Can an algorithm decide my insurance rate, and is that even legal?

The use of algorithms in insurance pricing is not a futuristic concept; it's a present-day reality that is reshaping how carriers assess risk and determine premiums. For consumers, this shift has introduced a layer of opacity, leading to a critical question: can an algorithm decide my insurance rate, and is that practice legal? The answer is complex, hinging on the intricate legal and ethical lines between permissible, risk-based pricing and prohibited, unfair discrimination. As insurers increasingly rely on automated models, understanding this distinction is a board-level strategic priority for medical directors, compliance officers, and reinsurance partners.

"Some of the world's largest insurance companies are charging minority communities up to 33 percent more for car insurance than they charge for similar drivers in non-minority communities with the same risk. We found that insurers were flagging people as higher risks based on factors that have nothing to do with driving, such as their TV viewing habits or whether they use generic or store-brand products." - The Markup, 2022

The legality of algorithmic insurance pricing and the question of fairness

The fundamental principle of insurance is pricing based on risk. For centuries, this has been a manual process, relying on broad demographic and actuarial data. The advent of artificial intelligence and machine learning has enabled a far more granular and individualized approach. However, this technological advancement has also introduced significant regulatory and ethical challenges, particularly around the concept of algorithm insurance pricing fairness legal frameworks.

State and federal laws have long prohibited unfair discrimination in insurance based on protected classes such as race, color, religion, or national origin. The challenge with algorithmic models is that they can inadvertently perpetuate or even amplify historical biases through "proxy discrimination." This occurs when a model uses seemingly neutral data points as substitutes for protected characteristics. For example, an algorithm might not use race as an input, but it could use ZIP codes, credit-based insurance scores, or even shopping habits, which can be highly correlated with race and socioeconomic status.

Regulators are acutely aware of this issue. The National Association of Insurance Commissioners (NAIC), a key standard-setting body for the U.S. insurance industry, has been proactive in addressing the use of AI. In December 2023, the NAIC adopted the Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. This bulletin requires insurers to develop a comprehensive AI governance program that includes, among other things, regular testing for fairness and bias. The focus is shifting from simply avoiding the use of prohibited data inputs to actively demonstrating that the outcomes of the models are not discriminatory.

Defining the line: permissible vs. prohibited discrimination

The core of the legal debate lies in distinguishing between "actuarial fairness," which justifies different prices for different levels of risk, and "unfair discrimination," which penalizes individuals based on their membership in a protected class.

Feature	Permissible Risk-Based Pricing	Unfair Discrimination (Prohibited)
Basis	Quantifiable risk factors with a causal link to the insured peril (e.g., driving history for auto insurance).	Characteristics of a protected class (e.g., race, religion, national origin) or proxies for them.
Data Inputs	Actuarially sound and directly related to risk. For example, in life insurance, this could be biometric data.	May include data that is a proxy for protected characteristics, such as ZIP code, credit score, or shopping habits.
Outcome	Premiums accurately reflect the expected cost of the risk being insured for an individual or group.	Systematically disadvantages individuals from protected classes without an actuarial justification.
Transparency	The insurer can explain the factors that led to a specific pricing decision in a clear and understandable way.	The model is a "black box," making it difficult or impossible to explain the rationale behind a decision.

Industry Applications

The challenge of ensuring fairness in algorithmic decision-making extends across the insurance value chain.

Underwriting

In underwriting, algorithms are used to assess the risk of an applicant and determine eligibility for coverage. A biased algorithm could unfairly decline applicants from certain demographic groups, even if their individual risk profile is acceptable.

Ratemaking

This is the process of setting premium rates. Algorithmic ratemaking models can lead to significant price disparities if they rely on data that is a proxy for protected characteristics.

Claims Processing

Algorithms can also be used to flag potentially fraudulent claims. However, if not properly designed and tested, these systems could disproportionately target certain groups for additional scrutiny.

Current research and evidence

A growing body of academic and industry research is exploring the issue of algorithmic bias in insurance. Researchers like Professor Andreas Tsanakas at Bayes Business School have highlighted the potential for "proxy discrimination" in insurance pricing. A 2020 study from the Casualty Actuarial Society (CAS) emphasized the need for new fairness criteria and testing methodologies for insurance pricing models. This research highlights the importance of a multi-faceted approach to fairness that considers The data used. The impact of the model's decisions on different population groups.

The key findings from this body of research include:

• The trade-off between predictive accuracy and fairness is a central challenge.
• Transparency and explainability are critical for building trust with both consumers and regulators.
• Ongoing monitoring and testing of models are essential to ensure they remain fair and accurate over time.

The future of algorithmic governance in insurance

The regulatory landscape for AI in insurance is expected to continue to evolve rapidly. We are likely to see a move toward more prescriptive regulations that require insurers to Have a governance framework in place. To be able to provide detailed evidence of their model's fairness and accuracy to regulators. This will require a significant investment in new technologies and processes for model risk management and compliance.

Frequently asked questions

What is "proxy discrimination"?

Proxy discrimination is an unintentional, and often illegal, form of discrimination that occurs when an algorithm uses a piece of information that is a substitute for a protected characteristic like race or gender. For example, an algorithm might use a person's ZIP code to set an insurance rate, but if that ZIP code is highly correlated with a particular race, it could lead to discriminatory pricing.

How are regulators addressing this issue?

Regulators, led by the NAIC, are moving from a rules-based approach (i.e., prohibiting the use of certain data) to an outcomes-based approach. This means that insurers must be able to demonstrate that their algorithms are not producing discriminatory outcomes, regardless of the data they use.

What can I do if I think my insurance rate was set unfairly by an algorithm?

Consumers have the right to question their insurance rates and to understand the basis for them. If you believe your rate is unfair, you can start by asking your insurer for a detailed explanation. If you are not satisfied with the response, you can file a complaint with your state's department of insurance.

The shift to algorithmic decision-making presents both opportunities and challenges for the insurance industry. For carriers, the ability to more accurately price risk can lead to improved profitability and new product innovation. However, this must be balanced with the legal and ethical imperative to ensure fairness and to avoid discrimination. As a leader in regulatory technology, Circadify is at the forefront of this issue, providing insurers with the tools and expertise they need to navigate this complex landscape. To learn more about how to build a compliance-first approach to digital underwriting, explore our compliance guides and regulatory insights at circadify.com/industries/payers-insurance.