How Do Insurers Use AI to Assess Risk and Set Premiums?

Analyzing Data at a Scale Traditional Methods Struggle With

Insurers use AI primarily to analyze large volumes of data more efficiently than traditional actuarial methods alone might allow, looking for patterns relevant to assessing risk across large populations of policyholders or applicants. This can include analyzing historical claims data, demographic information, and other data sources permitted under applicable insurance regulations, with the goal of identifying patterns that inform how risk is assessed and, ultimately, how premiums are priced. AI’s ability to process and find patterns across large, complex datasets makes it a natural fit for this kind of large-scale risk analysis work, which has long been central to how the insurance industry operates.

It’s worth emphasizing that this represents an enhancement of long-standing actuarial practices rather than an entirely new approach — insurers have always relied on data analysis to assess risk; AI has changed the scale and sophistication of that analysis rather than the underlying goal.

Regulatory Limits on What Data Can Be Used

An important constraint on how insurers use AI for risk assessment and pricing is that insurance regulations, which vary by jurisdiction and by the specific type of health plan involved, generally restrict what factors can legally be used in setting premiums. These restrictions exist independently of whether AI or more traditional methods are used to analyze the data, meaning AI doesn’t create a loophole around existing legal restrictions on permissible pricing factors, at least in principle. However, the growing sophistication of AI-driven analysis has raised questions among regulators and researchers about whether AI models might indirectly incorporate the effects of restricted factors through other correlated data points, even without directly using a restricted factor itself.

The Bias Concern Behind Growing Scrutiny

A significant concern that has drawn regulatory and public attention is the possibility that AI risk assessment models, trained on historical data, could learn and perpetuate patterns that effectively disadvantage certain groups, even without explicitly using legally protected characteristics as an input. This concern reflects a broader, well-recognized challenge with AI systems trained on historical data generally: if past patterns in the data reflect existing disparities, an AI model can sometimes learn to reproduce those patterns going forward. This has led to increased regulatory interest in understanding and auditing how AI models used in insurance risk assessment actually function, and what data patterns they may be picking up on.

Bottom Line

Insurers use AI to analyze large volumes of data and identify patterns relevant to assessing risk and setting premiums, generally building on established actuarial practices rather than replacing them, though this use remains subject to insurance regulations governing permissible pricing factors and has drawn growing scrutiny over concerns about AI potentially reinforcing unfair bias.

Go deeper

• Can AI underwriting models use data sources beyond traditional risk factors?
• Can AI help insurers price climate related risk more accurately?
• How do state insurance regulators oversee AI based pricing models?