Your renewal numbers dipped last quarter, and the question from leadership is simple: who is leaving, and why? A churn analysis answers it with data you already have, from billing and CRM records to survey scores and support tickets. This guide, written for insights and CX teams rather than data scientists, covers the churn rate formula, the main methods and how to link feedback to churn. It is part of our wider guide to customer experience analytics.
Key takeaways
- In any churn analysis, churn rate = customers lost during a period ÷ customers at the start of that period × 100. Leave out customers acquired during the period.
- Split voluntary churn (the customer chose to leave) from involuntary churn (a failed payment or expired card); they need different fixes.
- Cohorts, segment comparisons and survival curves show who leaves and when. Linking churn to NPS, CSAT and exit surveys shows why.
- A link between low scores and churn is an association, not proof of cause, and survey respondents are not all of your customers.
- Turn each finding into a retention action with a measurable target, and compare against a group that did not get the action.
What is customer churn analysis?
Churn is the loss of a customer: a cancelled subscription, a contract that is not renewed or, in non-contract businesses, a customer who stops buying. Customer churn analysis measures how much churn you have, then breaks it down by who, when and why.
The first two questions come from behavioral records such as billing, CRM and product usage. The third needs feedback: relationship and transactional surveys, support tickets, reviews and exit surveys. Adding feedback is what tells you what to change.
How do you calculate churn rate?
The standard formula is:
Churn rate = customers lost during the period ÷ customers at the start of the period × 100
Count only customers who were active on day one. New customers who join and leave within the same period distort the rate, so track them separately as early churn. Retention rate is the complement: 100% minus the churn rate.
Voluntary vs involuntary churn
Voluntary churn is a decision: the customer cancels, switches or stops buying. Involuntary churn happens without one, usually through a failed payment, an expired card or a closed account. Report the two separately. Involuntary churn is usually fixed by billing processes such as payment retries and reminders. Voluntary churn is where feedback analysis earns its keep.
Which churn analysis methods should you use?
Four methods cover most needs. Each answers a different question.
| Method | Question it answers | What you need |
|---|---|---|
| Cohort analysis | Do newer customers stay longer than older ones? | Start date and churn date per customer |
| Segment comparison | Which customer groups churn most? | Churn flag plus plan, region, channel or tenure |
| Survival curves | When in the customer lifetime does churn peak? | Start date and churn date, including customers still active |
| Feedback linkage | Why do customers leave? | Survey scores, open ends or tickets matched to customer IDs |
Cohort analysis groups customers by the month or quarter they joined and tracks each group’s churn over time. If the January cohort keeps customers longer than the October cohort, something improved, perhaps onboarding.
Segment comparison sets churn rates side by side by plan, region, tenure or acquisition channel. Compare rates, not counts, because large segments always lose more customers.
Survival curves plot the share of a cohort still active at each month since joining. A steep early drop points to onboarding problems; a drop at month 12 points to the first renewal. The method also counts customers who have not churned yet, rather than ignoring them.
Feedback linkage matches survey scores and comments to the same customer IDs, then compares churn by score group. This is where the customer experience metrics you already track, such as NPS and CSAT, become predictors of loss rather than scores on a slide.
Where do prediction models fit?
A churn prediction model scores each active customer on how likely they are to leave, based on patterns in past churners: falling usage, late payments, support contacts, low scores. The output is a ranked list for outreach. Build it after the descriptive work, because the model can only learn from data you have already cleaned, defined and understood.
What does a worked churn analysis look like?
The following numbers are illustrative. A company starts the quarter with a cohort of 2,000 customers. Earlier, 1,200 of them answered a relationship NPS survey; 800 did not. By the end of the quarter, 210 had churned.
| Group | Customers at start | Churned | Churn rate |
|---|---|---|---|
| Promoters (9–10) | 500 | 20 | 20 ÷ 500 = 4.0% |
| Passives (7–8) | 400 | 36 | 36 ÷ 400 = 9.0% |
| Detractors (0–6) | 300 | 66 | 66 ÷ 300 = 22.0% |
| Did not respond | 800 | 88 | 88 ÷ 800 = 11.0% |
| Total | 2,000 | 210 | 210 ÷ 2,000 = 10.5% |
Detractors churned at 22.0%, 5.5 times the promoter rate (22.0 ÷ 4.0). They were 25% of respondents (300 ÷ 1,200) but 54.1% of churn among respondents (66 ÷ 122).
Of the 66 churned detractors, 40 answered an exit survey. Coding their open ends gave four themes:
| Exit-survey theme | Mentions | Share of 40 |
|---|---|---|
| Price increase at renewal | 16 | 40% |
| Slow support resolution | 12 | 30% |
| Missing feature | 8 | 20% |
| Other | 4 | 10% |
The reading: price and support account for 70% of the reasons given by churned detractors (28 ÷ 40). That is a short list to test, not a verdict. Note the 800 non-respondents, who churned at 11.0%, above the passive rate. Silence is a churn signal of its own.
Why doesn’t churn analysis prove cause?
Detractors churn more, but a low score may be a symptom rather than the cause. A customer whose budget was cut may score low and leave for the same reason. Treat the link as a lead, then test it with a change and a comparison group.
Survey respondents are also not all of your customers. In the example, 40% of the cohort never answered, and only 40 of 66 churned detractors explained why. Report response rates next to every finding. Then check behavioral data, such as usage and tickets, for the customers who stayed silent.
How do you turn churn analysis into retention actions?
Each finding should become an action with an owner, a target group and a measure.
- Pick the driver with the biggest reachable group. In the example, renewal pricing touched 40% of churned detractors.
- Define the action and the group. For example, a renewal call with a pricing review for detractors due to renew next quarter.
- Hold back a comparison group. Give the action to a random share of the group and not the rest, so you can see the difference it makes.
- Measure churn in both groups over the same period, with the same churn definition.
- Feed the result back into the next analysis and your voice of customer program.
To see these drivers continuously instead of once a quarter, map what drives satisfaction and loyalty by segment and journey so issues surface before they become churn. Our guide to voice of customer analysis covers coding exit-survey and support verbatims into themes.
What are the common mistakes in churn analysis?
- Including new customers in the denominator. Adding customers who joined mid-period lowers the rate and hides early churn.
- Mixing voluntary and involuntary churn. A billing problem then looks like a service problem, and the wrong team gets the brief.
- Comparing counts instead of rates. The largest segment almost always loses the most customers in absolute terms.
- Reading exit surveys as the whole story. Churned customers who answer are a minority, and they often give the easiest reason, which is price.
- Changing the churn definition between reports. A new inactivity window or period length breaks every trend line.
How do you do churn analysis in practice?
Most of the work is joining records: billing, CRM, product usage, survey responses and support tickets, matched on one customer ID. If those sources sit in separate systems, connect customer, survey and operational data in one governed layer so the churn flag and the feedback describe the same people.
For your next step, pull last quarter’s starting customer list, flag who churned, split voluntary from involuntary, and compare churn rates by NPS group. That one table will tell you whether feedback-led retention work is worth the investment.