Observing Customers before Churning
-
Sometimes, it's helpful to observe a customer after they've churned
- In this case, it's easier to acquire customers who've already purchased or engaged
- But, it's easier to retain customers who are already purchasing or engaging
- However, it's most helpful to observe a customer before they've churned
- To do this, we can use behavioral features to measure their propensity to churn
-
This may include any of the following features:
- Logins
- Clicks
- Email opens
- Likes
-
Specifically, we can look at behavioral features of customers who've already churned
- And, we can determine common trends amongst particular segments of customers who've churned
-
This involves observing behavioral features in a window of time before customers churn
- This window of time should be shorter for customers who purchase or engage frequently
- However, this window of time should be longer for customers who purchase or engage infrequently
Observing Renewals with Churns
- While gathering our customers for clustering, we'll want to include a mix of churned and retained customers
- By doing this, we can compare churned customers with retained customers to determine any difference
- Typically, we'll pick enough retentions so that the number of retained observations are in proportion to the true retention rate
Identifying Active Periods for Non-Subscriptions
- Calculating active periods that accurately reflect periods when a customer interacted with a product is an important part of determining churn for nonsubscription services
- In this case, churn refers to a customer becoming inactive for more than some maximum allowed time
- Note, this time period is adjustable, but likely is around a month or a few months
-
This window of time should be decided for segments of customers, such that once they go inactive they don’t come back
- Or, it would be fair to consider it a fresh start if they do come back
Choosing Observation Dates for Churn Indicators
- Choosing observation windows is important when investigating leading indicators of churn
-
Churn leading indicators are behaviors that customers likely engage in before they've churned
- These behaviors are usually the underlying cause of churn
- Usually, lead times before churn may be a few weeks in advance of churn for a consumer product and one to three months for a business product
-
Churn lagging indicators are behaviors that customers likely engage in after they've churned
- These behaviors are usually the underlying cause of churn
-
When determining a churn indicator, both churns and non-churns should be included in the sample
- Roughly, reflecting the true proportion of actual churn and renewal rate
-
When creating a churn indicator, we typically first identify active periods for each customer
- Which, is when a customer has at least one event within a short time
- Events are aggregated into weeks as a single indicator of whether a customer had any events in that particular week
- Then, active periods are found from those active weeks