Property insurance underwriting has always been about information. The insurer wants to know what it is being asked to insure, what hazards exist, what the likely loss exposure may be, and what premium should be charged for accepting that risk. None of that is new. What is new is that Artificial Intelligence is transforming the speed, scope, and invisibility of how that information is gathered and used.
We are moving from application-based underwriting to intelligence-based underwriting. The underwriting file is no longer built only from what the policyholder, agent, broker, property manager, or condominium association submits. Increasingly, it is built by digital machines searching, collecting, comparing, and interpreting public and private information about a property before a human underwriter ever decides what to do with the account.
A property may now have a digital underwriting file assembled from aerial imagery, building permits, tax records, roof analytics, fire protection data, flood maps, wildfire models, claims history, inspection reports, code enforcement records, satellite imagery, vegetation measurements, elevation data, and surrounding-property conditions. Artificial intelligence can help find this information, organize it, interpret it, and flag what the insurer may consider a concern. The insurer may know, or believe it knows, a great deal about a property before the policyholder has any meaningful opportunity to explain the property's condition, improvements, maintenance, or mitigation.
The phrase "AI-driven risk score" is often used as if a single number is making the decision. Sometimes that may be close to true. A third-party vendor to an insurer may provide a wildfire score, roof condition score, hail vulnerability score, or other property-specific risk score that is used in rating, eligibility, or renewal decisions. But in many situations, the reality is more complicated. The insurance carrier's unique underwriting rules and appetite for risk guidelines still determine what will be accepted, rejected, surcharged, restricted, or non-renewed.
A roof condition algorithm may say the roof appears worn. A wildfire model may say the property has insufficient defensible space. A computer vision system may identify overhanging trees, debris, roof staining, a pool, a trampoline, solar panels, or neighboring exposures. A search agent may find that no permit was pulled for a roof replacement or that a building department record conflicts with an application. But the premium consequence comes when the insurer decides what that information means. One insurer may charge more. Another may demand repairs. Another may exclude certain coverage. Another may refuse to write the risk at all.
So, is this truly an AI risk score? Sometimes. But often it is something more difficult to see and challenge. It is AI-assisted underwriting judgment. The machine may gather the facts, arrange the file, flag the concern, and suggest the risk classification. The human underwriter or the carrier's rulebook then decides whether that concern becomes a higher premium, a repair demand, a coverage limitation, or a nonrenewal.
That has enormous consequences for policyholders. If a filed rating plan uses an approved score, regulators can at least ask how the score is developed, tested, and applied. But if AI is merely feeding the underwriting process with images, property characteristics, and red flags, the consumer may never see the actual source of the decision. The policyholder may receive a premium increase, repair demand, nonrenewal notice, or coverage restriction with a vague explanation that the property does not meet underwriting guidelines.
Those underwriting guidelines are where much of the real action occurs. Insurers are not all using the same definition of acceptable property risk. One carrier may tolerate an older roof if it has been well maintained. Another may decide that roof age alone is enough to move the account into a higher premium tier or out of eligibility. One insurer may reward documented mitigation. Another may not meaningfully credit it. One may use AI data as a starting point for a human inspection. Another may treat an aerial image as if it were conclusive.
The danger is not that insurers are using better information. Better information can lead to better underwriting. It can help distinguish one property from another rather than punishing an entire neighborhood, county, or state with broad-brush pricing. In a world of wildfire, hail, windstorm, flood, water damage, aging infrastructure, and rising reinsurance costs, insurers have a legitimate need to understand the risk they are assuming.
The danger is secret information being used in a way the policyholder cannot see, correct, challenge, or improve upon. Aerial images can be outdated. Shadows can be mistaken for damage. Roof staining may not mean functional deterioration. A permit record may be incomplete. Public records may lag behind actual repairs. A property may have undergone mitigation that the model does not recognize. A condominium building may have completed fire safety improvements or roof repairs that are not obvious from a satellite image. A wildfire score may not reflect new defensible space work. A flood model may not account for drainage improvements. The computer may find data, but it does not always find the truth.
This is where fairness must enter the underwriting conversation. If insurers can use artificial intelligence to build a property file, policyholders should have the right to know what is in that file when it affects premium, eligibility, renewal, or coverage. They should be told what specific property characteristics drove the decision. They should be allowed to correct inaccurate information. They should be given a meaningful opportunity to provide photos, permits, engineering reports, maintenance records, mitigation documentation, and other evidence showing the true condition of the property.
Regulators are beginning to understand that this is not science fiction. The National Association of Insurance Commissioners adopted a Model Bulletin on the use of artificial intelligence systems by insurers, making clear that insurance decisions supported by AI still must comply with existing laws against unfair discrimination, unfair trade practices, and improper rating or underwriting conduct. AI does not get a hall pass simply because the technology is new. If the decision affects a consumer, the insurer remains responsible for the decision.
Some states are moving even more directly. New York's insurance regulator has issued guidance on the use of artificial intelligence systems and external consumer data in underwriting and pricing. Colorado has been at the forefront of regulating insurer use of external consumer data, algorithms, and predictive models, requiring governance and risk management around the possibility of unfair discrimination. California has also warned insurers about the use of data, models, and algorithms in ways that may create unfair bias or improper underwriting results. The details differ by state, but the message to insurers is that using AI is not an unregulated playground.
This regulatory attention should be welcomed by honest insurers. The issue is not whether insurers may use technology. They will, and they should. The issue is whether they can explain what they used, why it mattered, whether it was accurate, and whether the policyholder had a fair chance to respond. A carrier should not be allowed to hide behind a third party vendor score, a model, or a vague underwriting guideline when the result is a higher premium or a lost policy.
The future of property underwriting should not be a one-way surveillance system. It should be a transparent risk dialogue.
This is especially important for condominium associations, commercial property owners, and homeowners in catastrophe-exposed regions. These policyholders are already facing dramatic premium increases, higher deductibles, restricted coverage, and limited market options. If an AI-assisted process is going to label their property as high risk, they deserve more than a mystery score and a take-it-or-leave-it renewal quote.
A black-box underwriting culture will invite regulatory pushback. A transparent process, by contrast, can help insurers defend their decisions, encourage loss mitigation, and maintain a more stable marketplace. If a carrier can say, "Here are the risk characteristics we identified, here is why they matter, and here is what you can do to improve the risk," that is a far better system than simply pricing people out or dropping them without meaningful dialogue.
Artificial intelligence will not disappear from property insurance underwriting. Nor should it. Used properly, it can help identify risk, encourage mitigation, and make pricing more accurate. But the industry must be honest about what is happening. This is not always a neutral "AI risk score" mechanically producing a premium. Often, AI is gathering information and giving insurers new power to choose what they will accept, what they will reject, and what they will charge.
This power must come with responsibility. Before a policyholder pays more, receives less, or loses coverage altogether, fairness demands the policyholder see the evidence and tell the full true story of the property.
