Churn Prediction and Transforming Data Using Bold Data Hub

In this article, we will demonstrate how to import tables from a CSV file, to predict churn using transformations, and move the cleaned data into the destination database using Bold Data Hub. Follow the step-by-step process below.

Sample Data Source:

Sample CSC Data

Creating Pipeline

Learn about Pipeline Creation

Applying Transformation

Learn more about transformation here

Churn Prediction Feature

Overview

Churn prediction models are used to forecast the likelihood of a customer discontinuing their relationship with a company. By creating features such as the time since the last contact, frequency of tickets, or changes in support issues, we can enhance the model’s ability to predict churn. These features provide valuable insights into customer behavior patterns and engagement.

Approach

We can derive the following features from support ticket data:

• Time Since Last Contact: The time difference between the most recent support ticket and the current date.
• Frequency of Tickets: The number of tickets raised by the customer within a given timeframe (e.g., monthly, quarterly).
• Changes in Support Issues: Tracking the nature or type of support issues over time to detect shifts that might indicate dissatisfaction.

SQL Query for Creating Churn Prediction Features

WITH Customer_Activity AS (
    SELECT 
        Customer_ID, 
        Customer_Name, 
        MAX(Ticket_Creation_Date) AS Last_Interaction, 
        COUNT(Ticket_ID) AS Total_Tickets, 
        SUM(CASE WHEN Ticket_Status = 'Resolved' THEN 1 ELSE 0 END) AS Resolved_Tickets, 
        AVG(Customer_Satisfaction) AS Avg_Satisfaction 
    FROM {pipeline_name}.sample_csc_data 
    GROUP BY Customer_ID, Customer_Name
)
SELECT 
    c.*, 
    (CURRENT_DATE - Last_Interaction) AS Days_Since_Last_Contact, 
    (Resolved_Tickets * 1.0 / NULLIF(Total_Tickets, 0)) AS Resolution_Rate 
FROM Customer_Activity c 
ORDER BY Days_Since_Last_Contact DESC;