Sandbox 2.0 is an Insurance Company based in India. The insurance products offered by this company are health insurance and vehicle insurance. In the fourth quarter of 2020, there was a decrease in the Monthly Recurring Revenue (MRR) in vehicle insurance products. Business Team wants to do cross selling on Health Insurance customer who has a vehicle by selling Vehicle Insurance.
Increase number of customer, revenue growth, cost efficiency.
Predictive Analysis: build a model that could predict health insurance customer response towards cross-selling offers.
Historical data of health insurance customer: demography, health insurance premium, vehicle owned, response towards cross-selling.
We did experiments on features, such as:
- Feature encoding: one hot encoding, label encoding, ordinal encoding, and target encoding.
- Outliers treatment: log transformation, IQR method.
- Class imbalanced: oversampling SMOTE 50:50
- Train-test split: 7:3
- Scaling: standardization
This is a supervised learning problem, to be precise, a binary classification problem. These are some algorithm we have tried:
- Random Forest
- Decision Tree
- Logistic Regression
After some experiments, we evaluated the model using ROC-AUC score due to its imbalanced class, we decided XGBoost as the chosen model for further step due to its performance on test set (not overfitting).
Result: Insights and Recommendation
We found some insights in the features, but this 2 are the most interesting.
West Bengal has the highest conversion rate, business team could focus more on this area to increase customer growth.
Exclusive/captive agent has the highest conversion rate, business team could optimize the use of this channel.
Sales and Marketing:
- Focusing on West Bengal Region to increase customer growth.
- Focusing on Captive Agents to approach our customer segment.
- Utilizing the facts of accident rate, age victim and economic impact to emphasize the importance of insurance on ads.
- Giving promotions/coupon for the first 1.000 customers as part of acquiring process, such as free towing, fuel voucher, etc.
- Providing special training for captive agents in persuading customers to recommend our service to their relatives (advocation).
- Once customers are acquired, maintain their loyalty by delivering best-in-class customer experiences.
- During conversation between captive agent and potential customers, try to dig feedback for future improvement.