You have probably heard about “churn” and “propensity” in conferences and meetings, but what exactly is a “propensity model?”
To put it simply, propensity models are like data equations that can quantify the chances of a phenomena happening by using data generated in the past.
In the case of churn, a propensity model ingests past behavioural data, transactional records, and other profiling data about each user, applies logic and algorithms, and then calculates a score against each user signifying the chances of churning by the end of their membership cycle.
Why do we need this model? Can’t we look at the churn trends of the past and make predictions for the future? Why do you have to calculate these churn risk scores for your entire user base?
Fair questions! The short answer is that knowing the propensity scores for each user enables your teams to take targeted actions to reduce churn.
Churn rate is a lag metric
If reporting business metrics is your only goal, then it’s absolutely fine to refer back to the time series of monthly churn data and make predictions for the year to come.
However, if your goal is to not just to know but to take effective actions to prevent readers from churning, as a starting point you should try to understand the following using data analytics:
- Who are the readers who have the highest risk of churning?
- What are the common traits among this group of high-risk readers in terms of engagement behaviour, work profile, demographics, satisfaction surveys, and activation of key sticky habits?
- In terms of behaviour on the platform, how do these users differ from the low-risk groups?
- At what stage in their membership are the high-risk readers transitioned from being low-risk to high-risk?
- What actions (or lack of) result in the user becoming a high-risk member?
This mindset is a good starting point when thinking about reducing churn. The answers to the above questions will unfold the targeted actions you should take for different cohorts of readers.
Here are some key variables that could have strong correlation with churn. For engagement, these are:
- Active days per month for the first and last month of membership.
- Volume consumption score per session.
- Strong engagement with at least two sticky or core topics.
- High degree of participation in virtual events.
- Typical time of the day for consuming content.
- Average session duration.
- Engagement levels during the free trial.
For the profile, these may be:
- Plan duration and net price paid.
- Reasons for signing up for the membership.
- Years of work experience.
- Job level.
To start with, all our friends working in different media companies may not have such comprehensive data available. Therefore, for those just getting started, I recommend focusing only on basic engagement data and running simple analyses that help you understand what’s driving renewals.
Analysing churned users who renewed
Assuming you are conducting this analysis in MS Excel (or in R/Python using data frames), work through the following steps.
Step 1: Pick a cohort of subscribers who expired during the past “x” months. Setting x = 3 is a reasonable choice.
Step 2: Insert the following data columns against each churned subscriber:
- A: New column containing original expiry date.
- B: Create another column with values = <expiry date> of subscriber + 30 (days).
- C: Whether the subscriber renewed by the date mentioned above. Put 0 for no and 1 for yes.
Step 3: For each subscriber, also insert other information points such as the following (if possible; this is purely optional at this stage):
- Number of active days during month 1. (This is a mandatory value.)
- Active days in month 2, month 3, and so on. Each is represented by a unique column. (This is a mandatory value.)
- Average number of stories read per active day for each of your top formats (e.g., long-form, opinions, videos, etc.). Calculate the total stories read and divide by only the number of days when the user visited your platform. (This is a mandatory value.)
- Number of unique newsletters subscribed.
- Newsletter open rates for the reader.
- Click percentage on push notifications sent on the mobile app.
- Any other data points that can be used to differentiate content consumption quality, such as scroll depth.
Step 4: Conduct a variable analysis. Group all churned users by the number of active days during their last month of their membership and calculate renewal rate for each active day value. Plot this on a graph.
For example, you with end up with something like this:
Notice in this sample plot that, based on the slope of the line representing the cohort renewal rate (in blue), you can easily identify the threshold of active days you could target to boost the chances of subscribers renewing.
For example, as per this plot, increasing active days from six to 12 per month almost doubles the chances of renewal. Similarly, increasing active days from 14 to 27+ increases renewal rate by five times.
Similarly, doing one-dimensional analysis with other variables captured in the second step (such as “volume of content consumed” by format, “active days” in the first month of membership, or profiling data of subscribers) can provide interesting insights for you to act upon.