Insurance law professor Daniel Schwarcz recently noted me in a LinkedIn comment to his post about a timely and important law review article he has written, Distributing Risk in an Age of AI: Procedural Bad Faith and AI Claims Handling. It is the type of article that many lawyers and judges will discuss in academic terms, using phrases like “procedural justice,” “algorithmic decision-making,” and “bad faith doctrine.” But I want to make certain that policyholders and public adjusters reading this blog understand how important this paper is to them. This article is about whether an insurance company can use artificial intelligence to help deny, delay, reduce, or steer a claim while pretending that a real human being made the decision.
The core point of Professor Schwarcz’s article is that bad faith law has usually focused on the result. Did the insurer wrongfully deny the claim? Did it delay payment without a reasonable basis? Did it fail to pay benefits owed under the policy? Schwarcz says that AI claims handling now creates another problem. The process itself can be unfair, even before we decide whether the final number is right or wrong. When a policyholder loses a home, business, roof, inventory, or livelihood, the claim should not be reduced to a data exercise where documents are uploaded into a portal, and a computer secretly decides the outcome.
That point matters enormously to policyholders and public adjusters. Public adjusters already know that claims are often won or lost in the details. What was inspected? Who inspected it? What was photographed? What was ignored? What pricing database was used? What assumptions were made? What communications shaped the estimate? What supervisors reviewed the file? What internal guidelines or claim severity goals influenced the final payment? Now add artificial intelligence to that mix, and the questions that need to be asked change. What machine, model, scoring system, estimating tool, fraud filter, or generative AI summary influenced the claim decision?
Insurance companies will say that AI merely helps trained adjusters do their jobs faster and more accurately. Sometimes that may be true. There is nothing inherently wrong with technology that helps pay claims faster, identify undisputed damage, organize a file, or make certain claims are fully paid. Nobody should be against newer or better tools used in the correct manner. A hammer is a fine tool until somebody starts using it on your thumb.
The concern raised by Schwarcz is different. The concern is whether AI becomes the invisible claims manager. The concern is whether the “human in the loop” is merely a human rubber stamp. Schwarcz makes the point that a human reviewer who simply accepts what an algorithm has already framed may not be exercising independent judgment at all. That is exactly what policyholders and public adjusters should be watching for in future claim files.
This is no longer future science fiction. Property insurers are already using tools to estimate damage from photographs, flag claims for fraud review, summarize claim files, draft claim communications, and support decisions about repair versus replacement. Those tools may be useful. They may also be dangerous if they are hidden from the policyholder, treated as proprietary black boxes, or used to justify underpayment without meaningful explanation.
The most important word in Schwarcz’s article may be “voice.” Policyholders are seeking a fair chance to explain what happened, what was damaged, challenge why the scope is incomplete, why matching matters, why code upgrades matter, why a building cannot be repaired the way a desk adjuster assumes, or why a contractor’s real-world estimate is not “inflated” simply because it is higher than a carrier’s software-generated number. If the insurance company’s process does not actually investigate facts of coverage and evaluate all considerations of damage, the policyholder has not received a fair claim adjustment.
This is where public adjusters should pay close attention. The future public adjuster will need to do more than write a better estimate. The future public adjuster will need to challenge the claim process. Was the field adjuster’s report altered by a desk review? Did a software tool remove line items? Did a generative AI summary omit key facts? Did a fraud algorithm flag the claim before anybody looked at the insured’s actual circumstances? Did the denial letter sound individualized, or did it look like a AI template dressed up with the insured’s name and claim number? We are already seeing these computerized letters written by AI.
I have written before about Professor Schwarcz’s scholarship because he studies the insurance marketplace from the consumer side of the transaction rather than simply accepting industry talking points. His work on insurance transparency and homeowner insurance markets has repeatedly shown that policyholders often cannot know what they are really buying, how claims will really be handled, or whether the promise of protection will be honored when the loss occurs. This new article fits that same theme. It asks whether policyholders are going to be treated as people with losses or as data points in a claims processing system. For those interested, I suggest also reading an article I wrote fourteen years ago, Insurance Regulators and Lawmakers, Judges and Insurance Consumer Advocates Should Study Professor Daniel Schwarcz’s Work.
The insurance industry will argue that AI lowers costs. But nobody is selling insurance policies by saying, “Buy from us because our computer may deny your claim faster.” Speed is wonderful when it results in prompt and full payment. Speed is not wonderful when it results in a quick, cheap, automated claims denial or underpayment.
The elephant in the room is disclosure. Policyholders and their representatives should be demanding to know whether AI or algorithmic tools were used in the claim. They should ask whether photographs were evaluated by computer vision, whether estimates were changed by automated rules, whether claim notes were summarized by generative AI, whether fraud scoring was used, whether supervisors were expected to follow the software recommendation, and whether the human adjuster had real authority to disagree with the computer.
Courts and regulators will eventually have to decide what “meaningful human review” means. This phrase cannot become another insurance company slogan. A meaningful review should mean that a trained adjuster actually looked at the claim and all its facts, considered the policyholder’s information, had authority to disagree with the software finding, and explained the decision in a human way the policyholder can understand. Anything less is not fair claim handling, but purely claim processing.
The practical lesson from the article is the claim file of the future will not just be about estimates, photographs, engineering reports, and correspondence. It will also be about workflow, software, prompts, algorithms, vendor systems, audit trails, override rates, and whether a human being truly made the claims decision. Public adjusters who understand that will be better prepared for the next generation of claim disputes.
Professor Schwarcz’s article matters because it gives lawyers, judges, regulators, policyholders, and public adjusters a vocabulary for something many in the field already suspect. A fair claim is not just a coverage decision and damage evaluation. Instead, it is a fair claims process. A policyholder who has been damaged by fire, wind, hail, water, collapse, theft, or catastrophe deserves more than a computerized answer wrapped in a polite AI written denial letter.
If artificial intelligence helps insurers keep the insurance promise, it should be welcomed. If it helps insurers hide how the promise is broken, it should be challenged early and loudly. Professor Schwarcz has raised this important issue with insurance academia for comment and further study.
Thought For The Day
“Insurance is not merely a financial product; it is an ongoing promise of protection purchased in anticipation of a future moment of need.”
— Professor Daniel Schwarcz, “Distributing Risk in an Age of AI: Procedural Bad Faith and AI Claims Handling”