has chosen. Beneath it sit the components: customers retained, and revenue per retained customer (frequency multiplied by order value, or plan value). Beneath customers retained sit voluntary and involuntary churn; beneath voluntary churn sit the drivers from the diagnosis, such as early activation, usage frequency, service quality and value for money; beneath each driver sit the leading indicators that measure it. The AARRR framework is a useful check that the tree includes activation, which so often drives the retention that follows. The decisions this tree forces, from choosing between customer and revenue retention to separating involuntary churn and reading each branch at a fixed cohort age, are worked through with a filled example in the KPI tree for retention.
The tree is also a planning tool: for each driver, it records who owns it and what is being done. Use the builder below to sketch the tree for your own business, starting from the outcome and working down to indicators you already collect.
Retention KPI tree builder
Common mistakes
- Undefined terms. "Active" and "churned" left to each team's judgement produce figures that never reconcile.
- Diluting the base. Including customers acquired during the period in the denominator makes churn look lower than it is.
- Blended rates only. Without cohorts, rising losses among new customers stay hidden until they are large.
- Customer churn alone. Revenue churn and net revenue retention can tell a very different story.
- Ignoring involuntary churn. Failed payments are often a large and easily fixed share of losses.
- One window for everyone. An inactivity window that ignores the natural purchase cycle mislabels customers as lapsed.
- Lagging measures only. Without leading indicators, the business learns about churn after it can act.
- Indicators never validated. Warning signs that do not predict churn in your own data waste effort.
