A general insurance company sought to improve its assessment of break-in risks associated with auto insurance leads, aiming to enhance decision-making for rejection or acceptance using predictive acquisition.
About Client
Objectives
The client aimed to develop a predictive acquisition model to evaluate the risk of break-in leads and improve underwriting processes.
Approach
- Over 100 variables were created, including factors such as the lag between inspection and lead creation, leads generated during holidays or long weekends, and flags for previous policy break-ins.
- A rejection model was developed using random forest methodology to score leads based on their risk of fraudulent claims post-conversion.
- Leads approved at this stage were assigned a policy number.
Outcome
The predictive acquisition model demonstrated high precision and recall, enabling the client to better assess risks associated with each lead and make informed decisions regarding rejection or acceptance.
- Performance in predicting rejections:
- Precision: 79%
- Recall: 60%
- F1-Score: 68%
- Performance in predicting approvals:
- Precision: 81%
- Recall: 92%
- F1-Score: 86%
- Random Forest was trained on 75% of the dataset, and 25% was reserved for validation. Two-stage prediction models were developed to optimize performance.
