Churn rate analysis: what your aggregate number hides

Your board deck shows 5% monthly customer churn. The percentage is correct. It still doesn't tell your customer success team which accounts to investigate on Monday.

Consider a hypothetical subscription business starting a month with 1,000 paying customers: 800 on a small-business plan and 200 on an enterprise plan. During the month, 48 small-business customers and two enterprise customers cancel. The blended rate is 50 divided by 1,000, or 5%. Within the segments, it is 6% and 1% respectively.

One headline, two different retention problems. Churn rate analysis becomes useful when you can move from that headline to a defined group of customers, a possible explanation and an action someone owns. Keep the aggregate number. Stop asking it to do the whole job.

Start churn rate analysis by fixing the denominator

For the examples here, monthly customer churn means customers from the opening paying-customer base who leave during the month, divided by that opening base. Count each customer once, even if they have several subscriptions. Keep new customers out of both sides of that calculation and report their early cancellations separately.

That last point corrects a tempting explanation for healthy-looking churn: "New signups are hiding the losses." Under this opening-base definition, customers added during the month do not increase its denominator. They cannot mathematically dilute that month's rate. Changes in the composition of the opening base across months can change the blended result, however.

Write the definition beside the chart. Decide how you handle pauses, reactivations, unpaid invoices and the effective cancellation date. A customer who asks to cancel today but has access until the end of an annual contract creates a different reporting event from an account whose service ends today. Choose the event that your report measures and apply it consistently.

Check your analytics tool's conventions too. ChartMogul's cohort documentation, for example, describes customer-churn cohorts that offset cancellations with reactivations. That differs from a gross cancellation count. Two reports can use the word "churn" and answer different questions.

The period needs equal attention. If a fixed cohort loses 5% of its remaining customers every month, with no reactivations, 12-month retention is 0.95 to the power of 12, approximately 54.0%. Cumulative churn is therefore about 46.0%, not 60%. This is a mathematical illustration, not a forecast for your customers.

Compare customers at the same age

A cohort groups customers with a shared starting point, such as the month of their first paid subscription. You then follow that group through its own first month, second month and so on. Stripe's guide to SaaS cohort analysis explains why these groups can reveal changes that company-wide averages hide.

Use a fixed definition of retention. In this hypothetical table, a customer counts as retained if their paid subscription is active at the specified age; there are no reactivations. Each percentage uses that cohort's original customer count.

February's 90-day result is ten percentage points below January's. March looks better at 30 days than either earlier cohort, but it has not reached 90 days. Leave that cell blank until it does. An unobserved outcome is not a zero, and it is not permission to extend a promising first month into a forecast.

The table tells you where to investigate. It does not tell you why February performed worse. Look for differences in acquisition source, customer needs, plan mix, pricing or product experience, then examine the affected accounts. If February also contained fewer customers, show the count prominently: one cancellation in a ten-customer cohort moves the rate by ten percentage points.

Add revenue before deciding which losses matter most

Customer churn, often called logo churn, treats every customer as one account. Revenue measures answer a different question.

In a second hypothetical example, losing ten customers paying $500 a month removes $5,000 in monthly recurring revenue. Losing one customer paying $5,000 removes the same MRR. The customer counts are ten and one, not one and one. You need both views to understand the loss.

Gross revenue retention, or GRR, measures how much of the opening recurring revenue remains after cancellations and downgrades, without credit for expansion. Net revenue retention, or NRR, includes expansion from the existing customer base. ChartMogul's NRR calculation also includes reactivation by default and provides an option to exclude it, another setting worth documenting.

Suppose the opening customer base contributes $100,000 in MRR. Over the month, cancellations remove $5,000 and downgrades remove $3,000. Expansion from those existing customers adds $12,000, with no reactivations. GRR is ($100,000 – $5,000 – $3,000) / $100,000 = 92%. NRR is ($100,000 – $5,000 – $3,000 + $12,000) / $100,000 = 104%.

New-customer revenue is excluded from both calculations. These are monthly results; don't compare them directly with a twelve-month retention benchmark.

The example's 104% NRR means expansion more than offsets lost recurring revenue. It does not mean every customer segment is healthy. Put GRR beside NRR, then show where the cancellations and downgrades occurred. Otherwise expansion can make a deteriorating part of the customer base easy to overlook.

Use benchmarks only after matching the business

A benchmark needs a population as well as a percentage. In its 2026 research on bootstrapped SaaS businesses, SaaS Capital reports median NRR of 103% and GRR of 91% for companies with $3 million to $20 million in ARR. The wider survey covered more than 1,000 private B2B SaaS companies; the post does not state the size of that specific subgroup.

Those figures describe a particular group. They are not a universal target for every subscription product, and they do not explain any one company's cancellations.

Before using an external benchmark, confirm its measurement period, customer definition, revenue treatment and business segment. If the methodology does not let you match those dimensions, use it as context rather than a pass/fail line. A board slide needs that qualification more than it needs another decimal place.

Your own same-age cohorts are the more direct place to start diagnosing a change. Begin with one split, such as customer size, and add another only when it answers a question. Dividing a small customer base into every possible combination of channel, plan and region produces lots of percentages and very little confidence.

Turn churn rate analysis into a short investigation queue

Build the first working dataset around customer IDs and billing history. You need paid-start dates, cancellation-effective dates, subscription status and recurring revenue movements. Add plan, acquisition source and relevant product events where they are available and consistently recorded. A signup date and last login alone cannot distinguish a paying but infrequent user from a cancelled subscription.

Then choose the largest unexplained loss in the period. The useful next action depends on what you find:

1. If a recent cohort loses more customers at the same age, review those accounts' onboarding, support history and reasons for leaving. Treat onboarding failure as a hypothesis. A missing feature or poor customer fit can produce an early cancellation too.
2. If one acquisition source has weaker retention, compare similar plans and customer sizes before blaming the channel. Investigate what was promised and which use cases were sold.
3. If cancellations follow failed payments, separate those accounts for billing recovery. Stripe's retry documentation confirms that retries can recover some failed payments, while hard declines and missing payment methods require different handling. There is no recovery percentage you should assume for your own base.
4. If losses cluster around renewal, review that contract cohort's usage, stakeholders and recorded objections. Set the timing of the intervention around your renewal process rather than borrowing a universal day count.

For each investigation, name an owner, record the evidence and set a review date. If you change onboarding, say what changed and which subsequent cohort received it. Compare that cohort at the same age, while recording other changes that could explain the result. An improvement after an intervention is useful evidence to investigate, not automatic proof that the intervention caused it.

Put the decision next to the number

For the next retention review, bring one page: the period and metric definitions, customer and revenue retention, the segment with the largest unexplained loss, and the accounts behind it. Add a named action with a date for checking the outcome.

Keep actual cancellations separate from accounts you think might leave. Forecast risk belongs in the discussion, but it has not yet entered the churn numerator. That distinction lets a team act early without quietly changing the reported result.

The next useful question for your board deck is specific: which group of customers will receive a different intervention because of this analysis? If the answer is nobody, you have finished the reporting and still have the retention work to do.

Images: AI-generated editorial illustrations.