Distinguishing potential child insurance customers : a statistical investigation
MetadataShow full item record
- Master Thesis 
In this thesis we try to illuminate possible reasons why the launch of a more affordable child insurance product by an established Norwegian insurance company failed to live up to the company’s expectations. We use three main approaches. First, to better understand the situation, we perform a change point analysis on the ratio of sales to offers from 2014 to 2020. We confirm the company’s problem by establishing that the one significant increase detected cannot have been caused by the new product. Secondly, to understand what drives sales in general, we create an easily interpretable logistic regression model to predict whether an offer is likely to result in a sale, taking into account both product types. Our most surprising finding here is the fact that the presence of certain data that the company gets from a third party business intelligence firm, and only has for about half the customers, by itself is associated with a significantly higher likelihood of purchasing child insurance. Regardless of the content of the data, its presence itself highly affects this likelihood. We suspect this is because only wealthier or more selective customers appear in this external database. Thirdly, we use two supervised methods to predict whether an offer involves the standard or new product, based on a range of customer characteristics. These fail. We then use two different unsupervised clustering methods, to see if it is possible to identify customer groups with clear preferences for one of the two products. This too fails. None of these statistical methods, successful in predicting and understanding sales, can identify characteristics or profiles associated with the new product. We interpret these failures as meaning that no customer segment significantly prefers the new product over the old. Slight evidence from one of the cluster analyses also suggests that a more premium product rather than a more economical one could have been more successful. Our final conclusion is therefore that the economy product was unsuccessful because it appealed to a non-existent customer segment.