The insurance industry has always been in the business of predicting the future. For centuries, actuaries built statistical models from mortality tables and claims histories, translating uncertainty into premiums with slide rules and, later, spreadsheets. That era is ending. Today, the underwriter reviewing your application is increasingly likely to be an algorithm trained on datasets so vast and granular that traditional risk categories—age, occupation, postcode—have become almost quaint in their simplicity.
The shift happened without fanfare. No press conferences, no congressional hearings, no viral demonstrations of machines outperforming humans. Insurance is not a glamorous industry, and its transformation has proceeded in the background while public attention focused on more photogenic applications of artificial intelligence. Yet the consequences for ordinary people are profound. The price you pay for coverage, whether you can obtain it at all, and how quickly your claim gets processed now depend substantially on machine learning systems that most consumers will never see or understand.
From actuarial tables to neural networks
Traditional underwriting relied on broad demographic categories because that was all the data could support. An actuary might know that male drivers under twenty-five filed more claims than middle-aged women, but the granularity stopped there. Modern AI systems ingest thousands of variables: telematics data from your car, purchasing patterns from data brokers, satellite imagery of your roof, even the cadence of how you type on a claims form. The goal is not to assess groups but individuals, building bespoke risk profiles that can justify bespoke pricing.
Health insurers, where regulations permit, have experimented with wearable data and pharmacy records. Property insurers use computer vision to assess building conditions from aerial photographs. Life insurers have tested models that predict mortality from facial images—a capability that sounds dystopian until you learn it correlates with physiological markers that traditional medical exams might miss. The accuracy gains are real. So are the questions about fairness, transparency, and the boundaries of acceptable surveillance.
The black box problem
Insurance is a regulated industry, and regulators have historically demanded that pricing be explainable. You cannot charge someone more simply because they live in a predominantly Black neighbourhood—a practice called redlining that was outlawed decades ago. But machine learning models are notoriously difficult to interpret. A neural network might arrive at discriminatory outcomes through proxies invisible to human reviewers: purchasing patterns that correlate with race, location data that maps onto income, health signals embedded in seemingly innocuous variables.
Some jurisdictions now require algorithmic audits. The European Union's AI Act imposes transparency obligations on high-risk automated decision systems, and insurance underwriting clearly qualifies. In the United States, regulation remains fragmented, with state insurance commissioners taking varied approaches. The industry itself is divided: some carriers see explainability as a competitive advantage, a way to build trust; others view it as an impediment to deploying the most powerful models.
The human underwriter's uncertain future
For professionals who spent careers mastering the art of risk assessment, the rise of AI presents an existential question. The algorithms do not replace underwriters entirely—someone must still validate edge cases, handle appeals, and exercise judgment where data is sparse. But the volume of decisions that require human attention is shrinking. Entry-level positions that once served as training grounds are disappearing, automated away. The career ladder is losing its lower rungs.
Some veteran underwriters have pivoted to oversight roles, becoming the humans who audit the machines. Others have moved into product design, where understanding customer needs still requires intuition that algorithms lack. But the profession is contracting, and the workers affected are not the young technologists who might easily retrain. They are mid-career specialists whose expertise was built for a world that is passing.
Our take
The AI transformation of insurance is neither dystopia nor utopia—it is simply the latest chapter in an industry that has always monetised uncertainty. The efficiency gains are genuine: faster claims processing, more accurate pricing, reduced fraud. But efficiency is not the same as equity. When algorithms determine insurability, the stakes are too high to accept opacity as the price of progress. The industry needs external scrutiny, and consumers deserve to know when a machine has judged them unworthy of coverage. The quiet revolution in underwriting should not remain quiet forever.
