The transformation happening inside insurance companies may be the most consequential AI deployment most people will never notice. Underwriting — the ancient art of deciding who gets coverage and at what price — is being systematically handed over to algorithms, and the implications extend far beyond quarterly earnings calls.
For centuries, underwriting relied on actuarial tables, historical claims data, and human judgment. An experienced underwriter might spend decades developing intuition about which applicants posed hidden risks. That institutional knowledge is now being compressed into neural networks that can process thousands of variables simultaneously, spotting correlations no human could perceive.
The data advantage
Modern underwriting AI doesn't just analyze your application — it ingests everything permissible about you. Credit behavior patterns, property records, social determinants of health, even the metadata around how you fill out forms. One major insurer reportedly found that applicants who complete forms between 2 a.m. and 4 a.m. file claims at higher rates, though the company declined to confirm whether this variable influences pricing.
The models are remarkably effective at their core task. Early adopters report loss ratios improving by several percentage points, which in an industry where margins are razor-thin translates to substantial competitive advantage. Smaller insurers that cannot afford sophisticated AI systems increasingly find themselves adversely selected — stuck with the risks the algorithms rejected elsewhere.
The fairness problem
Here lies the tension regulators are only beginning to grapple with. Insurance has always been, at its core, a business of discrimination — the legal kind, distinguishing between risk profiles. But when AI discovers that certain zip codes, purchasing patterns, or browsing behaviors correlate with claims, it can effectively recreate prohibited discrimination through proxies.
Several state insurance commissioners have begun requiring algorithmic audits, though the technical capacity to conduct meaningful oversight remains limited. The models are often black boxes even to the companies deploying them. When asked why a particular applicant received a higher premium, the honest answer is frequently that no one knows — the machine found a pattern in the training data.
What policyholders should understand
The AI revolution in underwriting means that your insurability is increasingly determined by factors you cannot see and may not control. The person who buys the same coverage, lives in the same neighborhood, and has the same claims history might pay dramatically different premiums based on behavioral signals neither of you are aware you're sending.
This isn't necessarily unjust — more accurate risk pricing can lower premiums for genuinely lower-risk individuals. But it represents a fundamental shift in the social contract of insurance, from broad risk pooling toward hyper-individualized pricing that some critics argue undermines the very concept of shared risk.
Our take
The insurance industry's AI adoption deserves far more public scrutiny than it receives. While we debate whether chatbots will take white-collar jobs, algorithms are already making consequential decisions about who can afford health coverage, whether a small business can get liability insurance, and how much families pay to protect their homes. The efficiency gains are real, but so is the accountability gap. Insurance commissioners need technical staff who can actually audit these systems, and consumers deserve to know when a machine is pricing their risk. The actuaries saw this coming years ago. The rest of us are only now catching up.
