Optimizing Customer Retention and Marketing Strategies with Machine Learning
Objective
To improve customer retention, optimize marketing campaigns, and accurately predict Customer Lifetime Value (CLV) for more effective targeting and resource allocation.
- Enhance marketing strategies and customer engagement efforts
- Reduce churn rate by targeting high-value customers with personalized campaigns
- Optimize marketing spend based on accurate CLV predictions and customer segmentation
Challenges
- 85.4% churn rate, indicating low customer engagement and retention
- Ineffective broad marketing campaigns, leading to resource wastage
- Manual CLV analysis is costly ($20 per customer) and time-consuming, taking approximately 2 hours per customer, leading to 15.4 weeks of work for the full customer base (6,161 customers)
Methodology
Our methodology followed these steps:
- Data Analysis & Feature Selection: Analyzed features such as Number of Policies, Monthly Premium Auto, Income, and Employment Status
- Predictive Modeling: Utilized Random Forest to predict CLV, ensuring accuracy with a MAPE of 3.3% and an adjusted R² of 93%
- Customer Segmentation: Applied machine learning clustering to segment customers based on their CLV, creating Gold, Silver, and Bronze clusters for more targeted campaigns
- Explainable AI: Used to identify important features influencing customer decisions and CLV predictions, enhancing understanding of customer behaviors
Analysis
We identified the following key insights:
- Gold Cluster: Represents the highest-value customers, with an average CLV of $13,277.98
- Marketing Campaign Performance: Offer 2 through the Agent channel showed the highest retention rate (28.01%) and has the potential to improve customer retention by 40%
- Cost Optimization: Machine learning-based segmentation led to a total marketing cost saving of $47,875 by more accurately targeting customers
- Manual vs. Automated CLV Analysis: By using a Data Scientist with machine learning, the company can save $6,322 in costs and reduce the project timeline by 74.03%, cutting down 11.4 weeks of work
