Customer churn: what it is and how to calculate it

Customer churn is the share of your customers who leave in a given period. To count it, divide the customers you lost by the customers you had at the start, then multiply by 100. A low survey score is one of the earliest signs a customer may leave, and an alert that reaches someone the same day gives your team a chance to act.

This guide covers the meaning, the formula with a made-up example, and a small early warning routine you can set up with any survey tool.

What customer churn means

IBM defines customer churn as the number of existing customers lost, for any reason, over a given period. ¹ Some writers call the same thing customer attrition. ³ Others say customer turnover. ⁴

Three distinctions come up often:

• Customer churn and revenue churn. Customer churn counts accounts. Revenue churn counts the money those accounts paid. ChartMogul shows how they can differ: losing one of three customers is 33% customer churn, but if that customer paid the most, revenue churn is 50%. ² This post is about customer churn.
• Voluntary and involuntary. IBM describes voluntary churn as a change in the customer's preference. It describes involuntary churn as something often beyond the customer's control. ¹ A customer who decides to cancel is voluntary. A customer whose card failed did not decide anything.
• Customers and logos. ChartMogul also calls customer churn "logo churn" and counts only paying customers, so trial and free-plan users stay out. ²

Survey feedback helps most with the voluntary kind. It tells you little about a failed payment.

How to calculate customer churn rate

The formula is the same on every source we read. ^{1 3}

Churn rate = customers lost during the period / customers at the start of the period x 100

Pick one period (a month, a quarter, a year) and use it every time.

A made-up example. A software company has 400 customers on 1 March. During March, 20 cancel.

• Customers lost: 20
• Customers at the start: 400
• Churn rate: 20 / 400 x 100 = 5%

That reads as "5% monthly churn." Graphite Note works a similar example the same way: 50 customers lost from a starting 1,000 is 5%. ⁴

Turning a monthly rate into a yearly one

A monthly rate does not multiply by twelve. Amplitude notes that 5% monthly churn does not mean losing 60% of customers in a year, and that it compounds to roughly 46%. ³

You can check the arithmetic. Each month, 95% of the remaining customers stay. After 12 months, 0.95 multiplied by itself 12 times is about 0.54, so about 54% remain and about 46% have left. This ignores new customers, who would be counted separately.

Common mistakes

• Counting new customers in the base. Use the number at the start of the period, not the number at the end.
• Mixing periods. A monthly rate and a yearly rate are different numbers. Label which one you mean.
• Counting people who never paid. ChartMogul excludes trial and free-plan users. ²
• Counting people who joined and left in the same period. ChartMogul subtracts them, and customers who left and came back, from the lost count so they do not skew the rate. ²

Retention is the other side of the same number: retention rate is 100 minus churn rate for the same period. The customer retention guide shows how to count it, including revenue retention.

What feedback tells you before a customer leaves

The churn rate tells you how many left. It cannot tell you who is about to leave. Survey answers can point at people while there is still time to talk to them.

IBM lists a sudden increase in negative customer feedback among the anomalies worth tracking. It also says that the more satisfied customers a company has, the less likely it is to have churn issues. ¹ Three signals are worth watching for in your own data:

1. A low score. On the 0 to 10 Net Promoter Score question, a detractor answers 0 to 6. ⁵ Our detractor definition explains the group.
2. A negative comment. The words often say what the number cannot, such as "I could not get my data in."
3. Silence. A customer who stops answering anything you send is not a measured signal, but it is worth a look next to the first two.

A survey after someone cancels is the same information too late. ChartMogul suggests asking departed customers why they left. ² That is useful for the next customer, not for this one.

Be honest about the limits. A low score does not mean a customer will leave, and a high score does not mean they will stay. Some unhappy customers stay because switching is hard (the customer loyalty guide shows how to find them). Some satisfied customers leave when their budget or their company changes. Treat a low answer as a reason to ask a question, not as a forecast.

Set up an early warning in four steps

This routine works with a spreadsheet and a calendar if you have to. With a feedback tool it runs on its own.

Step 1: Ask at the right moment

Ask when the customer has just had an experience worth rating. A solved support ticket, a first week of use or a payment are good moments. A survey sent at a random time gets a random answer. If you are unsure how often to ask, read how to ask for feedback without annoying people.

In Pulseri, an event survey can start from your app, from Stripe or from Zendesk. The customer voice page lists the options, and the Zendesk CSAT guide shows one in full.

Step 2: Write one alert rule

Start with one rule that matches two things: a detractor score, or a negative comment. Use "any of," so either one is enough.

A rule you can copy into any tool that supports conditions:

Alert when any of:

• Score bucket is detractor (0 to 6 on a 0 to 10 scale)
• Comment sentiment is negative

Send to: the channel your support team reads, with the score and the full comment.

In Pulseri, alert rules can match on score, bucket (detractor, passive or promoter) or sentiment, combined with "all of" or "any of." The alerts and automations page describes them. A rule with a score or bucket condition must be set on one campaign, so write the rule above on your NPS campaign. A rule that only checks sentiment can cover every campaign. Alerts go to Slack, Microsoft Teams, email or a signed webhook, and show the score and the comment.

Two details save confusion:

• A new rule is not run against old answers. It checks answers as they arrive, so your channel does not fill up on day one.
• Sentiment is worked out from the comment after the answer arrives, so a sentiment alert can come a little later than a score alert. An answer with no comment has no sentiment, so only the score part of the rule can match it.

The step-by-step setup is in turn detractors into a Slack alert.

Step 3: Name an owner and a time limit

An alert nobody owns is noise. Pick one person, such as the support lead or the founder, and set a rule: reply within one working day. "Same day" is better when the team is small enough.

The reply does not need to be a sales message. Ask what went wrong and, if it is fixable, fix it.

Step 4: Protect people from too many surveys

An early warning is useless if customers learn to ignore your surveys. Keep a limit on how often one person is asked. In Pulseri, relationship surveys such as NPS wait 90 days per person by default, and the relationship cooldown is explained in the glossary.

Check whether it works

You will not know whether the warnings predict anything unless you compare them with who actually left. After a quarter, split the customers who gave a low answer into two groups: those your team acted on within a day, and those it did not. Then count how many in each group left.

A made-up example, to show the layout. Every number here is invented.

At first glance the alerted group left half as often. Do not conclude that the alerts caused it. Three leavers against six is too few to tell luck from effect, and the two groups may differ in other ways, such as plan, age of account or product use. You would need many more customers, or several quarters, before the comparison means much.

What to compare:

• The share who left in each group, not the raw counts.
• The same time window for both.
• Whether the alerted group has more of the customers who were already easy to keep.

If the groups are small, say so in your notes and keep collecting. The routine is worth running even before the numbers are conclusive, because the conversation with the customer has value on its own.

Next step

Pick the one moment where an unhappy customer is most likely to speak up, send a short survey there and write the alert rule above. If you run a SaaS product, Pulseri for SaaS founders shows how that fits together, and the pricing page lists what it costs. Pulseri does not calculate your churn rate. You count that from your own customer list, and the survey gives you a reason to act before the number moves.