Set Up Churn Prediction Model

Built on Oracle Data Mining, the churn prediction model uses current and historical indicators to predict the churn probability. The churn probability is the probability expressed in percentage for a customer to discontinue the service or subscription. The churn prediction model helps you:

Add or remove custom and predefined attributes used for churn prediction.
Gather available customer behavior and usage pattern.
Build a robust predictive model by periodically training the model and calculating the prediction-based input data.
Review the accuracy of predictions using the View Model Accuracy task.
Use a predefined model to predict the churn probability on a subscription product.
Calibrate the top drivers contributing to the churn.
Identify how the churn probability has changed since the last time it was calculated.

Based on these factors, you can predict customers who are likely to churn and create opportunities to cross-sell and upsell.

The churn prediction model provides a flexible mechanism to identify the possible causes for churn probability so that the sales representatives can take corrective action. Here are the steps to set up the churn prediction model:

Set up the churn prediction model
Run the Populate Churn Prediction Features scheduled process
Run the Train with Historical data and Predict Churn Probability scheduled process
Review the accuracy of your churn prediction model

Set Up Churn Prediction Model

You can use the Manage Churn Prediction Model page to control the accepted threshold level and attributes that contribute to the prediction of churn probability.

Sign in as a setup user.
Go to the Subscription Management work area.
On the Subscriptions landing page, click the Subscription Configuration tab.
Click Manage Churn Prediction Model.
Enter a value in the Accuracy Threshold field. By default, the threshold is set to 70%.
Enter a value in the Account Level Threshold field. By default, the threshold is set to 70%.
Enter the number of days in the Training Data Duration field. All active records with a future renewal date, expired records, and closed records are included in the data set if the close date or the end date is within the time frame. By default, the number of days is set to 1,000 days.
Click the Add Row icon to add new attributes.
Select the Predictive Model check box for each attribute that you want to add for calculating the churn probability.
Click the Delete icon to remove the attributes that are used for churn prediction.
Click Save and Close.

Run Populate Churn Prediction Features Scheduled Process

The Populate Churn Prediction Features scheduled process aggregates the current values of subscription churn prediction objects such as subscription terms, past renewals, past amendments, and other supporting information such as invoices and service requests. You must run this scheduled process to transfer this aggregated data from Subscription Management to Oracle Data Mining.

Click Navigate > Tools > Scheduled Process.
On the Schedule Process window, click Schedule New Process.
On the Schedule New Process window, make sure the Job option is selected.
Enter Populate Churn Prediction Features in the Name field and click OK.
On the Process Details window, click Submit.
In the Confirmation window, click OK.

Run Train with Historical Data and Predict Churn Probability Scheduled Process

Use the Train with Historical Data and Predict Churn Probability scheduled process to train the churn prediction model with historical data and predict the churn probability of a subscription product. The objective of this scheduled process is to:

Rebuild the selected predictive models.
Make predictions based on scores derived during the build process.

You must have at least 30 churned and 30 renewed subscriptions to use this feature. When there are fewer than 30 records of churned and renewed subscriptions, the Train with Historical data and Predict Churn Probability scheduled process still runs successfully, but with the message to add more subscriptions as the churned data is insufficient to train the churn prediction model.

A churned subscription means one that's closed or terminated before the end date or a subscription that expires without a renewal. When you renew a subscription, the original subscription record is set in the expired state. The application considers these renewed subscriptions and their corresponding parent subscriptions as not churned. When a subscription expires without renewal, the application considers it as churned.

You must run this scheduled process every time you add or removes attributes used for churn prediction on the Manage Churn Prediction Model page.

Click Navigate > Tools > Scheduled Process.
On the Schedule Process window, click Schedule New Process.
On the Schedule New Process window, make sure the Job option is selected.
Enter Train with Historical Data and Predict Churn Probability in the Name field and click OK.
In the Basic Options section, select a value in the Mode drop-down list. You can select:
1. Train Data and Run Prediction: Creates a new model based on selected attributes to train data and uses it to predict churn probability.
2. Run Prediction: Uses the model created in the previous run of this job to predict churn.
3. Train Data: Creates a new model to train data based on selected attributes.
On the Process Details window, click Submit.
In the Confirmation window, click OK.

You must repeat these steps each time you want to predict churn for new subscriptions.