News Release: ORCAA and Octagram Analytics Form Strategic Alliance Addressing Unfair Bias in Insurance
Cathy O’Neil and Jessica Leong
June 13, 2023 — O’Neil Risk Consulting and Algorithmic Auditing, Inc. (ORCAA) and Octagram Analytics announce a strategic alliance and a new service to help P&C insurers address emerging state regulations designed to detect unintended unfair bias.
The service is called INFER, which stands for INsurance Fairness Explainability Review. INFER helps insurers develop governance policies and practices to detect and prevent future bias. “Combining the expertise of both companies allows insurers to confidently approach model bias and compliance in the era of artificial intelligence,” said Jessica Leong, CEO of Octagram Analytics and former president of the Casualty Actuarial Society.
Leong leverages her extensive experience building predictive models within the P&C insurance industry. ORCAA’s President, Cathy O’Neil, is a leading algorithmic auditor and author of “Weapons of Math Destruction,” which raised awareness about the potential of algorithms in many industries, including insurance, to perpetuate bias against certain groups and cause other harms.
Although insurers adhere to traditional standards to prevent unfair discrimination, the underlying data and escalating model intricacies can introduce bias in insurance practices. The bias can exist where potentially discriminating data and complex models are applied to insurance practices, including pricing, rating, marketing and claims handling.
O’Neil and Leong joined forces in response to growing regulatory scrutiny of insurance practices. Some states are developing regulations to ensure insurers are not unfairly discriminating against policyholders. Specifically, the Colorado Division of Insurance is developing regulations model governance and testing required by SB 169, enacted in 2021.
“We are on the front lines of this issue,” said O’Neil, who is assisting with the proposed regulations. Further, the insurance department in Washington, D.C., recently conducted a personal auto data call for an initiative to evaluate unintentional bias. Connecticut also posted a notice concerning insurers’ use of big data and avoiding discriminatory practices.
Similar state initiatives are expected to spread throughout the country.
Recognizing that insurers may want to avoid holding or accessing granular data on protected classes, INFER offers ORCAA’s Pilot analysis platform. It includes a double firewall: Insurers cannot access inferred data on race and gender, and INFER cannot access an insurer’s personally identifiable information.
INFER’s key offerings and deliverables for governance include:
Gap analysis of existing model and data governance structures and procedures
Guidance for creating a rubric to prioritize risks that consider consumer impact.
Advice on how to address unfairly discriminatory outcomes, if found
INFER’s key offerings and deliverables for bias testing include:
The Explainable Fairness analysis framework to identify and measure bias in outcomes
Audit reports for specific models
Custom dashboards for continuous outcomes monitoring
ORCAA is an algorithmic auditing consultancy that helps clients identify and manage risks related to the use of artificial intelligence and machine learning systems – especially risks related to fairness, bias and discrimination. The company’s DNA is based on CEO Cathy O’Neil’s 2016 book, Weapons of Math Destruction. Our mission is to help establish and uphold standards for the safe and fair use of algorithms, housing and credit, as well as other unregulated industries. For more information, please visit www.orcaarisk.com.
Octagram Analytics helps property & casualty insurers bolster their competitive advantage through data-driven pricing strategies. Founded by Jessica Leong, former president of the Casualty Actuarial Society, the Octagram team has decades of experience building predictive pricing models that deliver effective underwriting results. For more information, please write Jessica Leong, President, Octagram Analytics.