Sunday, 6 September 2026
Artificial intelligence is beginning to change one of the most traditional parts of the financial sector, but the biggest opportunity for insurers may not come from replacing employees or simply processing policies faster. Instead, AI could become a tool for understanding risk more precisely, expanding underwriting capacity and allowing insurers to enter markets that have become increasingly difficult to cover. That was one of the central messages from an insurance panel at AI4 2026 in Las Vegas, where executives from The Hartford, Tokio Marine and TomTom joined Scale Venture Partners to discuss how artificial intelligence is moving into underwriting and claims.
For insurers, the attraction is straightforward. Underwriters spend substantial amounts of time collecting information, reviewing documents, checking property details and summarising material before they can perform the higher-value work of assessing risk and structuring coverage. AI can increasingly perform some of this preparatory work, allowing experienced professionals to concentrate on complex risks, customer relationships, loss prevention and difficult underwriting decisions. The Hartford is pursuing this approach by using AI to support rather than replace underwriters, with the objective of automating information gathering and routine analysis so underwriting teams can spend more time solving problems for customers.
This becomes particularly important in commercial insurance, where individual accounts can involve numerous properties, operating risks and coverage requirements. A company seeking property insurance might have ten locations, for example, of which eight present relatively straightforward risks while two require detailed investigation. Traditionally, an underwriter may need to review all ten. AI could increasingly handle much of the preliminary work associated with the simpler properties, leaving the underwriter with more time to investigate the difficult locations. The potential consequence is greater underwriting capacity rather than simply lower staffing costs.
Insurers could therefore devote more expertise to risks that are currently difficult to price or insure, including properties exposed to wildfire, flooding and other increasingly complex hazards. That could become important as parts of the insurance market struggle with changing catastrophe exposure. In locations where insurers have reduced capacity or withdrawn coverage, better information and more sophisticated risk modelling could potentially allow carriers to distinguish between properties more precisely rather than treating entire areas as similarly risky. AI therefore has the potential to change insurance at a more fundamental level than administrative automation. If insurers can understand individual assets and businesses more accurately, they may be able to expand the range of risks they are willing to cover while pricing them more precisely.
The panel distinguished between several stages of AI adoption. Giving employees access to generative AI tools can improve individual productivity, while automating established workflows can reduce processing times. Neither necessarily provides a lasting competitive advantage because rival insurers can adopt similar technologies. If one carrier reduces the time required to turn an insurance submission into a quotation from a week to a day, competitors will eventually be expected to provide the same service. What initially creates an advantage gradually becomes the new market standard.
The more valuable opportunity lies in combining AI with proprietary information and underwriting expertise in ways competitors cannot easily reproduce. Insurance companies possess decades of claims histories, underwriting decisions, broker relationships, policy information and risk-engineering experience. Connecting those assets through AI could help companies make materially better decisions about which risks to accept, how much capacity to provide and what conditions should apply. This is where AI begins moving from productivity technology towards a source of underwriting advantage.
Experienced underwriters accumulate considerable knowledge during their careers, but much of that expertise historically remains with the individual. Two professionals reviewing the same account can reach different conclusions because their experience, judgement and interpretation of risk differ. AI creates an opportunity to capture more of that institutional knowledge and make it available across the organisation. Rather than eliminating professional judgement, the technology could give every underwriter access to a broader body of historical experience and information.
That is particularly significant because the insurance industry faces a demographic challenge. Many experienced underwriters and claims specialists are approaching retirement, while insurance does not always attract enough younger professionals to replace them. Capturing institutional knowledge before experienced employees leave could therefore become an important part of the industry’s AI strategy.
Location intelligence demonstrates how external data can improve these decisions. TomTom, historically known for consumer navigation, has increasingly developed its business around geospatial information covering roads, traffic, addresses, speeds, incidents and other location characteristics. Insurance risk is inherently connected to geography. Two properties in the same postcode can have materially different exposure depending on their precise position relative to roads, vegetation, terrain, flood zones or other physical characteristics. Moving from broad geographic assumptions towards increasingly granular location analysis can therefore improve risk assessment.
TomTom described applications where detailed road and traffic information is being incorporated into insurance analysis. The company also discussed work involving claims verification, where the reported location and timing of a vehicle accident can be compared with traffic and incident information to assess whether the circumstances are consistent with the claim. This illustrates another important shift in enterprise AI. Large language models themselves are becoming increasingly available across the market, meaning the differentiating factor may increasingly become the information surrounding the model rather than the model itself.
For insurers, this means proprietary claims information, detailed geospatial data, weather information, building characteristics, historical losses and other specialist datasets could become more valuable as AI makes them easier to combine and analyse. The challenge is avoiding information overload. Underwriters are not necessarily data scientists. Providing hundreds of additional variables does not automatically produce better underwriting. The system needs to determine which information matters for the particular risk being considered and present it in a form that supports a decision. This could become one of the most valuable functions of insurance AI: converting increasingly large quantities of data into a manageable explanation of the factors that genuinely affect a particular property, company or policy.
Claims provide another area where this combination of AI and external information could change existing processes. Crop insurance offers a useful example. When hail damages a large agricultural property, traditional claims assessment can involve an adjuster physically inspecting portions of a field and estimating how much of the crop has been affected. Satellite imagery, aerial information and AI-based image analysis could provide a broader view of the damage, potentially allowing insurers and farmers to establish the affected area more consistently.
The value is not simply faster claims handling. Better evidence can also reduce disagreement between insurer and policyholder. That question of trust is particularly important because insurance depends on customers believing that claims and underwriting decisions have been reached through a legitimate process. The panel repeatedly returned to the need for accountability and auditability when AI becomes involved in decisions. Insurance differs from many consumer applications of artificial intelligence because errors have financial and regulatory consequences. An automated system cannot simply make an unexplained decision and transfer responsibility away from the insurer.
This becomes even more important as companies experiment with AI agents capable of taking actions rather than merely providing information. Insurers may be beyond the initial pilot stage in selected applications, but widespread autonomous deployment remains constrained by governance, regulation and the difficulty of controlling non-deterministic systems. An insurer may successfully automate one workflow within one line of business while still operating dozens or even more than a hundred other insurance products through conventional processes. Scaling AI across such organisations requires considerably more than proving that an individual application works.
The industry’s preference for consistency also matters. Insurance companies make decisions based on long histories of losses and probabilities. Property underwriting can involve catastrophe models based on events expected to occur once in a century or even less frequently. Technology that has existed for only months therefore needs to demonstrate that it can operate reliably within organisations accustomed to measuring risk over decades. This is why the transition from AI pilot to enterprise infrastructure is likely to take time. Insurers need systems that can be monitored, audited and integrated with established controls before they can rely on them for material underwriting decisions.
The process may resemble earlier technological transformations in which businesses initially used new technology to reproduce old workflows before eventually redesigning operations around its capabilities. Simply inserting AI into an existing underwriting process may create efficiencies, but the larger opportunity comes from reconsidering why each stage exists and whether it remains necessary. That distinction is increasingly shaping how insurers allocate AI investment.
Companies need to determine which capabilities are central to their competitive position and which can be purchased from technology providers. Underwriting, claims management and risk engineering are generally regarded as core insurance capabilities. The data, models and decision systems supporting those activities can directly affect profitability and therefore represent areas where insurers may want to retain greater control. More generic corporate applications can potentially remain with external software providers. Building every AI application internally would require insurers to maintain software, security, user interfaces and continual product development in areas that provide little underwriting differentiation.
The dividing line increasingly appears to be the insurer’s proprietary data and decision architecture. An insurer may use external models, cloud infrastructure and software, while retaining control over the systems that combine those technologies with decades of internal claims and underwriting knowledge. This also creates opportunities and risks for insurance technology start-ups. Insurers are interested in specialised providers capable of solving difficult problems or supplying information they cannot easily obtain themselves, but technology companies attempting to replace large portions of the insurance workflow may encounter resistance from organisations that already employ substantial teams and operate highly specialised processes.
The economics of AI vendors are another concern. Enterprise customers can redesign workflows around a technology only to discover that the supplier later changes from user-based pricing to consumption-based pricing as AI usage increases. Once a process has become dependent on an external platform, a substantial price increase can create operational and financial risk. For insurers, this means AI procurement is increasingly becoming part of risk management. Companies need to consider whether a supplier will remain economically viable at scale, how easily technology can be replaced and which institutional capabilities should never become dependent on a single outside provider.
The underlying AI models themselves may eventually become less important. As competing models improve and open-weight alternatives become more capable, enterprises could increasingly treat the model as interchangeable infrastructure. The more defensible value would then sit in proprietary information, workflow integration and the surrounding technology ecosystem. For insurance, that reinforces the importance of context. A general AI model may be capable of reading an insurance submission, but it does not automatically possess the insurer’s historical loss experience, risk appetite, broker knowledge, policy interpretations or underwriting strategy. Combining those elements is where insurers expect AI to become strategically important.
The financial case also differs from industries where labour represents the primary cost. Insurance profitability depends heavily on the relationship between premiums collected, claims paid and operating expenses. Saving a relatively small amount of administrative cost is useful, but making a materially better decision about a large risk can be far more valuable. Automating a process that saves a few dollars while introducing additional poorly priced risks would be economically counterproductive. This is one reason the panel resisted the idea that AI’s main purpose should be reducing insurance employment.
Both The Hartford and Tokio Marine representatives argued that AI should make insurance professionals more capable rather than simply remove them. The expectation is that underwriters and claims specialists equipped with better information can handle more complex work, investigate difficult risks and provide better service. That could ultimately increase demand for skilled professionals if insurers use AI to expand into risks they currently avoid. Climate exposure, cyber aggregation and emerging technologies are creating categories of risk that require increasingly sophisticated analysis. Greater automation of routine work could free human expertise to concentrate on these areas.
The consequences could extend beyond insurance companies themselves. Insurance is an important component of investment, property development and corporate financing. Assets and businesses that cannot obtain affordable insurance can become difficult to finance, transact or operate. This is particularly relevant for real estate. Increasing catastrophe exposure has already complicated property insurance in some markets. If AI, geospatial intelligence and better catastrophe modelling allow insurers to assess individual assets more accurately, the technology could influence not only insurance pricing but property liquidity and investment decisions.
A building in a broad high-risk area might nevertheless demonstrate characteristics that make it more resilient than neighbouring properties. More granular underwriting could potentially recognise those differences, giving owners stronger incentives to invest in resilience measures. The same principle could eventually apply to cyber insurance, supply-chain disruption and other difficult-to-model risks. Better information could expand the boundaries of what insurers are prepared to cover.
That may prove to be AI’s most important contribution to the sector. The first phase of insurance AI has concentrated heavily on documents, productivity and automation. The next phase is moving towards underwriting intelligence: deciding which risks are acceptable, identifying which characteristics genuinely matter and allocating insurance capacity more effectively.
The industry still faces significant obstacles around governance, model reliability, data quality, vendor dependence and organisational change. It also needs to ensure that increasingly automated decisions remain explainable and that customers continue to believe they are being treated through a fair and accountable process. But the economic incentive is substantial. Insurers that use AI merely to process the same business faster may achieve temporary efficiencies. Those that use it to understand risk better could create a more durable advantage.
For businesses, property owners and consumers, the difference could eventually be significant. Better underwriting should not simply mean faster quotations. At its most effective, it could mean more accurately priced risk, quicker claims, greater insurance capacity and coverage becoming available for assets and activities that insurers currently find difficult to understand. The transformation of insurance through AI is therefore likely to be measured less by how many underwriting tasks become automated and more by whether the industry becomes better at deciding which risks it is prepared to take.
Source: CIJ.World Research & Analysis Team
