Advanced rating models built in record time to significantly improved pricing for Allstate customers
Published: February 15, 2024
Our D3 team within Allstate NI embarked on the ambitious task of creating advanced rating plan models in collaboration with MMi. Focusing on enhancing the accuracy of coverage level statistical models, the team, predominantly based in Allstate NI, achieved an unprecedented feat by building a comprehensive suite of models within just four months.
A rating plan is a table of values, based on an underlying statistical model that predicts the amount a customer should pay for their insurance based on their characteristics or "risk profile". The specific objective was to build out coverage level statistical models that have improved accuracy over their historical counterparts. Although these models were the initial starting point, they continue to be iterated and improved on over time as more information becomes available.
These models provide more accurate rating which leads to more affordable insurance and simple in that they make use of data collected via external reports meaning less for the customer to fill in when obtaining a quote.
Adapting with Flexibility and Navigating Risks for Timely and Successful Project Deliver
The group involved met almost daily to ensure that any findings were shared and that any concerns were escalated as quickly as possible. Any decisions made during the project were tracked and any decisions that altered the course of the project were discussed explicitly with stakeholders. All models went through D3 model validation framework to ensure effective model risk management.
A rapid was established with key decision makers before project kick off as well as all the key deliverables. This kept things on track and made conversations around issues that arose easy because we knew exactly where to go for these decisions to be made. The iterative approach to building the models meant that if any issues arose or any changes in requirements from stakeholders, these could be incorporated into one of the upcoming iterations.
Confluence was used as the main documentation tool, this meant that the global team could asynchronously ensure that everything was communicated and tracked appropriately. Tasks for the individual modellers were managed using JIRA and the group met daily to discuss progress.
Assessing Success, Planning Improvement
Through the increased accuracy of these models, this has helped generate significantly improved pricing. The scale of the work was unprecedented, with an entire suite of models being built within 4 months, an exercise that historically had taken over a year. Although these models were the initial starting point, they continue to be iterated and improved on over time as more information becomes available.
