To understand subscriber churn, it is not necessary to start with complex machine-learning models. Regular tracking of thoughtfully selected data segments can also be very effective.
Here is a simple and practical way to effectively analyse subscriber churn.
Step 1: Collect reader engagement data
For every subscriber due for renewal during the previous month, we need the following data points grouped at an event level:
- Subscriber ID.
- Article ID.
Properties of the article:
- Category tag.
- Format type.
- Length of article.
Please note, all events that occurred during the subscriber’s last plan should be considered.
Step 2: Calculate scores for each subscriber
Score for volume of content consumed:
- Include all stories read by the subscriber within the duration of his last plan.
- Allocate points for each article read based on its length, format, category of content, etc.
- Aggregate all story points to arrive at a total score for each subscriber.
Score for frequency of visit:
- Evaluate the ratio of total active days by plan duration for each subscriber.
Step 3: Create subscriber cohorts based on variables calculated in step 2
Since we only have two variables for each subscriber, we can represent each subscriber on a scatter plot.
Plot all users in a scatter plot where x = volume score and y = frequency score. Mark the vertical and horizontal boundaries wherever users are clustered. Give a unique name to each of the clusters/cohorts for easy reference.
Sometimes, you might even have to try exponential or a log function of x and y values to get a clearer segmentation.
Step 4: Calculate churn rate for each cluster/cohort
This is a simple definition of subscriber churn you could use:
- Let y = all subscribers due for renewal during an assessment period.
- Let x = amongst y, select the set of subscribers who actually renewed on or before their date of expiry.
Your equation is: Churn = 1 -(X/Y)
Write down the churn rates next to the cohort on the 2-D plot. Now you will begin to clearly note how the churn rates vary by subscriber segments.
From this analysis, you will be able to identify subscriber cohorts with a high risk of churn, understand the extent to which engagement influences churn rates, and identify the engagement thresholds to be crossed for driving meaningful reduction in churn.
You can include more variables such as lifetime value (LTV), type of subscription plan, and acquiring channel for defining your subscribers in step 1. Adding such variables may further sharpen your analysis as these factors are known to influence churn rate for a lot of subscription products.
If you plan to use three or more variables to bucket subscribers into cohorts, you would no longer be able to represent your subscribers on a 2-D plot (as demonstrated in step 3). In this case, for completing step 3, you would need to use Excel or apply a clustering algorithm.