One of the UK’s largest insurance companies was preparing to launch a website with a suite of products for commercial vehicle fleets, providing live quotes to customers on the spot. Before they could launch, they needed to develop robust pricing models as quickly as possible. The company engaged StatCore based on their decade of expertise in this area.
StatCore’s challenge was to
- Test a range of variable answers customers could give to each question. There were more than 150 variables and 15 of them proved to be predictive of the risk a customer presented.
- Develop a new pricing model based on these 15 variables. This new pricing model boasted three times the predictive power of their previous model, which identified only 5 variables that were predictive of risk.
- Demonstrate the predictive powers of the new model to non-technical audiences, using visualisations of model performance (Gains Charts, Lift Curves)
- Build claim predictive models for perils (fire, theft) to determine the potential cost of a claim, using the Emblem software
- Train in-house staff to use Emblem to test whether a variable is predictive
- Devise a no-claims methodology to determine how much to reward good claims experience; this was done through machine learning to enhance the accuracy of the live price offered.
- Automate calculations in the program Emblem in order to meet the challenging, fixed date for the website launch
Following a successful presentation by StatCore to the Pricing Committee, the new site was launched on time and to the client’s satisfaction.