Churn prediction is one of several measures to reduce churn by providing you with data about which customers are more likely to cancel their subscription. Find out how to use churn forecasting for improved customer experiences and higher retention and prevent churn before it happens.
6 steps for successful churn prediction #
- Identify risk segments
- Identify behavioral signals that indicate churn risks
- Build a predictive model for churn prediction and LTV
- Turn insights into actionable KPIs
- Act on macro and micro levels
- Use churn prediction for campaign optimization
Forecasting churn enables you to: #
- Prioritize high-value (e.g. with a high lifetime value – LTV) accounts at churn risk
- Improve friction and gaps in the customer journey
- Adjust CAC (customer acquisition costs) based on predicted LTV
In the following, we will go into details regarding the six steps for churn prediction.
Step 1: Identify risk segments #
Define historical churn rates based on high-level data. High-level churn data is static data that doesn’t change over time.
Examples:
- Signup cohort (0-3 months, 3-6 months, 6+ months)
- Date of signup
- Subscription plan
- Demographic data (country, age, industry, etc.)
- Campaign source
Analyze churn rates across these segments and identify where churn is happening and which segments are most at risk.
High churn in the first 3 months
Early churn can indicate that expectations don’t match your actual product or service. This could mean that your messaging might be at odds with your actual product/service value. Another reason could be that certain discounts, freemiums or free trials attracted the wrong user type (who just wanted to use your product/service temporarily) or that there are technical payment issues once the paid subscription starts (read more about involuntary churn here).
Since you often see a higher churn rate in the first three months, it is better to differentiate different signup cohorts for clearer insights. This way, you can easily determine whether something was wrong during the “onboarding” phase (e.g. wrong campaign audience, false expectations, issues during the onboarding). Connecting new customers to specific campaigns improves measuring results and return on investment.
High churn in later cohorts
Later cohorts that churn tend to have different churn reasons that require another approach (e.g. support issues, pricing, new competitors, etc.) which we will discuss in the next step.
Step 2: Identify behavioral signals that indicate risk of churn #
To get the full picture, churn prediction also requires so-called low-level behavioral data to learn more about individual and clustered user behavior that might indicate churn intentions.
Low-level churn data is based on dynamic user data that constantly changes and is very individual.
Examples:
- Login frequency
- Purchase frequency
- Time spent in product
- Content viewed/bought
- Engagement and activity trends over time
- What do customers buy/stream/use?
Since these data sets can be incredibly large, this is usually the point where data analytics models are introduced to structure data and identify churn patterns.
Step 3: Build a predictive model for churn prediction and CLV #
Combine static high-level data (plan, cohort, campaign) and dynamic low-level (behavioral) data to define who is likely to churn based on user behavior (specific actions or lack of actions such as engagement with your product or communication, etc.).
Use statistical machine learning models* to find patterns that can:
- identify decreasing engagement patterns
- predict churn probability per customer
- forecast LTV per customer
*This usually requires data analysts or external support, however, there are software solutions available that already cover churn use cases. For example, the Frisbii recurring billing platform offers a churn prediction lab that uses foundational customer models to predict churn and identify churn reasons.
Step 4: Turn insights into actionable KPIs #
The churn prediction model you built in step 3 provides you with insights about what behavior causes churn. This will help you to refine your KPIs. Instead of KPIs such as “reduce churn by 5%”, you’re now able to define more operational KPIs that – according to the model – positively impact churn reduction and prevention.
For example:
- Increase of login frequency
- Increase of time spent in product
- Increase of repeat purchases (up-sells, add-ons, etc.)
Read more about how to reduce churn throughout the customer journey in our churn prevention guide
Step 5: Act on macro and micro levels #
In this step, you will have gained enough information to act on two different levels: Product-based (macro) and customer-based (micro).
If the data indicates that your product needs improvement:
- Improve onboarding if 0-3 month churn is high
- Improve feature engagement if long-term churn rises
- Fix expectation mismatch from marketing campaigns
- Refine your ICPs (ideal customer profile)
If the data indicates that you need targeted retention activities on a personalized level:
- Offer discounts to high-risk, high-LTV customers
- “Nudge” users with declining activity (e.g. via email offers, push notifications, etc.)
- Run re-engagement campaigns for inactive users
Step 6: Use your churn prediction for campaign optimization #
By identifying churn patterns within each campaign, you can
- compare behavioral patters within campaigns
- compare campaign churn vs. average churn
- forecast expected LTV per campaign
- adjust CAC (Customer Acquisition Cost) based on predicted LTV
This means that you can use churn forecasting to pro-actively optimize your campaigns, allocate proper campaign budgets and focus on campaigns that will result in customers with a higher retention rate and high LTV.
With Frisbii, you save 4 out of 6 steps. No need to hire a data agency or build up your own data analyst team. Our recurring billing platform provides you with churn forecasting models based on your own data. Interested? Let’s have a chat!
