Churn rate analysis: how to measure churn by segment
Analyze churn beyond one average: split voluntary from involuntary churn, segment by plan, tenure and channel, read cohort curves, compare matched benchmarks.
By Florian
Churn rate analysis means splitting one churn number into the parts that need different fixes: voluntary versus involuntary churn first, then segments such as plan, tenure, channel and payment method, then the timing of departures across cohorts. A single company-wide rate cannot show whether customers are rejecting your product or payments are failing on their own, and it hides the small segments where churn concentrates. This guide shows the formulas, the splits worth running first, how to read cohort curves, which benchmarks to compare against and how to run a monthly churn review.
If you need the basic formulas first, start with how to calculate churn rate. This guide picks up where that one stops: what to do with the number once you have it.
Which churn formulas do you need, and what does each one tell you?
| Metric | Formula | What it reveals |
|---|---|---|
| Logo churn rate | Customers lost in period ÷ customers at period start | How many accounts leave. Example: 40 lost ÷ 800 starting customers = 5.0%. |
| Revenue churn rate | MRR lost to cancellations ÷ MRR at period start | How much revenue leaves. Example: $12,000 ÷ $200,000 = 6.0%. |
| Gross revenue retention | (Starting MRR − churn − contraction) ÷ starting MRR | Revenue kept before expansion. Example: ($200,000 − $12,000 − $4,000) ÷ $200,000 = 92.0%. |
| Net revenue retention | (Starting MRR − churn − contraction + expansion) ÷ starting MRR | Whether existing customers grow revenue. Example: ($200,000 − $12,000 − $4,000 + $9,000) ÷ $200,000 = 96.5%. |
| Monthly-to-annual conversion | 1 − (1 − monthly rate)^12 | What a monthly rate means over a year. At 5% monthly churn, the compounded annual loss is about 46%. |
These examples use hypothetical numbers on the same starting base to show why the metrics answer different questions. Logo churn counts accounts equally, while revenue churn weights departures by the MRR they represent. Gross revenue retention is stricter than net revenue retention because it does not let expansion from growing accounts offset losses elsewhere; see net revenue retention. A strong net figure can coexist with shrinking revenue among smaller customers.
The annual conversion matters because monthly churn compounds. At 5% a month, about 54% of the starting customers remain after a year (0.95 to the power of 12), so roughly 46% are lost, not 60%. Monthly and annual plans also churn on different calendars, as annual vs monthly churn explains.
Denominator rule: Write down who counts before calculating. Specify whether trials, paused accounts, refunds and reactivated customers are included, and record the exact date used for the starting customer and MRR snapshots. Keep those definitions fixed from one reporting period to the next.
How do you split voluntary from involuntary churn?
A total churn figure bundles together departures that need different fixes. Voluntary churn follows a customer’s decision to cancel or not renew. Involuntary churn follows a payment failure, such as an expired card or a bank decline. Treating both as one outcome can send a product team searching for dissatisfaction when the immediate problem is billing recovery.
Use event records to classify exits. A cancellation request and declining usage before departure point toward a voluntary exit. A failed charge code on the final invoice points toward an involuntary one. In the example below, 20 of 400 customers left in a month: 14 canceled and 6 were lost to failed payments that were never recovered, so voluntary churn is 3.5% and involuntary churn 1.5%. The mix in your own data will differ; involuntary churn covers how to measure it.
How do you reconcile payment failures before counting churn?
A failed renewal is not always a lasting departure. Set a recovery window that matches your billing retry process, then count the customer as involuntarily churned only if they are still inactive when that window closes. If the payment succeeds and the customer remains active within the window, mark the event as recovered rather than as churn. This prevents a temporary billing problem from being recorded as a product rejection, or counted again when the customer returns. How retries work, and how long platforms keep trying, is in what a payment retry means.
In the billing table, keep the failure event, recovery status, cancellation status and reactivation status distinct. A refund or chargeback also needs its own record: it can create revenue loss without proving that the customer formally canceled. A pause or downgrade changes the account or its value, but should not be folded into logo loss.
| Churn type | Typical cause | How to confirm it | Retention lever |
|---|---|---|---|
| Voluntary churn | Price, low usage, unmet need, or a competitor | Cancellation reason and falling usage before exit | Improve onboarding or offer a relevant cancel-flow alternative |
| Involuntary churn | Expired card, insufficient funds, or bank decline | Payment failure code on the final invoice | Use automatic retries and card-update links |
| Recovered failure | Failed charge paid within the retry window | Customer active again before the window closes | Count as recovered, not churned |
| Downgrade or pause | Plan change or temporary break | Plan or status change in billing history | Track as contraction, separately from logo loss |
| Refund or chargeback | Dispute or refund request | Refund or dispute record | Track as revenue loss and review separately |
For a spreadsheet, report total churned, voluntary churned and involuntary churned net of recoveries as separate lines. That simple breakdown gives the team a clearer first question: are customers choosing to leave, or is a billing event interrupting an otherwise continuing subscription?
Which segments should you compare before trusting the average?
Two cancellations in a month sound minor until you learn they came from a twelve-customer plan tier. That is a 16.7% churn rate inside a segment that a company-wide average of a few percent would never surface. Averages smooth over exactly the pockets where churn concentrates, so compare like with like before calling a rate healthy.
The cuts worth running first, because each points at a different fix:
- Plan or price tier: a lower-priced plan may lose customers faster than a premium tier, while the premium tier accounts for more lost revenue per departure.
- Tenure: compare 0–30 days, 31–90 days, 91–365 days and over 365 days. A spike among new customers points at onboarding or first value (see onboarding, activation and early churn); a rise among long-standing accounts points at product drift or renewal friction.
- Acquisition channel: compare paid social with referrals. A high-volume channel can deliver customers whose expectations never matched the product.
- Contract type: monthly plans create a cancellation opportunity every cycle; annual plans concentrate risk into renewal windows.
- Country or time zone: payment outcomes vary with geography and renewal timing, and a pattern around local paydays or bank declines warrants a billing review.
- Payment method: card and direct-debit customers fail differently, and some card types carry more risk (see prepaid and virtual cards), so check recovery outcomes before blending them.
- Product or device: for an app, split iOS and Android. A gap narrows the investigation to a platform-specific experience.
Run the voluntary and involuntary split inside each segment as well. A premium tier losing customers to card declines needs retries and card-update links; the same rate driven by deliberate cancellations needs a product or pricing answer. Two segments can post identical churn rates and require opposite responses.
For every cut, show the segment rate beside the overall rate and print the customer count behind each. Set a tolerance for differences that trigger review, and treat small samples as a reason to collect more months of data rather than a signal to act.
Finally, rank segments by lost revenue as well as rate. A modest churn rate in a large, high-MRR segment can cost more than a dramatic rate in a small one. Give the three segments with the largest revenue loss their own review, so a striking percentage never pulls attention from the larger commercial hit.
How do cohorts and survival curves show when customers leave?
Two subscription businesses can report the same monthly churn rate and need opposite fixes. One loses customers in the first weeks because cards fail or onboarding stalls. The other looks stable for eleven months, then drops at renewal. Group customers by the month they started paying and follow each group’s active share at consistent milestones. A cohort curve starts at 100% and steps down as customers leave; its shape shows when to investigate before you argue about causes.
A steep early drop usually separates into different work: involuntary loss around the first charges, or voluntary exits when setup and first value miss. A level curve that falls near month 12 points to renewal readiness and the customer’s decision window. The curve does not prove cause. It narrows the calendar.
A worked example with illustrative values:
| Signup cohort | Active at month 1 | Active at month 3 | Active at month 6 | Active at month 12 |
|---|---|---|---|---|
| Oldest cohort in data | 97% | 94% | 91% | 78% |
| Middle cohort | 90% | 81% | 77% | 75% |
| Newest cohort with 12 months of data | 95% | 89% | 86% | 82% |
Read the oldest cohort as a renewal problem: stable through month 6, then a sharp slide into month 12. Read the middle cohort as an early-life problem: 10 points gone by month 1, another 9 by month 3, then a flatter path. The newest cohort reads as better early retention, with erosion still building later.
Fill each cell with the share still active at that milestone. If a cohort has not reached a milestone, leave it unobserved rather than entering zero. Customers still active when observation ends are right-censored: their time counts, but their outcome is unknown. A SAS conference paper on telecom churn models churn this way over a 15-month observation window, a reminder that unequal observation windows distort simple churned-or-not counts.
The Kaplan-Meier method estimates a survival curve while accounting for customers whose observation ends before they churn. A proportional-hazards model, such as the Cox model, can estimate how an attribute changes the instantaneous risk of leaving. If an attribute matters mostly around the first invoice or only at renewal, a single average hazard can hide that timing; stratify or test a time interaction before trusting one coefficient.
For a practical cohort review, group by signup month, record status at fixed intervals, estimate each curve, then compare cohorts by segment. The step-by-step build, with SQL, is in cohort retention analysis. Compare cohorts with similar observation periods. When two similarly sized groups show different curves, ask which segment mix, billing pattern or contract event changed before treating the company average as the explanation.
A cohort curve can expose when retention changes; it cannot identify the cause alone. Pair the timing with payment method, signup source, usage, contract structure and cancellation reason. Involuntary failures near billing events call for checks on recovery timing and card update friction. Voluntary exits near renewal call for value proof and an offer review. Test any change against a comparable cohort before you credit it with the improvement.
Which benchmarks can you compare against, and what are their limits?
A benchmark is useful only when business model, contract value, metric and period match. A monthly logo-churn rate for B2B SaaS cannot be set beside annual revenue retention or first-month cancellations among app subscriptions and treated as the same measurement. Use published figures to frame a question; resist turning them into a universal target.
- 01 Business model B2B SaaS, consumer app or ecommerce
- 02 Contract Monthly or annual, and the price point
- 03 Metric Logo or revenue, voluntary or involuntary
- 04 Period and definition Monthly or annual, and who counts in the base
| Source | Business model | Reported figure | Compare only with |
|---|---|---|---|
| UserMotion 2024 | More than 1,000 subscription businesses | Average churn of 4.2%, made of 3.5% voluntary and 0.7% involuntary (UserMotion benchmark) | B2B SaaS churn with a comparable definition and period |
| SaaS Capital 2025 | Private B2B SaaS companies | Median gross revenue retention of 91% and net revenue retention of 101% (SaaS Capital retention benchmarks) | Revenue retention comparisons only |
| Lighter Capital 2025 | 155 private B2B SaaS startups, data through 2024 | Median customer churn of 16.25% (Lighter Capital benchmark) | Early-stage B2B SaaS startups |
| RevenueCat 2025 | Subscription apps | Nearly 30% of annual subscriptions canceled in the first month (RevenueCat report) | Annual mobile app subscriptions |
| Recurly research published in 2025 | Ecommerce subscription customers in the United States and Europe | 74% of ecommerce subscribers dropped off after their first year, from a survey of 1,000 consumers (Internet Retailing coverage) | Consumer ecommerce subscriptions |
Read each row by unit and population. UserMotion separates voluntary and involuntary churn, so its involuntary share gives you a distinct check on payment failure rather than product dissatisfaction. SaaS Capital belongs in revenue-retention conversations, and says little about seat-based enterprise logo churn, where one lost account can erase the gains of several expansions. Lighter Capital describes private startups; RevenueCat measures the first month of annual app plans; Recurly’s survey measures first-year consumer ecommerce drop-off. A company-wide average can hide all of that mix.
A benchmark earns its place when it changes the comparison you run next. Match customer type, contract structure and metric first. Then compare your segment with the closest row and record every mismatch: annual versus monthly billing, consumer versus B2B, logo versus revenue, voluntary versus involuntary. Recurly’s public figures by industry and price point are collected in churn rate benchmarks.
The simplest useful check is your monthly logo churn beside the monthly rate of a matched benchmark, with your involuntary share reported separately. If the benchmark is annual, revenue-based or drawn from a different customer population, keep it as context and build the target from your own cohorts.
How do you find the factors that move churn?
Once you know which customers are leaving, compare what leavers had in common with what stayers had. Useful fields include onboarding completion, early active usage, seats used versus seats purchased, support tickets, NPS, signup source and payment method. The point is not to collect every attribute. It is to find a difference that gives you a testable question.
Say, as a hypothetical example, that 55% of customers who leave completed onboarding, compared with 80% of those who stay. That gap suggests an investigation: perhaps incomplete onboarding goes with higher churn, or perhaps another difference between the groups explains both. Check how many customers sit behind each share before prioritizing the result. A small sample can make a large-looking difference unstable. To gather cancellation context, use a cancellation reasons survey; for models that score every customer, see predicting churn with customer data.
Test before you standardize: If incomplete onboarding is associated with departures, change one part of the onboarding path for a defined group of new customers. Compare that group’s cohort curve with a control group over the same observation period. A comparison can identify a promising factor; a controlled test shows whether changing it makes a difference.
Keep a short log with the attribute, observed sample sizes, interpretation, proposed test and outcome. That record makes the next review a continuation of evidence gathering, not a fresh round of guesses. It also prevents a correlation from becoming a policy before anyone has tested the mechanism.
How do you run a monthly churn review?
Make the monthly review repeatable. Use one set of definitions, preserve the starting snapshot, and assign an action to the clearest driver rather than ending with a dashboard of unexplained changes. The metrics themselves are laid out in the churn dashboard.
- 01 Starting base MRR and customers on a fixed snapshot date
- 02 Logo churn Customers lost ÷ customers at start
- 03 Revenue retention Lost MRR, gross and net retention
- 04 Split the churn Voluntary, involuntary, recovered
- 05 Biggest losses Segments with the most lost MRR
- 06 Owner and test One person, one next action, logged
- Pull the starting base. Record MRR and customer counts at period start, using the same fixed snapshot date each month.
- Calculate logo churn. Count customers lost during the period and divide by the starting customer count.
- Calculate revenue retention. Measure lost MRR, gross revenue retention and net revenue retention, with consistent definitions for contraction and expansion.
- Split the churn. Separate voluntary from involuntary exits and keep recovered payment failures distinct.
- Review the biggest revenue losses. Give the segments with the largest lost MRR a review, not only those with the highest churn rates. Compare each with the closest matching benchmark and note the limits of that comparison.
- Assign an owner and a test. Give each confirmed factor one person responsible and one next action, then record the result in the same review log.
If involuntary churn stands out, start with billing operations. Check for expired cards and failed payments before renewal dates, then confirm that automatic retries and card-update links are running. Close the review by recording the metric definitions, segment findings, recovery outcomes and assigned tests in one place. That gives the next monthly review a consistent baseline and makes it easier to see whether an intervention changed the pattern.
FAQ
What is churn rate analysis?
It is the work of breaking one churn rate into its causes and locations: voluntary versus involuntary, by segment, and by when in the customer’s life the departure happens. The goal is a specific, testable next action rather than a single number to watch.
What is the difference between logo churn and revenue churn?
Logo churn counts lost customers, each one equal. Revenue churn counts lost MRR, so a large account weighs more than a small one. Report both: they can move in opposite directions.
How many customers does a segment need before its churn rate means something?
There is no fixed threshold. With twelve customers, one cancellation moves the rate by more than 8 points. Always print the customer count beside the rate, and merge months or segments until the count is large enough that one departure does not swing the result.
Should failed payments count as churn?
Only once the recovery window has closed. A failed renewal that is paid within your retry window is a recovered payment, not churn. Count it as involuntary churn only if the customer is still inactive when the window ends.
How NoChurn helps
NoChurn reads subscription events from your billing tool (Stripe in one click, Chargebee, Recurly, Paddle and others by connector, or any billing through the API), so voluntary and involuntary exits are separated from real billing events. You can send your own customer attributes in one API call, such as signup date, source, seats, usage, onboarding or NPS, and NoChurn shows which ones go with churn, with the sample size behind each result: correlation, not causation, and only a hint while samples are small. It detects when a customer removes a card shortly before renewal, raises the churn score and starts a save flow before the payment fails. To act on what the review shows, it runs failed payment recovery, cancel flows with an offer per reason and a one-click cancel always visible, and win-back emails at 30, 60 and 90 days. See Custom data & AI, churn score, failed payments and cancel flows.