Customer retention metrics: retention, churn and LTV explained

Customer retention metrics tell a lifecycle team how much of its customer base it keeps. The retention rate is the share of customers you had at the start of a period who are still customers at the end. The churn rate is the share of that same base you lost. Customer lifetime value (LTV) is an estimate of what a customer is worth over the whole relationship.

These are the core lifecycle marketing metrics, but none of them means much on its own. Unless the unit, the period and what counts as "still a customer" are fixed up front, the same data produces different numbers.

This guide explains what each metric measures, how analytics and billing tools define them differently, and which metric leads for each business model. The calculations themselves live in the site's calculators, so this post doesn't repeat formulas.

When two teams report different churn rates for the same period, the problem is usually the definition, not the data. Three decisions set what a retention number means: what you count, what counts as still a customer, and how the period is counted.

Users, customers or revenue: choose the unit first

Retention can be measured in users, paying customers or revenue, and each tells a different story. ChartMogul calls watching only customer churn a common mistake, because a handful of high-value cancellations can hurt revenue far more than the customer count suggests. The same page says only paying customers belong in customer churn, with free trial and free plan users excluded.

For retention measured in customers, the site's retention rate calculator does the math. The revenue side has its own net revenue retention (NRR) calculator.

A note on terms: a customer retention rate sits between 0% and 100% and can't go negative. That is because a group can't lose more customers than it started with. "Negative churn" is a revenue concept.

ChartMogul's revenue churn page notes that net revenue churn can go negative. That happens when expansion revenue from existing customers outweighs what downgrades and cancellations take away. It has nothing to do with winning new customers.

What counts as still a customer

When a customer counts as lost also varies by tool. According to Stripe's billing analytics documentation, a subscriber churns when all of their subscriptions are canceled or marked unpaid. Baremetrics treats customers who are delinquent for more than 30 days as canceled. Recurly doesn't count a subscription on hold as churned.

The denominator differs too. Stripe's subscriber churn rate adds the past 30 days' new subscribers to the starting base. Baremetrics uses only the customers active 30 days ago.

ChartMogul starts from customers at the beginning of the period and removes those who joined and churned or churned and reactivated in the same period. Before comparing churn rates from two tools, check that their definitions match.

Days, weeks and months differ between tools

How a period is counted changes the result as well. Amplitude's page on how time works in retention says it treats a day as a rolling 24-hour window by default, different for each user. It also computes a month as 30 days.

In Mixpanel, day 1 retention means 24 to 48 hours after the user's start. In Google Analytics 4's cohort exploration, a week runs Sunday to Saturday rather than a rolling 7 days. The same group of users can show three different week 1 retention numbers in these three tools.

What it means for a customer to stay depends on the business. In a subscription business, the cancel button answers that question. In e-commerce and in mobile apps, you have to define the answer yourself.

Subscriptions: customer churn, revenue churn and NRR

Subscription businesses track two kinds of churn together: how many customers left and how much revenue was lost. Net revenue retention (NRR) shows how much recurring revenue remains from existing customers after upgrades, downgrades and cancellations. Gross revenue retention (GRR) leaves expansion out. ChartMogul notes that GRR is capped at 100% by definition and that comparing your GRR to someone else's NRR is a common mistake.

The window has to be consistent as well. ChartMogul says it is common to measure NRR over 12 months, while Baremetrics calculates NRR and GRR on a rolling 30-day basis.

For app subscriptions, Apple reports voluntary churn (the user canceled) separately from involuntary churn (a billing issue Apple couldn't recover within its 60-day retry window). The second is fixed in the payment flow, not with a marketing message.

E-commerce: repeat purchase inside your own window

In e-commerce nobody cancels anything. A customer doesn't come back.

Shopify's customer reports define a returning customer as one whose order history already includes at least one order. Shopify also notes that the reports use each customer's entire order history, not only orders in the selected timeframe.

There is no ready-made definition of a lost customer. Look at how long your own customers usually wait between orders, set a window from that, and keep it fixed when you compare periods. Segmenting customers by time since their last order is the subject of the RFM analysis post. RFM stands for recency, frequency and monetary value.

Mobile apps: N-day and on-or-after retention

In mobile apps, retention measures whether users come back after a starting event, usually the install or first open. There are two basic methods. N-day retention asks whether the user came back on exactly day N. On-or-after retention, also called unbounded, asks whether the user came back on day N or any day after it.

Amplitude's retention documentation calls these Return On and Return On or After. Amplitude and Mixpanel default to the second, PostHog to the first. On-or-after retention always gives the higher number, so results from the two methods can't be compared.

Which one fits depends on how often the app is meant to be used. N-day suits an app people use every day, and on-or-after suits an app opened a few times a month.

A hypothetical example shows how large the gap can be. Out of 1,000 users who installed the app on the same day, say 120 opened it on exactly day 7. Also say 310 opened it at least once on day 7 or later. For the same cohort, the two methods give these results:

Both numbers are correct, but they answer different questions. Putting one team's 31% on-or-after result next to another team's 12% N-day result suggests users are retained more than two and a half times better. But the whole gap comes from the method.

The app stores' own numbers measure different things too. Apple's App Store Connect retention counts devices that installed the app and opened it a given number of days later. Installs that were never opened don't count.

Google Play Console's retained installers count users who kept the app on a device. The app doesn't have to have been opened in that time.

A single monthly retention rate averages customers who arrived at different times. Cohort retention follows each starting group in its own row and shows which cohorts are retained better or worse. Anchor each cohort on one event: the install for an app, the first order for a store, the first payment for a subscription.

Shopify's default cohort report groups customers by the date of their first order. Stripe assigns subscribers to cohorts by when they first started generating recurring revenue.

Two traps come up when reading cohorts. The first is incomplete periods. Amplitude warns that while an analysis is still in progress, the retention curve can bend upward and appear to increase. It says you should wait until the time frame is complete to read the trend.

The second is the average. PostHog notes that a weighted mean accounts for different cohort sizes while a simple mean treats all cohorts equally. Averaging a cohort of a few hundred users with one of tens of thousands at equal weight gives a misleading figure.

Because tools use the same words for different things, check these before comparing numbers from two of them:

• "Rolling": In Amplitude and Mixpanel, 24-hour windows from each user's own start. In PostHog, on-or-after retention. In GA4's cohort exploration, returning in every period up to that one. In Baremetrics and Braze, a trailing 30-day window.
• "Returning user": In GA4, a user who has visited at least once before. In Google Play, a user who reinstalled after uninstalling. In Shopify, a customer with an earlier order.
• "New user": Amplitude classifies new users by when the account was first seen, not by the first time they performed the selected event.
• Device and identity: GA4's cohort exploration is based on device data only and doesn't consider User-ID.

LTV is a planning number that shows how valuable a customer is, not a precise measurement. Tools estimate it with different formulas. Stripe and Baremetrics divide average revenue per user by the churn rate.

ChartMogul's LTV page adds gross margin and says LTV and CLV are the same metric under different names. It also warns that the basic formula assumes churn is linear. In reality churn is highest in the first months, so the formula overstates LTV. Klaviyo's predictive analytics add the value of past orders (historic CLV) to a prediction for one year by default (predicted CLV) to get total CLV.

Analytics tools use revenue within a window. GA4 shows the average revenue from new users over their first 120 days. Amplitude shows the cumulative revenue a cohort generates in its first N days.

The practical upshot: before comparing two LTVs, check which model produced each, whether it counts revenue or margin, and which window it covers. To get a simple or margin-adjusted LTV from historical order data, use the LTV calculator. Churn and CLV predictions for individual customers are covered in the AI in lifecycle marketing post.

LTV:CAC and CAC payback belong to acquisition planning

LTV is most often used next to customer acquisition cost (CAC). ChartMogul notes that CAC on its own says nothing about whether the spend paid off. It should be read with the CAC payback period and LTV:CAC. The same page calls it a common mistake to compare CAC across companies or channels without checking that both used the same fully loaded definition.

Both sides of the ratio need the same customer definition and the same period. The LTV:CAC ratio calculator and the CAC payback period calculator handle the math. These two metrics mostly serve acquisition budgeting, while a lifecycle team's day-to-day decisions run on retention and churn.

A customer doesn't go from active to lost in a single day, and a reactivated customer is not the same as one who never left. Amplitude's lifecycle chart splits active users into new, current and resurrected (previously inactive). It counts users who were active in the previous period but not this one as dormant.

PostHog adds a useful warning when describing the same states. If the interval you choose doesn't match your product's natural usage frequency, the results can mislead.

There is one more reason to count reactivated customers separately. Suppose a win-back campaign works well and the returning customers are folded into the retention rate. The rate then looks as if existing customers are being retained better.

ChartMogul removes customers who churned and reactivated within the same period from its customer churn rate. Baremetrics, by contrast, adds a reactivated customer back into their cohort. Whichever approach your tool takes, reporting reactivations as their own number shows a program's effect more accurately.

Each of the three states calls for a different customer relationship management (CRM) flow. A few flows from the Journey Library as examples:

• Tiered churn intervention: For a lapsing customer, choosing the level of intervention by the strength of the risk signal.
• Lapsed customer win-back: For a customer counted as lost after no order inside the window.
• Usage habit recovery: For an app user whose usage frequency has dropped.

If retention went up after you launched a CRM program, that number alone can't tell you what caused it. The cause could be the program, the season or a new product feature.

To see the program's effect, hold out a randomly chosen group that doesn't get the program. Compare the two groups' retention with the same definition and the same window. The window should match the behavior the program targets. That means the first 30 days for an onboarding flow, or the customer's normal order interval for a win-back flow.

Channel metrics are not retention metrics. An email's open or click rate, or an in-app message's dismissal rate, tells you whether a message was seen, not whether the customer stayed. Those metrics and what they measure are covered in the email deliverability and in-app messaging vs push notifications posts.

Agreeing on a few definitions in writing before the first lifecycle report saves every argument that would follow. The table below is a filled-in example definition sheet for a mobile app with subscriptions. Rewrite each row for your own business model.

Once written, the sheet should stay fixed. If a definition changes, older and newer periods stop being comparable. If you do need to change one, note the date and the reason next to the report. That is the simplest way to explain the break in the numbers months later.