To Proactively Give Discounts or Not?

Is there anything to be gained from foreseeing who's likely to wake up a week, month or year from today with a strong feeling of subscription fatigue or mere unsatisfaction?

Posted by Alfred Prah on March 31, 2023 · 6 mins read
In today's highly competitive business environment, it's essential for companies to retain their customers. One way to do this is to reduce churn rates - the rate at which people unsubscribe from a good or service or choose alternatives over yours.

In this article, we'll explore how Data Science and Machine Learning can be used to reduce churn rates in the Telecommunications Industry. Let's assume we're creating a new hypothetical cell phone carrier, EF&G, which has the largest market share in the industry. The million-dollar question is, how can we determine which of EF&G's customers are likely to cancel their phone plans in the near future, and what should we do with that information?

Cost of Acquiring A Customer

The cost of acquiring customers has skyrocketed in recent years. For example, in 2023, assuming EF&G wanted to purchase a 30-second Super Bowl Ad to brand recognition, it would cost the company $7 million, which is $500,000 more than the previous year and $1.5 million more than 2021. For an ad that expensive, the expectation would be to have whichever customers were acquired through the ad (assuming our marketing team voted to purchase one for the just-ended Super Bowl) to stay with the company long enough to turn a profit (lifetime value of a customer).

Data Science and Machine Learning Methodology

To reduce churn rates, EF&G can leverage Data Science and Machine Learning methodologies. The first step is to collect and analyze customer data. Then, using predictive analytics, we can develop a churn prediction model to identify which customers are likely to churn. The model can take into account various factors, such as customer demographics, usage patterns, billing history, and customer service interactions. For a phone carrier specifically, other nuanced attributes such as the quality of service provided to a given customer (using telemetry on network strength in a geographical region for example), and segmented data usage per quarter, month or week (to determine if someone is potentially feeling like their getting their money's worth over time) can be useful. Several machine learning models can be used for churn prediction, depending on the specific needs of a business and the available data. Here are some good traditional options I'd personally use to generate some benchmark model performance:
  • Logistic Regression: This is a simple yet powerful machine learning model for binary classification tasks like churn prediction. It is easy to interpret and can be trained quickly on large datasets. It makes predictions based on a weighted sum of input features, and its output can be passed through a sigmoid function to obtain a probability of churn.
  • Decision Trees: Decision trees are another popular machine learning model for churn prediction. A great value-add is its ability to recursively split the data into subsets based on the most significant feature at each step. Through this, we can infer features that are highly correlated with churn prediction or otherwise. Additionally, they are easy to interpret and can handle both numerical and categorical data. However, a downside of Decision Trees is how prone they are are to overfitting, which can lead to poor generalization on unseen data down the line.
  • Random Forest: Random forest is an ensemble learning method that combines multiple decision trees to improve the accuracy and robustness of the model. It is a popular model for churn prediction because it can handle a large number of features and is less prone to overfitting than decision trees. Random forest works by building multiple decision trees on random subsets of the data and then combining their predictions.
  • Gradient Boosting: Gradient boosting is another ensemble learning method that combines multiple weak models to create a strong model. It is a popular model for churn prediction because it can handle a large number of features and has high predictive accuracy. Gradient boosting works by iteratively adding weak models that correct the errors of the previous models. Think of it as "the model that learns from its mistakes over time".

  • Ultimately, the choice of machine learning model for churn prediction depends on the specific needs of the business and the characteristics of the data. Once we've identified customers who are likely to churn, we can take proactive measures to retain them. For example, we could offer them a discount or promotion, personalized recommendations based on their usage patterns, or improved customer service.

    Benefits of Proactively Predicting (and ultimately reducing) Churn

    Using Data Science and Machine Learning to reduce churn rates has several benefits. First, it helps companies retain their customers, which is crucial for their long-term success (imagine squandering all that Super Bowl ad money!). Second, it helps companies save money on customer acquisition costs, as retaining existing customers is generally less expensive than acquiring new ones. Third, it allows companies to offer personalized recommendations and services to their customers, improving the customer experience and building customer loyalty long-term.

    Conclusion

    The Telecommunication Industry was my use case for this article but Data Science and Machine Learning can be used to reduce churn rates in any other industry. By collecting and analyzing customer data to proactively develop a churn prediction model, companies can identify which customers are likely to look elsewhere, and consequently take measures to retain them.