Next Recharge Churn Prediction: A Machine Learning Approach

Author: Shivam Pal

Welcome to my blog! Today, we’ll explore the fascinating world of Churn Prediction Models, focusing on their importance, implementation, and potential to boost business growth. Let’s dive in!

Understanding the Problem Statement

In today’s digital economy, customer retention is critical for business growth, especially in subscription and recharge-based models. Predicting whether a user will continue to recharge their service or churn (stop recharging) can significantly impact a company’s revenue and retention strategies. This blog delves into the challenge of predicting user churn within India’s FASTag recharge system, a digital toll collection system.

Impact and Benefits of the Solution

A successful churn prediction model enables proactive customer retention. By identifying users at risk of not recharging their FASTag, businesses can tailor their marketing and retention strategies to address these users specifically. The benefits include:

• Improved Customer Retention: Targeted strategies can significantly reduce churn rates.
• Increased Revenue: Preventing churn directly translates to sustained or increased revenue.
• Optimized Marketing Spend: Focusing resources on users most likely to churn makes marketing efforts more cost-effective.
• Enhanced Customer Experience: Personalized interventions can enhance the user experience, fostering loyalty and long-term engagement.

Insights for Solving the Problem

Customers have many options for recharging their FASTag. Analyzing the recharge journeys of users reveals critical insights. From our analysis of 1 million users, we discovered:

• It takes at least three recharges on the app to ensure that 80% of subsequent recharges are also made on the app.
• Factors such as brand recall value, user engagement, recharge experience, frequency, and rewards significantly influence the user’s choice of recharge platform.

Solution Building Approach

To build the Churn Prediction Model, we employed a Machine Learning approach, following these steps:

Data Collection and Preparation

We collected data on various parameters, including:

• Recharge Behavior: Details of current and previous recharges.
• App Interactions: Number of times the user opened the app, clicked notifications, etc.
• User Demographics: Information like city tier and acquisition source.
• Rewards and Complaints: Data on rewards earned and complaints filed by the user.

We used a balanced dataset of around 300,000 users to train our model.

Feature Engineering

Key features used in the model included:

• Current Recharge Source: Whether the current recharge was done through our app or not (target variable).
• Last Recharge Source: Was the previous recharge done on our or another app?
• Credit and Debit Counts: Number of times the user added or debited amounts.
• App and Notification Clicks: Interaction metrics with the app and notifications.
• Time-to-Recharge (TTR): Time taken between recharge initiation and reflection in the account.
• Rewards and Convenience Fees: Rewards earned and convenience fees paid.
• Points Earned through Engagement: Virtual points are earned by platform interaction.

Model Selection and Training

We selected a Random Forest classifier, a tree-based method, to predict recharge churn. The process included:

• Normalization: Normalizing user data values.
• Model Training: Training model for the nth recharge.

Model Evaluation

We evaluated the model’s performance using the F1 score and accuracy. The model achieved an accuracy of around 90%, which increased to 95% after the fourth recharge.

Building Retention Strategies

Once it is predicted that the user will churn, we have designed various retention and engagement campaigns that will prevent the user from churning.

Applications of the Churn Prediction Model

The methodology for predicting recharge churn can extend beyond the FASTag system to other domains, such as:

• Subscription Services: Predicting user churn for streaming platforms, SaaS products, and online memberships.
• Telecommunications: Anticipating customer churn in mobile and internet service subscriptions.
• E-commerce: Identifying customers at risk of not returning for future purchases.
• Banking and Fintech: Predicting churn in digital banking services and financial products.

By leveraging machine learning for churn prediction, businesses can proactively address customer retention, enhance user experience, and drive sustainable growth.

Thank you for reading! Stay tuned for the next article, where we’ll go deeper into the applications of machine learning for business growth.