Customer churn: what it is, how to calculate it and how to reduce it before it happens
By Paco Herranz · November 19, 2026 · 8 min read
Churn is the percentage of customers who stop buying from you in a given period. In a subscription business you can see it: the customer cancels. In retail and ecommerce it gives no warning: the customer simply stops coming back, and by the time you notice, they are gone. Measuring it properly and spotting it before it happens is one of the cheapest ways to grow sales, because keeping a customer costs far less than replacing one.
In this guide you will see what customer churn is, how to calculate churn rate with a formula and an example, which signals predict it and what you can do to reduce it.
What is customer churn?
Churn (also called customer attrition or customer turnover) is the share of customers you lose in a given period. Its opposite is the retention rate: if you lose 30% of your customers in a year, you retain 70%.
In both cases there is also voluntary churn (the customer decides to leave) and involuntary churn (a declined payment or an expired card). Involuntary churn is the easiest to win back.
How to calculate churn rate: formula and example
The basic formula is:
Churn rate = customers lost in the period ÷ customers at the start of the period
Monthly example: 600 ÷ 10,000 = 6%
Annual: 1 − (1 − 0.06)¹² ≈ 52%
An example. You start the month with 10,000 active customers and lose 600 during the month. Your monthly churn is 6%. New customers acquired that month are not part of the calculation: they are measured separately.
Be careful when annualising it: multiplying by twelve is not enough. A 6% monthly churn means losing just over half of your base in a year, because each month applies to the customers who remain: 1 − (1 − 0.06)¹² ≈ 52%.
Active customers from a base of 10,000 with 6% monthly churn
Month 0 to month 12. After a year about 4,760 remain: 52% have gone.
Besides customer churn, it is worth tracking revenue churn: how much of your revenue you lose. Losing 600 customers who buy once a year is not the same as losing 600 of your best customers.
When is a customer considered lost?
In non-contractual churn, the key to the calculation is the inactivity window: how long without a purchase marks a customer as lost. Set it too short and you will see churn where there are only infrequent buyers; set it too long and you will find out too late.
A practical rule is to start from the usual buying rhythm. If your customers buy every 45 days on average, a customer who has gone 90 to 120 days without buying is at serious risk. Even better if you calculate it by segment or customer by customer: someone who buys every week does not have the same rhythm as someone who buys twice a year.
The signals that predict churn
Customers rarely leave all at once. Before they stop buying, they almost always fade out, and those signals are in your data:
Purchase frequency drops
Where it lives: ecommerce and POSOr the time between orders gets longer.
Average order value falls
Where it lives: ecommerce and POSOr they buy from fewer categories than before.
They stop engaging
Where it lives: email and appThey open emails but do not buy, or no longer open them at all.
They have an unresolved issue
Where it lives: customer serviceA return, a complaint or an order that arrived wrong.
They stop using the app or the loyalty card
Where it lives: app and loyaltyThe problem is that each signal lives in a different system: purchases in the ecommerce platform and the point of sale, emails in the email tool, issues in customer service. To see them together you need to unify each customer's data into a single profile, which is what a Customer Data Platform (CDP) does.
How to predict churn
With unified data, a machine learning model learns which combination of signals comes before a customer leaves in your business and calculates, for each customer, the probability that they will stop buying in the next 30, 60 or 90 days.
The difference from a fixed rule ("more than 90 days without buying") is timing: the model spots the customer who is cooling off while they are still buying, not once they have already gone. And it does so customer by customer, taking their own rhythm into account.
| Customer | Probability | Main signal | Suggested action |
|---|---|---|---|
| Customer A · high CLV | 74% | Frequency down from 2 to 0.5 purchases a month | Call from the loyalty team |
| Customer B · medium CLV | 68% | Unresolved return | Resolve the issue; no campaign |
| Customer C · medium CLV | 41% | Opens emails but does not buy | Recommendation based on history |
| Customer D · low CLV | 38% | Stopped using the loyalty card | Points reminder |
Illustrative example; not taken from any customer.
Two things worth demanding from a churn model: that it shows its accuracy, so you know how far to trust it, and that it explains which signals weigh on each customer, so you know what to do about them.
How to reduce churn: six actions
Look after the second purchase
Many customers buy once and never come back. Onboarding designed to lead to a second purchase reduces churn in the first months, which is usually the highest.
Act when frequency drops, not when the customer cancels
A retention campaign works much better on a customer who is cooling off than on one who has already left.
Resolve issues
A customer with an open complaint should not receive a sales campaign, but a solution.
Personalise based on purchase history
Recommend what fits what they buy, not the latest item in the catalogue.
Prioritise by value
Not every at-risk customer deserves the same effort. Cross churn probability with customer lifetime value (CLV) and start with the most valuable.
Measure with a control group
Keep part of the at-risk customers without a campaign so you know how many you really retained. Without a control group you cannot tell whether the campaign worked or those customers would have come back anyway. We explain it in what incrementality is and how to measure it.
Churn and CLV: why every point counts
Churn determines how long a customer stays and therefore how much they are worth. A customer's average lifespan is roughly 1 divided by annual churn: with 33% churn, a customer lasts about 3 years; with 25%, about 4.
Following the example in our CLV guide, a customer who brings €72 of profit a year is worth €216 if they stay 3 years and €288 if they stay 4. Cutting churn by eight points raises the value of every customer by 33%, without acquiring a single new one.
With €72 of profit per customer per year, the example from our CLV guide.
Frequently asked questions about churn
What does churn mean? Churn is customer attrition: the share of customers who stop buying from you or cancel in a given period. It is also called customer turnover or attrition rate.
How is churn rate calculated? Divide the customers lost in the period by the customers you had at the start. If you start the month with 10,000 customers and lose 600, monthly churn is 6%.
What is the difference between churn and retention? They are two sides of the same metric: retention rate is 100% minus churn. With 30% annual churn, you retain 70% of your customers.
What is a good churn rate? It depends heavily on the sector, the business model and how you define a lost customer. More useful than comparing yourself with an average is comparing your churn by cohort, acquisition channel and segment, and tracking how it evolves.
How is churn predicted? With a machine learning model that learns from your history which signals come before a customer leaves (lower frequency, smaller orders, less engagement, issues) and calculates for each customer the probability of leaving in the coming months.
How FLYDE detects churn
FLYDE brings together in a single profile each customer's online and in-store purchases, emails, app activity and issues. On that profile, the churn model in Predict Pack identifies which customers are going to stop buying with time to act, and the Unsubscribe model anticipates cancellations in subscription businesses. Each model shows its accuracy metrics and is retrained with new data.
Churn probability appears on each customer's record and becomes an audience in one click, ready for email, ads or your sales team. In the Scoring module you can combine it with other signals, such as RFM, to build your own priority score. You can also ask FLYDE Brain in plain language: "who is at risk of churning this month?". And with the Incrementality module you set aside the control group when you build the audience, so you know how many customers you really retained.
Want to know how many customers are about to leave? Request a demo and we will look at it with your data.