Data is essential to the rapid digitization of the underwriting process but as with any system designed to optimize efficiency, AI programs can leave behind people who are “difficult to assess” in a model based paradigm. Consumer groups have noted that the use of data tools for automated decision-making in the underwriting process can inherit prejudice of prior decision makers or reflect biases that persist in society at large. In the absence of a formal guidance, what language and tools can life insurers use to have a productive discussion about bias and fairness in models to support risk assessment?

Hareem Naveed, Director, Integrated Analytics, and Amy Atkinson, Deputy General Counsel and Chief Compliance Officer, present a technical framework for bias mitigation in model based underwriting with illustrated examples from the life insurance industry. 

• Leveraging third-party data: From our extensive experience working with carriers on designing and monitoring AUW programs, we present a compilation of third-party data used as inputs to model-based risk assessment frameworks and the questions that need to be applied in order to examine for fairness. 
• Auditing model bias: All model audit approaches specify metrics that can be used to determine if models are fair or biased. In addition to a model performance metric, how can insurers incorporate a fairness metric in the model selection process? We define a flexible and model-agnostic framework that can be used to evaluate third-party scoring tools or internally developed models.
• Following regulatory guidance: The National Association of Insurance Commissioners (NAIC) recently adopted Artificial Intelligence Guiding Principles. We explore how legal counsel can be incorporated into the modeling process to ensure close adherence to regulatory guidelines.

Watch a recording of the webinar and download the executive summary for more information.