Customer retention
Churn analysis
On this page
- Four things the analysis should show
- What the analysis should deliver
- Two types of churn
- Define churn
- Collect historical data
- Collect data on active customers
- Descriptive analysis (what happened)
- Comparative analysis (differences between groups)
- Follow customer groups over time with cohort analysis
- Root cause analysis (why it happened)
- Predictive analysis (who will churn)
- Shared traits and signals
- Step 1: Define churn and collect data
- Step 2: Calculate the baseline
- Step 3: Compare churned and active customers
- Step 4: Find warning signs
- Step 5: Create risk scores
- Step 6: Automate the tracking
- Step 7: Validate and adjust
- A hypothetical SaaS company with a high churn rate carries out the analysis
- Analytics
- Customer success platforms
- BI tools
- Machine learning platforms
- How to go from analysis to action
- Contact us about churn analysis
Churn analysis is a systematic review of which customers leave, when they leave and why. The goal is to understand what has already happened and to find warning signs, so that you can act before the customer decides to leave.
To keep customers, you first need to understand where you stand. How high is the churn rate? Which customers leave? Are there shared patterns that set those who leave apart from those who stay?
When you collect historical data on churned customers and compare it with data on customers who stayed, you can see which behaviours, signals and moments come before churn. Based on those patterns, you can take action in time to get at-risk customers to stay.
What is churn analysis?
Four things the analysis should show
A churn analysis systematically evaluates
- Which customers leave (demographics, segment, value)
- When they leave (point in the customer journey, season, after an event)
- Why they leave (the reasons behind the decision)
- Warning signs that come before churn (behaviour before the decision)
What the analysis should deliver
- Understanding
- What causes churn in the business?
- Prediction
- Which customers are at risk right now?
- Prevention
- Which actions can prevent churn?
Two types of churn
Voluntary churn: The customer actively chooses to leave.
Examples
- Cancels the subscription
- Switches supplier
- Stops buying products
Involuntary churn: The customer wants to continue but cannot.
Examples
- The payment method fails (expired credit card)
- Moves abroad (service not available)
- The customer has died
Voluntary churn can be influenced through retention work.
How to collect data for the churn analysis
Define churn
The first step is to define exactly what counts as churn.
The definition depends on the industry
Subscription service
- The customer cancels the subscription = churn
- The customer pauses but does not cancel = not churn (yet)
E-commerce
- The customer bought regularly but has not made a purchase in X months = churn
- (X depends on the product category: 3 months for food, 12 months for electronics)
B2B service
- The contract is not renewed = churn
- The contract is renewed but at a lower level = partial churn (reduced revenue)
Important: Use the same definition throughout, so that the data can be compared over time.
Collect historical data
Time period: At least six to twelve months of churn data to find patterns.
Data on churned customers
From the CRM or customer register
- Customer ID
- Customer since (start date)
- Churn date
- Total time as a customer
- Total amount spent (LTV)
- Segment/category
- Demographics (age, location, company size for B2B)
From transaction systems
- Purchase history (frequency, amount, products)
- Latest purchase (how long ago?)
- Average order value
- Total number of orders
From the website or app (if the customer is logged in)
- Number of visits in the past 30, 60 and 90 days
- Number of logins
- Pages visited
- Features used (for SaaS)
- Time on the platform
From customer support or customer service
- Number of contacts
- Type of contacts (questions, complaints)
- Time of the latest contact
- Resolved and unresolved problems
From other sources
- Details the customer has provided, such as a change of address
- Changed payment methods
- Participation in surveys and feedback
Analysing individual customers’ behaviour and demographics is processing of personal data and requires a legal basis under the GDPR. Collect only the data you need.
Collect data on active customers
Comparison group: To identify what is unique to churned customers, you need the same data for customers who have not churned.
Collect the same data points
- Demographics
- Purchase history
- Behaviour on the website or in the app
- Contacts with customer support
Result: Two datasets, churned and active, ready for comparison.
How to analyse churn
Descriptive analysis (what happened)
Basic statistics
Churn rate: (Number of churned customers ÷ Total customers at the start of the period) × 100
Example
- 1,000 customers at the start of the month
- 50 churned
- Churn rate = 5 per cent
Segmented churn rate: Compare churn between segments.
Example
| Segment | Customers | Churned | Churn rate |
|---|---|---|---|
| New (<3 months) | 300 | 30 | 10 per cent |
| Active (3-12 months) | 500 | 15 | 3 per cent |
| Loyal (>12 months) | 200 | 5 | 2.5 per cent |
- Insight
- New customers churn at four times the rate of loyal ones. Focus on onboarding.
- Timing of churn
- When in the customer journey do people leave?
Example
- 40 per cent leave within the first three months
- 30 per cent leave at renewal after 12 months
- 20 per cent leave at renewal after 24 months
- 10 per cent leave at other, scattered points
Insight: The critical moments are onboarding and the annual renewals.
Comparative analysis (differences between groups)
Find what churned customers have in common
Behavioural differences
Example (SaaS)
| Metric | Churned | Active | Difference |
|---|---|---|---|
| Logins/month | 2 | 15 | Churned customers use it less |
| Features used | 1-2 | 4-6 | Churned customers explore less |
| Support contacts | 5 | 1 | Churned customers have more problems |
| Time since latest activity | 20 days | 3 days | Churned customers stay inactive for longer |
Insight: Low usage and many support tickets are warning signs.
Demographic differences
Example (e-commerce)
- Churned: 70 per cent men, 60 per cent aged 18-25, 80 per cent Stockholm
- Active: 50 per cent men, 40 per cent aged 18-25, 30 per cent Stockholm
Insight: Men, young people and Stockholm residents are over-represented among churned customers. Analyse the combinations of gender, age and location before you draw conclusions about, for example, young men in Stockholm.
Follow customer groups over time with cohort analysis
Analyse churn per cohort (a group of customers who started at the same time).
Example
Cohort Jan 2025: 100 new customers
- After one month: 90 remain (10 per cent churn)
- After three months: 75 remain (25 per cent churn in total)
- After six months: 60 remain (40 per cent churn in total)
- After twelve months: 50 remain (50 per cent churn in total)
Cohort Feb 2025: 100 new customers
- After one month: 85 remain (15 per cent churn, worse than the January cohort)
Insight: Something changed between January and February and increased early churn. Find out what.
Root cause analysis (why it happened)
Exit surveys: Ask customers who cancel why they are leaving.
Common answers
- Too expensive
- Found a better alternative
- No longer use the product (their needs have changed)
- Poor customer service
- The product or service did not work as expected
Categorise the answers: Create categories and count how common each reason is.
Example
| Reason | Share |
|---|---|
| Price | 35 per cent |
| Competitor | 25 per cent |
| Need no longer exists | 20 per cent |
| Poor service | 15 per cent |
| Technical problems | 5 per cent |
- Insight
- Price and competitors account for 60 per cent of churn. Review your pricing strategy and what sets you apart from the competition.
- Analysis of support tickets
- Review complaints from churned customers before they left.
- Identify recurring problems
- If many churned customers complained about the same thing before leaving, it is probably a main cause.
Predictive analysis (who will churn)
- Machine learning models
- Train a model on historical data to predict future churn.
- Input
- All data points (behaviour, demographics, transactions, support).
- Result
- Churn risk per customer (0-100 per cent).
- Use
- Identify customers with a high churn risk and target efforts to keep them.
Requires
- A large dataset (at least 1,000 customers, ideally 10,000)
- Data science skills
- Tools (Python, R or ML platforms)
An alternative for smaller companies: Rule-based scoring instead of ML.
Example
How the risk score is calculated
- No activity in the past 30 days: +40 points
- Support ticket in the past 14 days: +30 points
- Purchase in the past 60 days: -20 points
- Loyal customer (>12 months): -30 points
Add up the points and count a negative total as 0. Then place the total in a band. The scale goes up to 100, so there is room for more factors than the examples use
- 0-30: Low risk
- 31-60: Medium risk
- 61-100: High risk
Identify warning signs
How to do a churn analysis in seven steps
Step 1: Define churn and collect data
Define churn: What counts as churn in the business?
Collect data
- Churned customers (6-12 months of history)
- Active customers (the same data points)
- Sources: CRM, transaction systems, website, customer support
Step 2: Calculate the baseline
- Current churn rate
- What is the churn rate today?
- Segmented churn rate
- How does churn differ between segments?
- The cost of churn
- How much revenue is lost per month and per year?
Example
- Churn rate: 5 per cent a month
- 1,000 customers × SEK 300 a month = SEK 300,000 in monthly recurring revenue (MRR)
- Monthly revenue lost in the first month: 300,000 × 0.05 = SEK 15,000
- After twelve months, with no new customers: around SEK 138,000 less in monthly revenue (300,000 × (1 − 0.95 to the power of 12))
- Accumulated revenue loss over the year: around SEK 980,000 (the sum of the twelve months’ losses)
Step 3: Compare churned and active customers
Identify differences
- Behaviour
- Demographics
- Transaction patterns
- Support contacts
Document: List all significant differences.
Step 4: Find warning signs
Which behaviours come before churn?
- Inactivity
- Lower purchase frequency
- Support complaints
How long before churn do the signals appear?
- 30 days? 60 days? 90 days?
How common are the signals?
- Do they appear in 80 per cent of churned customers? Or 20 per cent?
Prioritise the strongest signals: Those that are most common and appear earliest.
Step 5: Create risk scores
Define the risk score: Which factors add points?
Example
Raises the risk
- +30: No activity for 30 days
- +20: Support ticket with a complaint
- +15: Lower purchase frequency
- +10: Visited the pricing page
Lowers the risk
- -20: Purchase in the past 30 days
- -30: Loyal customer (>12 months)
- -10: Positive NPS response
Add up the points and count a negative total as 0. Then place the total in a band. The scale goes up to 100, so there is room for more factors than the examples use
- 0-30: Low risk
- 31-60: Medium risk
- 61-100: High risk
Step 6: Automate the tracking
Automate the identification: Let a system calculate risk scores continuously for all customers.
Platforms
- CRM with custom fields
- Customer success platforms (Gainsight, ChurnZero)
- Custom dashboards (Tableau, Power BI)
Alerts: A notification when a customer moves into the “High risk” category.
Step 7: Validate and adjust
- Test the predictions
- How many of the “High risk” customers churn?
- Compare with the baseline
- Do high-risk customers churn clearly more often than the average customer? Only if they do is the model working.
- Coverage
- What share of all churned customers had the model flagged in advance?
- Control group
- Leave a randomly selected share of high-risk customers without any intervention. Only then can you separate the effect of the intervention from the accuracy of the model.
- Adjust
- Change the scores based on the actual outcome and keep improving the model.
What a complete churn analysis can look like
A hypothetical SaaS company with a high churn rate carries out the analysis
Current situation
- 2,000 customers
- 10 per cent churn a month (200 customers lost)
- MRR: SEK 500 per customer
- Lost recurring revenue: SEK 100,000 a month
Step 1: Collect data
- Churned customers (the past six months)
- 1,200 customers
- Active customers
- 2,000 customers
Step 2: Compare the groups
| Metric | Churned | Active |
|---|---|---|
| Logins/month | 3 | 20 |
| Features used | 1 | 5 |
| Support tickets/month | 2 | 0.3 |
| Days since activity | 25 | 2 |
Step 3: Find the warning signs
Signal 1: No login for 14+ days
- 85 per cent of churned customers had this
- 10 per cent of active customers had this
Signal 2: Uses only one feature
- 90 per cent of churned customers
- 15 per cent of active customers
Signal 3: Two or more support tickets in 30 days
- 70 per cent of churned customers
- 5 per cent of active customers
Step 4: Set risk scores
Points
- +40: No login for 14+ days
- +30: Uses only one feature
- +20: 2+ support tickets
- -30: Logged in during the past 7 days
- -20: Uses at least four features
Step 5: 400 customers are identified as “High risk”
Action
- Email campaign: “We noticed you haven’t used feature X, so here’s a guide”
- Personal contact from customer success with high-value customers
- Offer a one-to-one onboarding session
Hypothetical result after three months
- Churn fell from 10 per cent to 7 per cent
- 60 customers saved = SEK 30,000 in retained MRR
- ROI: The efforts cost SEK 15,000 and retained SEK 30,000 in MRR (annual impact: SEK 360,000)
Tools for churn analysis
Analytics
- Google Analytics / Mixpanel
- Behavioural data on the website and in the app.
- CRM (Salesforce, HubSpot)
- Customer data, transaction history, segmentation.
Customer success platforms
Gainsight, ChurnZero
- Churn forecasts
- Health scores
- Automated workflows
Use: For SaaS and subscription businesses.
BI tools
- Tableau, Power BI, Looker
- Custom dashboards for churn analysis.
- Use
- Visualise churn data, segments and trends.
Machine learning platforms
- H2O.ai, DataRobot, Azure ML
- Build predictive models.
- Use
- Large companies with data science teams.
How to go from analysis to action
The churn analysis shows where and why customers leave. The next step is to act on the results.
Process
1. Churn analysis
- Identify where and why customers churn
- Find warning signs
2. Audience segmentation
- Segment customers by churn risk
- High, medium or low risk
3. Channel matrix
- Plan communication per segment
- What is said, in which channel and when
4. Carry it out
- Send campaigns to keep customers
- Personal contact with high-value customers
5. Measure
- Did the churn rate fall?
- How many customers were saved?
6. Adjust
- Analyse continuously
- Improve the models based on new data
Contact us about churn analysis
Carnaby helps companies carry out systematic churn analyses, from data collection and identifying warning signs to implementing predictive models that reduce churn.
Want to know where you stand today?
Call us, or fill in the form and we will get back to you.