INFER: Gain confidence in your approach to model bias and compliance in the AI era

INFER is a consulting service for P&C insurers that want to be on the front foot of the evolving regulation around bias. INFER: Insurance Fairness Explainability Review will guide insurers  through establishing good governance and bias testing in models.

Insurers know it is important for models to be “fair,” but are uncertain how to stay compliant as the definition is evolving. While regulations have long managed the impact of algorithms in the insurance industry, the increased sophistication and wider application of models have brought governance to a new stage. Some insurance regulators and other public actors are requiring increased scrutiny and reporting on  how private entities are using advanced analytical tools.. The year 2022 brought two examples in insurance: Colorado’s SB 169 requiring model governance and testing and Connecticut’s Notice concerning the usage of big data and avoidance of discriminatory practices.

ORCAA and Octagram have formed a strategic alliance to help insurance companies prepare for this new era of regulation. We can help with both (1) governance structures and processes around the use of big data models, and (2) testing of the models for bias.

You can trust ORCAA and Octagram to guide you

Cathy O’Neil founded ORCAA, a leading algorithmic auditor with clients in insurance and other regulated industries. ORCAA are on the front lines of this issue, including assisting the Colorado Division of Insurance with its implementation of SB 169. Octagram is a data and analytics consulting firm with deep expertise in insurance, founded by Jessica Leong, former president of the Casualty Actuarial Society.  With our combined expertise, we can help insurers gain confidence in managing this complex issue.

How can you test models without having access to data on protected classes?

To test models for bias, you can infer race and ethnicity using the BISFG method (Bayesian Improved Firstname Surname Geocoding), a standard methodology developed by RAND. However, we know that insurers are sensitive about this data, even when inferred. That’s why we developed the Pilot software platform for analysis. It provides a double-firewall: the insurer will not have access to the inferred data on gender or race/ethnicity, and ORCAA and Octagram will not have access to the insurer’s personally identifiable information.  This overcomes a major hurdle for many carriers to start testing.