This article details the options that are available to define churn within the customer churn prediction template.
Defining churn in the AI & Analytics Engine
The Customer Churn Prediction template automates the entire process of creating an end-to-end solution to predict which active customers are likely to churn in the near future. An essential part of this process is defining which customer behaviors should be classified as churned.
After importing necessary datasets, users can define churn to suit the objectives of their business, by providing two parameter values:
-
The time it takes to put in place churn preventive actions.
-
The period of time for predicting customer activities.
.png?width=688&height=516&name=Defining%20churn%20(1).png)
Note: An example illustrating the significance of these two numbers can be found in the Appendix below.
By default, the “total number of transactions” is used to determine whether a customer will churn or not. If the number of transactions is fewer than a certain threshold (e.g., one transaction), they are considered churned.
Another way of defining churn is through the “total transaction amount”. The image below illustrates that a user will be considered churned if they have less than 10 (currency/unit) within 14 consecutive days:
%20(1).png?width=688&height=516&name=Defining%20churn%20(part%202)%20(1).png)
After clarifying the churn definition, the next step is to define the contributing factors. For more information, read this article.
Appendix: The churn definition
Customer Churn Prediction helps businesses increase their retention rate by providing early warnings of customers with a high risk of churn. With this knowledge, businesses can take preventive actions, such as offering benefits, which often reduces the likelihood of churn. However, if the time of the warning is too close (for example, tomorrow), it will not be sufficient to take any practical action. Therefore, an appropriate interval should be added after the prediction time in the churn definition.
For example, on 01/01/2023, one wants to predict which of their customers will churn in the near future, and they need 7 days to respond. The situation can be described with the illustration image below:
In the period after 08/07/2023, assuming that the following rule is set: if a customer has no activity for 14 consecutive days, they will be considered as churned. Based on this rule, the machine learning models will predict the customer activities in the 14-day period (the red period in the illustration below), and determine the churn status.
The trained model estimates the likelihood (for every active customer) that there will be no transactions in the 14 days period following the 7 days period.