Project that will help underwriters score/rank the submissions and hence saving time and look at more submissons in the stipulated time.

The Client Group is a leading wholesale provider of reinsurance, insurance and other insurance-based forms of risk transfer. Dealing directly and working through brokers, its global client base consists of insurance companies, mid-to-large-sized corporations and public sector clients. From standard products to tailor-made coverage across all lines of business, Client deploys its capital strength, expertise and innovation power to enable the risk-taking upon which growth and progress depend on. This is a largely underutilized asset that can benefit from the application of analytical techniques. The initial working process of Client’s reinsurance follows four steps: Broker sends submission; Genpact indexes submission; Underwriter sees submission; Decision- work on it or not. Currently, with the annual submission significantly increasing from 3242 in 2012 to 10305 in 2015, submission to bound ratio dropped from 6.1% in 2012 to 5.5% in 2015. E&S Property LOB is a high flow business that requires quick responses. Significant growth in submissions over past four years brought more workload to each underwriter which means they had more submissions to look at than the time permitted. Since each underwriter has a process to look at deals, it was hard to cut time on each submission or increase work efficiency. Lack of labor resource made the hit ratio slump. Considering that hiring more employees will raise the cost, Client hopes to set up a mechanism to “score” each deal on submission so that the efficiency of the underwriting process and hit ratio will be increased without adding capital and resource. The goal of this study is to develop an analytical model to help underwriters triage deals that are more likely to be bound. The underwriters can utilize the model to not only identify the potential deals on the numerous submissions but also provide the best opportunity to drive more wins for the same amount of effort. This report will discuss how the historical submission data are rearranged and rectified to variables selected to construct models to fulfill the goal. K-fold Cross Validation Technique is used to split the dataset into two: training dataset and testing dataset. Then four models, Logistic Regression, Decision Tree, and Random Forest are conducted and we can find out which is the winning model that fits the data best through confusion matrix. Finally, key variables that significantly impact the model will be applied to interpret and explore business values and recommendations on how to enhance the model robustness and deploy the model to production will be given in detail.