Business Intelligence
11 min
Financial Dashboard: Examples From One Month-End Export
Choose useful finance KPIs, explain changes in profit and cash, and check the figures before your monthly review. Follow one downloadable, worked example.
Read More
Business Intelligence — Guide · 2026
Customer Churn Analysis: Metrics, Steps and Reducing Churn
What customer churn analysis is, the metrics and formulas to use, seven steps worked on a downloadable agency sample, and how to reduce churn before renewal.
Updated 16 min readQuick answer — customer churn analysis
Customer churn analysis measures how many customers stopped buying from you, how much recurring revenue left with them, and what they had in common beforehand. To do one, write down a churn rule, calculate churn among customers active at the start of each period, and compare groups with uncertainty ranges. Then check warning signs before renewal and act first where the most revenue is at stake.
Your Monday retention report shows recurring revenue down again. Paid acquisition seems to have the highest churn rate, so marketing is considering a budget cut and finance wants to know how much revenue is at risk. Before either team acts, someone needs to check whether the channel is really the problem.
This guide follows that investigation from the first churn calculation to a list of renewals ranked by revenue. The worked example is a fictional marketing analytics agency with 1,500 client accounts on recurring retainers, and every number comes from three downloadable files you can check.
It is for anyone who owns retention or recurring revenue, from growth marketers and customer-success leads to finance analysts and founders.
What is customer churn analysis?
Customer churn analysis is the work of measuring customer losses and explaining where they come from. Churn means a customer stopped paying or stopped using your product under a rule you define, such as a cancelled contract. The analysis usually answers four practical questions:
- How many customers left, and how much recurring revenue went with them?
- Is churn getting better or worse?
- Which contract terms, acquisition channels or customer groups are losing the most?
- What changed before customers cancelled?
A churn rate alone answers only the first question. The others need the records behind it: contracts, acquisition channels, product usage and support history.
How to identify customer churn
You identify churned customers by applying one written rule to every customer record. The right rule depends on how customers leave:
- With contracts or subscriptions, a customer has churned when they cancel or do not renew. The cancellation date in your billing or CRM export marks the moment.
- Without contracts, as in an online shop, a customer has churned after a set time without activity, such as 90 days with no purchase or login.
- A failed payment that ends the account is involuntary churn, because the customer did not choose to leave. Count it separately from voluntary cancellations.
Apply the rule to customers who were active at the start of the period to list who churned. To spot customers who may leave soon, look for warning signs before cancellation, such as falling usage, as Step 5 shows.
Churn metrics and formulas, worked on the example
The table gives the main churn metrics, each formula and a worked figure from the example agency, counting voluntary cancellations only. Monthly retainer revenue is the fee each client pays every month.
| Metric | Formula | Worked on the example |
|---|---|---|
| Customer churn rate (logo churn) | Customers who cancelled during the period ÷ customers active at its start | March 2025: 25 ÷ 917 = 2.73% |
| Revenue churn | Monthly revenue of customers who cancelled ÷ monthly revenue at the start | March 2025: $4,013 ÷ $182,641 = 2.20% |
| Retention rate | 100% minus the churn rate, for the same customers and period | March 2025: 100% − 2.73% = 97.27% |
| Annual churn rate | Customers who cancelled during the year ÷ customers active on its first day | 2025: 140 ÷ 813 = 17.2% |
| Customer lifetime value (simple version) | Average monthly revenue per customer ÷ monthly churn rate | 2025: $202.10 ÷ 1.60% ≈ $12,600 of revenue |
| Cohort retention | Share of customers who signed up in the same month and are still active a set number of months later | 2024 sign-ups: 82.0% still active after 12 months (Step 7) |
Treat lifetime value as a rough planning figure. The 1.60% is the 2025 average monthly churn rate: 205 cancellations divided by 12,799, the twelve month-opening client counts added together. The $202.10 is what those clients paid on average each month. The formula assumes churn stays at that rate, which implies an average client lifetime of about 62 months, far beyond the two years in the sample. It also counts revenue, not profit. The files hold no cost or spend data, so customer acquisition cost (the cost of winning one customer) cannot be calculated here.
A churn flag is a yes-or-no column that marks whether each customer churned in a period under your rule. The example files have none, so you build it from cancel_date and cancel_reason, and every flag traces back to a dated record. A risk flag is different: it marks a customer who has not left but shows a warning sign, like the engagement-decline flag in Step 6.
How teams can use AI for churn analysis
AI makes follow-up questions faster. Instead of rebuilding a spreadsheet each time, you can ask how a number was counted, compare Paid with Referral clients, or list the accounts renewing in the next 90 days. Marketing, finance and customer success can each ask their own questions of the same data.
Speed does not make the answer right. The same month's churn can range from 3.64% to 6.00% depending on which churn formula is used (Flowsery). Customers who join mid-period can also distort the rate (Shopify Engineering). The tool can vary too: one study repeated the same data-analysis task 480 times and found substantial differences between runs (Cui & Alexander, 2026). OpenAI's help page advises users to review generated code, outputs and assumptions (OpenAI Help).
AI explores and explains; a person defines, checks and decides. AI does not turn an association into a cause, predict the next client to leave, refresh an uploaded file or push a list into a CRM.
The seven steps below show how to do a customer churn analysis on the example files, from writing the churn rule to scheduling a checked monthly report.
Step 1: Define churn and learn how to calculate churn rate
Start the churn data analysis with raw records rather than a file that already has a churn flag. The sample uses three files:
agency_client_accounts.csvholds contract term, monthly retainer revenue, acquisition channel, renewal, cancellation, and reactivation data.monthly_client_engagement.csvholds monthly report views and active client users.client_success_activity.csvholds dated support and client-success activity.
Download the three-file sample
agency_client_accounts.csv, monthly_client_engagement.csv, and client_success_activity.csv cover 1,500 synthetic agency client accounts over two years and reproduce every number below. No real customer data is included.
agency_client_accounts.csv · monthly_client_engagement.csv · client_success_activity.csv
Clean the export before calculating anything. Remove duplicate customer IDs so an account is not counted twice. If active customers and cancellations come from separate exports, compare the two lists for missing records before joining them.
Then make four decisions:
- Contract or no contract? This agency records cancellations, so the date is explicit. Without contracts you need an inactivity rule, and the window matters: on this sample, 30, 60, and 90 days without activity give churn figures of 18.9%, 17.5%, and 16.1%.
- Customers or revenue? Customer churn counts accounts, and revenue churn weights each loss by its monthly retainer. The top 10% of accounts in this sample hold about 37% of monthly revenue, but the clients who leave are close to average size. The two rates therefore stay within about one percentage point of each other.
- Voluntary churn or failed payment? Keep the two separate. Here, 43 of 318 cancellation events (13.5%) are failed payments. That is 13.5% of cancellation events, not 13.5% of customers.
- Gross or net of reactivations? March has 25 voluntary cancellations among 917 accounts active on the first day, a 2.73% gross churn rate. Fourteen of those clients reactivate later, and subtracting them after the fact gives a net rate of 1.20%. Use the gross rate in the monthly report, because at month-end nobody knows yet who will come back.
With those choices made, apply the customer churn rate formula from the metrics table, leaving out new sign-ups, and save the rule so every report counts churn the same way.
Step 2: Inspect the calculation before trusting the result
The first April result looks plausible: 22 voluntary cancellations divided by 958 accounts active on the first day, or 2.30%. Opening the calculation reveals the problem. Three of the 22 cancelled accounts signed up during April, so they were never among the 958 accounts the rate is divided by.
The correction is plain English: “Only count cancellations from customers who were active on the first day.” The count of cancellations, the top of the fraction, falls from 22 to 19, and the rate moves from 2.30% to 1.98%.
Apply the boundary consistently. A customer who signs up on the first day was not active when the period began, so exclude them. A customer who cancels on the first day was active at that boundary, so include them. This way, every cancelled customer you count is also part of the starting group you divide by.
Step 3: Check whether the churn trend is real
The agency's monthly churn rate falls from about 2.3% in January to 1.0% in December. That looks like an improvement, but clients in their first year cancel more often than older clients: about 1.6% to 1.8% a month, against 1.0% after the first year. When new sign-ups slow, the client base gets older, and the headline rate can fall even when retention has not improved.
Ask the AI to compare the raw rate with a rate adjusted for tenure, meaning how long each client has been a customer. That separates a real retention change from a change in customer mix.
Result: after accounting for tenure, no calendar month stands out from the others. That holds after correcting for multiple comparisons, which allows for the chance that one of twelve months looks unusual by luck alone. Marketing should not claim that retention improved, and finance should not build a forecast around the apparent decline.
How it was checked: voluntary cancellation is the event, failed payments are counted separately, and tenure is the time since a client first signed up. Anomaly's tenure-adjusted check and a separate calculation from the source rows reached the same conclusion.
Step 4: Test whether an acquisition channel is really the problem
To compare channels fairly, follow the 890 client accounts that signed up in 2024 for the same 12 months. Paid clients churned at 19.9%, against 17.6% for Referral. Each rate has a 95% confidence interval, the range the true rate probably falls in given the group's size. The Paid and Referral ranges overlap, so the data cannot rank the two channels.
Next, check whether another factor, here monthly versus annual contract term, explains the apparent channel gap. Before any adjustment, contract term shows a bigger gap: 25.0% of monthly-contract accounts churn within 12 months, compared with 11.5% of annual-contract accounts. Test channel and contract term together before acting on either pattern.
Result: the data does not support blaming Paid acquisition. The test of channel and contract together compares odds, the chance a client leaves set against the chance it stays, and an odds ratio of 1 means no difference. With contract term held constant, Paid clients have 0.73 times the churn odds of Referral clients. Their 95% range of 0.45 to 1.19 includes 1, so there may be no channel difference at all. Monthly contracts show the clearer association: 2.95 times the churn odds of annual contracts, with a range of 1.97 to 4.40 that stays above 1.
Report this as an association, not a cause: “Monthly contract terms are associated with higher churn, holding acquisition channel constant.” The marketing decision is to investigate the monthly-contract experience before cutting Paid spend.
Step 5: Track churn indicators before renewal
Customer churn indicators are measurable changes that tend to come before a cancellation, such as falling usage or more support tickets. To identify customer churn early, use only signals that exist before the client leaves. A note saying “asked about cancelling” records churn already under way, so treat client-success notes as context, not evidence.
To test an indicator fairly, choose a fixed date. Measure the signal before that date, then count cancellations afterward. At the 1 May 2025 checkpoint, 937 accounts have enough history. Their churn over the next four months rises from 3.2% among accounts with no support tickets in March and April to 7.7% among accounts with six or more.
The pattern points one way, but the ranges overlap, so do not turn the four groups into a precise risk score. A drop against each client's own normal is worth testing too. Among clients who cancelled for good, report views in the two months before cancellation average about half of the two months before that.
That look-back covers only clients who left. A fixed-date check also covers clients who stayed. On 1 September 2025, 1,061 active accounts had report views for May to August, and 530 of them viewed fewer reports in July and August than in May and June. Of those 530, 7.2% cancelled for good by the end of December, against 2.3% of the 531 whose views held steady. The 95% ranges, 5.3% to 9.7% and 1.3% to 3.9%, do not overlap.
Can you predict which customers will churn 90 days before renewal?
You can rank them, which is not the same as predicting them. A fixed-date check like the one above shows which warning signs separate leavers from stayers, and Step 6 applies one to every annual client renewing in the next 90 days. Even in the declining group, 92.8% of clients did not cancel for good by December. Predicting individual clients needs the extra tests covered in “Can AI predict churn?” below.
Step 6: Turn the analysis into a revenue-ranked action list
A churn analysis does not reduce churn by itself. The useful output is a list of client accounts, renewal dates, monthly retainer revenue and the reason each account needs attention. Sort by revenue at stake so marketing, finance and client success can agree where to start.
On 1 September 2025, 122 annual client accounts renew within the next 90 days. Together they pay $28,063 in monthly retainer revenue. The list flags an account when its July and August report views add up to less than its May and June views. Sixty-seven accounts carry that flag, and they pay $14,301 of that monthly revenue. Because any drop counts, about half of the accounts get the flag, so use it to decide whom to call first rather than as a forecast.
The map helps decide whom to call first; it does not predict who will leave. The ranked view below names the twelve largest upcoming retainers.
Step 7: Verify the numbers, then schedule the report
Before the result reaches leadership, re-derive the important numbers from the source rows. Verification should confirm what was checked and flag weak evidence, such as a cohort that is too small to interpret.
Cohort retention groups client accounts by signup month and shows the share still active after 1, 3, 6, 9, and 12 months. To make the boundary reproducible, an account permanently cancelled on its exact signup anniversary counts as lost at that point.
The twelve cohorts that signed up in 2024 can each be followed for a full year. Across them, average retention is 98.2% after one month. It falls to 95.6%, 92.0%, 88.8%, and 82.0% after 3, 6, 9, and 12 months. Six of the sample's 24 monthly sign-up cohorts (12 in 2024 and 12 in 2025) contain fewer than 30 accounts, so treat their movement as uncertain rather than as a marketing-channel verdict.
Once the numbers hold, put the churn rate, cohort retention, channel and contract views, indicators and renewal list in one dashboard that uses the saved definition.
On a paid plan, the dashboard can become a monthly emailed report. Set the reporting question, schedule it for the first day of the month, and send it to the people who act on it.
For subscription software teams, the same workflow produces a defensible SaaS churn rate beside the other SaaS metrics you can track with AI, including activation, expansion, and revenue retention. If the source is an uploaded file, upload the latest export before the scheduled run. Otherwise, the report will use the last file available.
Can AI predict churn?
Not from the analysis in this guide, which explains past churn and finds signals that came before cancellations. A model that predicts which customer will leave needs a long activity history and enough churned customers to learn from. It also needs a held-back test group the model never saw while it was built. Ask any tool that gives individual churn scores how it performed on such a group.
Even a good prediction does not explain why a customer leaves. An engagement drop can help rank risk without proving its cause. To act on risk you can measure today, use the renewal list from Step 6 and the actions below.
How to reduce churn based on what the analysis finds
In the example, each finding points to a different action:
- Review the monthly-contract experience. Monthly-contract clients churned at 25.0% within a year, against 11.5% for annual contracts. Because that is an association, test a change, such as an annual-contract offer, on some new monthly clients first.
- Give first-year clients more attention. Clients in their first year cancelled more often each month than older clients, so plan check-ins and training early, then check whether first-year churn falls.
- Fix failed payments separately. Forty-three of the 318 cancellation events were failed payments. Those clients did not choose to leave, so start with payment reminders and card-update requests rather than a retention offer.
- Call the flagged renewals first. The 67 annual clients with declining report views pay $14,301 a month between them. Contact them in revenue order and ask what changed before the renewal conversation.
- Do not cut Paid acquisition on this evidence. The channel gap disappeared once contract term was considered, so cutting Paid spend would act on a pattern the data does not support.
Measure each action with the same churn rule, and keep a comparison group that does not get the change. Without it, a lower rate next month could just reflect a change in customer mix, as Step 3 showed.
Eight checks before you trust an AI churn number
Run these checks before a churn number goes into a report. Each one comes from a step above:
- Write down the churn rule and reuse it in every report.
- Count only customers who were active when the period began.
- Report gross cancellations separately from later reactivations, and failed payments separately from voluntary cancellations.
- Adjust a trend for customer tenure before calling it an improvement.
- Show uncertainty ranges around contract-term, channel, and cohort comparisons.
- Check at least one plausible hidden factor before blaming a segment.
- Measure indicators before the outcome and compare clients who left with clients who stayed.
- Rank renewals by revenue at stake and re-derive the key numbers from source rows before scheduling the report.
The result is a churn number finance can defend, a channel comparison marketing can use, and a renewal list client success can act on.
Frequently asked questions
How do you calculate churn rate in Excel?
Use two COUNTIFS formulas. Count customers who signed up before the first day of the month and have no cancel date or cancelled on or after that day. Then count voluntary cancellations that month among customers who signed up before it began, and divide the second count by the first. In the example file, March 2025 gives 25 ÷ 917 = 2.73%.
Is 20% churn high?
There is no universal benchmark. The answer depends on the period, business model, billing or contract term, customer size, and churn definition. Compare the rate with your own history and customer segments before comparing it with an external number.
What is the difference between churn and attrition?
In customer analytics, customer attrition analysis and customer churn analysis usually mean the same thing. “Attrition” can also describe employees leaving, which is a separate HR use case.
Can ChatGPT do churn analysis?
Yes, it can explore customer exports and calculate churn. Define churn explicitly, ask to see the calculation and assumptions, and independently re-check the important numbers. Repeated runs can differ, so do not treat the first fluent answer as final.
What should I remove from a customer file before uploading it?
Remove names, emails, phone numbers, and identifying free text unless they are essential. Keep a stable customer ID so files can still be joined. If you do not own the customer data, get permission before uploading it to any AI tool.
Anubhav
Product Growth and UX at Anomaly AI
Anubhav works on product growth and UX at Anomaly AI, focused on AI products, B2B SaaS and enterprise workflows.
LinkedInRelated Articles
Continue exploring AI data analysis with these related insights and guides.
Business Intelligence
11 min
How to Choose Financial Reporting Software When Your Numbers Live in Excel
A vendor-neutral way to choose financial reporting software when your numbers live in Excel: three routes, the criteria that decide, and a worked example you can reproduce.
Read More
Business Intelligence
10 min
How to Build an Inventory Aging Report From Receipts
Turn receipt and issue records into an inventory aging report — surviving stock aged as of a date into 0-30, 31-60, 61-90, and 91+ day buckets under a stated FIFO assumption, reconciled to units on hand.
Read More
