The insurance industry spent three centuries perfecting the art of saying no. Underwriters pored over applications, actuaries built mortality tables, and adjusters investigated claims with the thoroughness of detectives. Today, an algorithm can price your life insurance policy before you finish your coffee, scanning everything from your fitness tracker data to your grocery purchases.
The end of the questionnaire
Traditional underwriting relied on forms. How many cigarettes do you smoke? Does diabetes run in your family? The new model barely asks. Instead, it watches. Telematics devices in cars track harsh braking and late-night drives. Health insurers parse pharmacy records and wearable device data. Home insurers analyze satellite imagery to spot overhanging trees or aging roofs. One major carrier's AI system processes over 130 external data sources to price a single auto policy.
The shift happened gradually, then suddenly. Early adopters like Progressive started with simple telematic devices over a decade ago. But the real transformation came when insurers realized they could combine disparate data streams. Your credit score predicts your likelihood of filing a home insurance claim better than most traditional metrics. Your social media activity correlates with your health outcomes. Your typing patterns on insurance forms can indicate fraud.
Winners, losers, and the new risk pools
This granular risk assessment creates clear winners. Safe drivers with good credit see premiums drop by 30-40%. Young professionals with healthy lifestyles get life insurance at rates previously reserved for those a decade older. The precision is remarkable: insurers can now identify the safest 10% of drivers with 90% accuracy.
But precision cuts both ways. The same algorithms that reward low-risk behavior penalize those they deem risky. A construction worker who drives to job sites at 4 AM faces higher auto premiums, regardless of their safety record. Genetic testing, though currently regulated, looms as the next frontier. Insurers are already using proxy data to infer what direct genetic information might reveal.
The industry's fundamental promise—pooling risk across populations—starts to fray when risk assessment becomes too precise. If insurers can perfectly predict who will file claims, insurance becomes unaffordable for those who need it most. Several European countries have already banned certain types of algorithmic discrimination in insurance pricing.
The black box problem
Modern insurance AI systems are remarkably opaque. A deep learning model might consider thousands of variables and their interactions, making decisions no human can fully explain. When your premium doubles, the insurer may genuinely not know why—only that the model predicts higher risk.
This opacity creates new problems. Regulators struggle to audit algorithms for bias. Consumers can't dispute decisions they don't understand. Errors in training data can systematically disadvantage entire groups. One widely-used model incorrectly associated certain zip codes with high risk, effectively redlining communities based on flawed correlations.
Some insurers are developing "explainable AI" that can articulate its reasoning. But the trade-off is stark: more explainable models are often less accurate. The industry faces a choice between precision and transparency.
Our take
AI has made insurance more accurate but less insurance-like. The industry's social function—spreading risk across society—weakens when algorithms can slice risk into ever-finer segments. The most sophisticated pricing models might actually undermine the product they're meant to improve. Insurance works because we don't know who will need it. When AI knows too much, the pooling mechanism that makes insurance possible begins to break down. The industry's next challenge isn't building better algorithms—it's deciding how much we want them to know.
