Artificial intelligence in insurance
Artificial intelligence in insurance is the use of machine learning and other artificial intelligence (AI) techniques by insurance companies and intermediaries, for tasks such as underwriting, pricing, claims handling, fraud detection and customer service. Insurers have used statistical models to classify and price risks for decades, and actuaries adopted generalized linear models in the
Overview
Artificial intelligence in insurance is the use of machine learning and other artificial intelligence (AI) techniques by insurance companies and intermediaries, for tasks such as underwriting, pricing, claims handling, fraud detection and customer service. Insurers have used statistical models to classify and price risks for decades, and actuaries adopted generalized linear models in the 1980s. Machine learning methods spread in the industry from the mid-2010s. Surveys by supervisors in the United States, the United Kingdom and the European Union found in the 2020s that most insurers were using or testing AI, with claims handling, fraud detection and marketing as the most common uses. The use of AI by insurers has drawn the attention of regulators. The European Union's Artificial Intelligence Act classifies AI systems used for risk assessment and pricing of individuals in life and health insurance as high-risk. In the United States, where insurance is regulated by the states, the National Association of Insurance Commissioners (NAIC) adopted a model bulletin on insurers' use of AI in December 2023, and several states have passed laws on the use of AI in health insurance decisions. Criticism has focused on the risk that models reproduce discrimination through proxy variables, on the effect of individual pricing on risk pooling, and on lawsuits alleging that American health insurers used algorithms to deny claims.
History
Insurers have always analysed data to classify and price risks. EIOPA, the European insurance supervisor, has described data processing as having "historically been at the very core" of their business. In a 2021 review of the actuarial literature, Christopher Blier-Wong and co-authors wrote that statistical learning models had been common practice in actuarial science since the 1980s, when the field adopted linear and generalized linear models for ratemaking and reserving. A 2003 study had recommended neural networks for ratemaking, but according to the review it took fifteen years for the suggestion to be taken up, and the use of machine learning in pricing and reserving grew from about 2015, with gradient boosting for structured data and neural networks for unstructured data. When the NAIC surveyed American auto insurers on AI in 2022, it left linear regression, generalized linear models and generalized additive models out of its definition, as established ratemaking methods. In the United States, insurers of cars and homes also began to use external data such as credit-based insurance scores. In a 2007 report to Congress, the Federal Trade Commission found that these scores were effective predictors of risk under car insurance policies. In a thematic review published in 2019, EIOPA surveyed 222 insurance undertakings and intermediaries from 28 jurisdictions on their use of big data in motor and health insurance. It found that 31 per cent of firms were actively using tools such as artificial intelligence or machine learning, and that another 24 per cent were at the proof-of-concept stage. In a 2022 review of the research, Martin Eling, Davide Nuessle and Julian Staubli described a shift in the business model of insurance from compensating losses towards predicting and preventing them.
Applications
Surveys by insurance supervisors show AI in use across most of the insurance business, though unevenly. In a survey of 193 private passenger auto insurers reported in December 2022, the NAIC found that 88.6 per cent used, planned to use or planned to explore AI or machine learning. Claims was the most common area, with models in use at 70 per cent of companies, followed by marketing and fraud detection. The shares were lower for rating, at 27 per cent, and for underwriting, at 18 per cent. In later NAIC surveys, 70 per cent of homeowners insurers and 58 per cent of life insurers said they used AI or were interested in using it. Of 93 health insurers surveyed in 2025, 92 per cent used, planned to use or were exploring AI, most often in utilization management and claims adjudication. In the United Kingdom, a survey by the Bank of England and the Financial Conduct Authority published in November 2024 found that insurance had the highest share of firms using AI of any sector, at 95 per cent, just ahead of international banks.
Underwriting and risk assessment
In life insurance, accelerated underwriting lets insurers issue policies without a medical examination. The insurer adds external data to the application, such as credit reports, motor vehicle records, prescription drug history and records of the Medical Information Bureau, and uses predictive models and machine learning to decide eligibility, often within hours. The NAIC has said that such models must rest on sound actuarial principles and be evaluated for unfair discrimination. According to the NAIC, life insurers use AI mainly for marketing and underwriting. In property insurance, companies buy aerial and satellite photographs that computer models analyse for signs such as damaged roofs or overhanging trees. In Texas, where insurers nearly doubled their non-renewal rates between 2020 and 2023, homeowners have reported losing cover on the basis of such images, and one provider of the analysis acknowledged that aerial photos are less accurate than inspections in person. In 2025 the California Insurance Commissioner backed a bill requiring 30 days' notice before aerial inspections and giving homeowners access to the images, after complaints that inaccurate drone or satellite photos had led to cancellations.
Pricing
EIOPA's 2019 review found that pricing and underwriting were the activities where big data analytics had the largest impact. A contested use is price optimisation, in which premiums are adjusted using data not related to the expected cost of the risk. In 2015 the NAIC's Casualty Actuarial and Statistical Task Force published a white paper on the practice, and by November of that year regulators in eleven states and the District of Columbia had issued bulletins restricting it in personal lines. In the United Kingdom, the Financial Conduct Authority ended price walking in home and motor insurance. Since 1 January 2022, the price quoted to a renewing customer may be no higher than the price the same customer would be quoted as a new customer through the same channel.
Telematics
Usage-based insurance uses telematics to set premiums from the policyholder's own behaviour rather than from the past experience of a group. In motor insurance, a device or smartphone app records data such as miles driven, time of day, location, rapid acceleration and hard braking, and the insurer adjusts the premium accordingly. The NAIC names privacy as the main concern, because location and behaviour are tracked continuously. EIOPA's 2019 review found motor insurers using on-board devices, phone apps and GPS, and health insurers using wearable fitness trackers and smart watches. The services ranged from risk prevention to premium adjustments and emergency assistance.
Claims handling
Among American auto insurers, claims is the area where AI is most widely used. In the NAIC survey, 96 companies used it to provide information to adjusters, 58 to assign claims, 55 to evaluate images and 50 to help set settlement amounts. Nine used it for claim approvals, and none reported using it to deny claims. In the NAIC's 2025 survey of health insurers, between five and nine companies, depending on the product line, used AI in decisions to deny prior authorization. Computer vision systems estimate damage to vehicles and buildings from photographs. Tractable, one provider of such systems, said in 2023 that it processed about $7 billion of claims a year for insurers including Aviva, Geico and Admiral.
Fraud detection
Fraud detection was among the most common uses in the NAIC's auto survey, with models in use at 49 per cent of companies and under construction at another 9 per cent. In the Bank of England and FCA survey it was one of the three most frequent uses of AI across UK financial services, reported by 33 per cent of firms. Fraud screening can itself raise questions of discrimination. In December 2022 Black homeowners in Illinois filed a class action against State Farm under the Fair Housing Act, alleging that the program the insurer used to predict whether claims might be fraudulent was biased against Black policyholders. The plaintiffs cite statistics indicating that claims by white policyholders were processed faster and with less documentation. In September 2023 the federal court in Chicago allowed a disparate impact claim to proceed, and in 2026 the case was in discovery, with the first phase limited to State Farm's algorithmic tools for screening claims.
Customer service and generative AI
In a survey of 347 insurance undertakings in 25 European countries, published in February 2026, EIOPA found that nearly two thirds were using generative artificial intelligence, although most applications were still at the proof-of-concept stage. Of the use cases, 64 per cent concerned internal productivity, for example extracting data from invoices and medical reports or drafting emails and contracts, while 36 per cent were tools for customers, such as chatbots and voice assistants. Respondents named hallucinations as the main risk, ahead of cybersecurity and data protection. The share of insurers with a dedicated AI policy had risen from 25 per cent in 2023 to 49 per cent.
Wikipedia contributors · Artificial intelligence in insurance · revision 1380549084 · CC BY-SA 4.0 · Plain text extracted and formatting adapted; tags derived from linked topics.