The Algorithm That Decides Your House Fire
The insurance claims adjuster was once the human face of disaster. When a house burned down, a flood swallowed a living room, or a tree crashed through a roof, this person arrived with a clipboard, a flashlight, and a practiced sense of empathy. They walked through the wreckage, catalogued the damage, and translated a family’s loss into a number that would let them rebuild. It was a job built on judgment, on the ability to read a situation beyond what the visible damage suggested. That role is now being systematically dismantled, not by a rival company or a regulatory shift, but by software that analyzes photos and generates estimates in seconds. The adjuster is becoming a relic, a human backup system for a machine that never needs to see the tears, smell the smoke, or understand what it means to lose everything at once.
The numbers tell a story that is hard to ignore. On the job review platform Glassdoor, one faction hates artificial intelligence more than any other. Claims adjusters who mention AI in their posts are critical a whopping 98 percent of the time, enough to win them the mantle of the biggest AI haters in the American workforce. [2] That statistic is not a quirk of survey methodology or a handful of disgruntled outliers. It is the sound of a profession watching itself become obsolete in real time. The Bureau of Labor Statistics projected that the number of claims adjusters in the United States would fall by 18,900, or 5 percent, over the coming decade. [2] Between May 2024 and May 2025, employment in the sector dropped a staggering 21 percent, according to BLS data. [2] For early-career adjusters, the decline was even sharper: Entry-level postings have fallen 50 percent since 2024, according to Glassdoor.
The speed of this collapse is what makes it remarkable. Technological displacement usually takes generations, giving workers time to adapt, retrain, or find adjacent roles. The adjuster’s decline has been compressed into a few short years, a sudden cliff rather than a gradual slope. Insurance companies, always hungry for efficiency and cost reduction, have embraced AI with an enthusiasm that borders on recklessness. AI startups like Liberate and Pace have raised millions of dollars based on promises to “reinvent” insurance, while insurers have scaled their use of AI to handle more and more claims. The pitch is always the same: faster payouts, lower overhead, happier customers. The reality, as the Glassdoor reviews make clear, is a workforce that feels betrayed by the very tools designed to replace them.
Ahmad Jackson knows this betrayal firsthand. About a year ago, he was working in the claims department for a major insurance company. His employer decided to use AI for initial loss reporting, which involves setting up claims and gathering information when someone first discloses an incident. It was supposed to be a boon for employees and policyholders alike, streamlining simple claims while transferring more complex situations to actual people. Instead, Jackson says, he and other adjusters faced a sudden influx of misclassified claims that required rerouting to the correct department. He encountered AI-driven hallucinations while reading claims summaries; when he inadvertently relayed those errors to claimants or their attorneys, he bore the brunt of their fury. He quit the company not long after, switching to a different insurance carrier. AI is “getting things wrong,” Jackson tells WIRED. “And it’s implementing more work onto the adjusters.” [1]
The irony is thick enough to cut. The technology that was supposed to make adjusters more efficient has instead made their jobs harder, at least in the short term. When an AI misclassifies a claim, a human has to untangle the mess. When a hallucinated detail slips into a summary, the adjuster becomes the face of the error, absorbing the anger of people who are already having the worst week of their lives. The machine does not feel the sting of a claimant’s accusation that they are incompetent, heartless, or corrupt. The machine does not have to live with the knowledge that a mistake delayed a family’s recovery by weeks. Those costs are externalized onto the human workers who remain, the ones tasked with cleaning up after an algorithm that was supposed to make them redundant.
“There’s an AI fatigue,” says Geoffrey Conrad, a claims executive in Mobile, Alabama. “We’re pretty much exhausted as far as the amount of AI being shoved down our throats.” Conrad has spent decades in the industry, and he speaks with the weary authority of someone who has seen the business transform from a people-centered profession into a data-processing operation. He insists that a good claims adjuster is one who is both empathetic and determined to get the policyholder as much money as possible. He knows, he says, because he’s been there. Twenty-five years ago, before Conrad started working in the insurance industry, his house burned down. It was a total loss. “If you would have told me that someone was just gonna come take photographs, and an estimate would be generated [by AI] off of photographs, I probably would have lost my mind,” Conrad says. “I needed somebody to make sure that I’m OK, my family’s OK, that we’re safe.” [1]
The human need that Conrad describes is not something an algorithm can quantify. When a family stands in the charred remains of their home, they are not looking for a payout figure. They are looking for reassurance, for a sign that someone else understands the enormity of their loss and will guide them through the bureaucratic maze that follows. That is the essence of the adjuster’s craft, the part that cannot be captured in a training dataset or replicated by a neural network. It is the ability to sit with a stranger in their moment of crisis and say, without words, that you will take care of them. AI can process a claim in seconds, but it cannot hold a hand. It cannot look a person in the eye and convey that their suffering matters. It cannot make the distinction between a house and a home.
Glassdoor senior economist Chris Martin didn’t expect claims adjusters to hate AI quite this much. When he saw the results, he tells WIRED, he did a double take—then started digging. Martin’s conclusion: The profession is in the midst of a reckoning. The Glassdoor data revealed a pattern that went beyond simple job insecurity. Adjusters complained about “AI-obsessed leaders forcing error-prone AI on them and their clients,” according to the research. When AI falls short in this job, adjusters say, they have to clean up a very human mess. Martin says that reviewers also express skepticism about AI when they feel a subpar product is being forced upon them—or upon clients. The resentment is not just about losing jobs; it is about being forced to rely on tools that are demonstrably worse at the job than the humans they are replacing.
The gap between AI’s promise and its performance is where the real damage happens. Claims adjusters don’t have “a lot of faith in the output,” says Sandy Avina, a claims adjuster turned insurance industry consultant. A smudge in a document from an attorney can result in a hallucination, which leads to an incorrect payout. A missing point in an AI summary of a medical report can do the same. Customers often don’t realize when AI is the root cause of confusion or misinformation, and they assume the adjuster is the one who made the mistake. The human worker becomes the scapegoat for errors they did not commit, the fall guy for a system that was supposed to eliminate fallibility but merely relocated it. The adjuster’s professional reputation, built over years of careful work, can be destroyed by a single algorithmic glitch.
There are instances where AI can make claims adjusters’ lives easier; Jackson says it’s helpful for administrative tasks. He uses it for what he considers “nuisance calls,” like extending someone’s rental car booking by a few days. This is the mundane side of the job, the paperwork that eats up hours and provides little satisfaction. Automating these tasks seems like an unqualified win, a way to free up humans for the work that actually matters. But the slippery slope is real. Once the mundane tasks are automated, the next logical step is to automate the slightly more complex tasks, and then the genuinely difficult ones. Each step is justified as a way to help the human workers, but each step also brings the industry closer to the point where the humans are no longer needed at all. The training data for the next AI model is being generated by the very people who are training their replacements.
That is the goal of Lemonade. Since its founding in 2015, the insurance company has promised to replace bureaucracy with “bots and machine learning.” By the end of last year, its proprietary chatbot, AI Jim, handled initial reports 96 percent of the time, with automation handling roughly 55 percent of all claims. [3] While more traditional insurance companies like State Farm emphasize that claims require a mix of human and digital expertise, some claim adjusters still worry they’re training their replacements. “AI is going to impact many jobs and threaten many incumbents in insurance and the economy,” a Lemonade spokesperson, Paul Staats, tells WIRED, adding that automating allows employees to focus their “empathy, care, and expertise on the most complex claims.” [3] Glassdoor’s report “should be taken seriously across the industry,” Staats adds.
The Lemonade model is the clearest articulation of what the insurance industry is becoming. The company does not pretend to be doing anything other than replacing human judgment with machine processing. Its entire business strategy is premised on the idea that an algorithm can do the job better, faster, and cheaper than a person ever could. The chatbot handles the initial report, the automation processes the claim, and the payout arrives in seconds. The human employees who remain are reserved for the edge cases, the ones too complex or too sensitive for the machine to handle. This is presented as a virtue, a way to deploy human empathy where it matters most. But it also means that the vast majority of policyholders will never interact with a human being during their moment of crisis. They will talk to a chatbot, upload their photos, and receive a number that an algorithm decided was fair.
The question of fairness is central to the adjuster’s anxiety. Insurance is, at its core, a system of trust. Policyholders pay premiums for years, trusting that when disaster strikes, the company will honor its obligations. The adjuster is the embodiment of that trust, the person who ensures that the promise is kept. When an AI makes the decision, that trust is transferred to a black box. The policyholder has no way to understand why they received a certain amount, no way to appeal to a human judgment that might see the full picture. The algorithm operates on patterns, on statistical correlations, on the data it has been trained on. It does not know that the family photos in the living room were irreplaceable, that the antique dresser had been passed down for generations, that the workshop in the garage was a father’s sanctuary. It only sees the square footage and the type of flooring and the estimated cost of replacement.
Conrad, the Alabama claims adjuster, says people in his profession do feel like they’re being replaced by AI models. “AI is just a tool,” he says. “It should never be given the keys.” The metaphor is apt. A tool is something you use, something that extends your capabilities without replacing your judgment. Keys are what you hand to someone you trust to drive the car. The industry, in its rush to efficiency, has handed the keys to a system that does not understand the road, the weather, or the passengers. It is driving blind, and the adjusters are the ones left to clean up the wreckage. They are the ones who have to explain to a grieving widow that the algorithm made a mistake, that the payout will have to be recalculated, that the process will take longer than expected because the machine got it wrong.
The psychological toll of this arrangement is difficult to overstate. Adjusters are trained to be the calm in the storm, the professional who can handle the most stressful situations with grace and competence. Now they are forced to be apologists for a system that does not deserve their defense. They are asked to vouch for outputs they do not trust, to stand behind decisions they did not make, to absorb blame for errors they did not commit. The cognitive dissonance is corrosive. It eats away at professional pride, at the sense of purpose that comes from doing a difficult job well. It is no wonder that the Glassdoor reviews are so bitter, so uniformly negative. The people writing those reviews are not Luddites or technophobes. They are professionals who have watched their craft be devalued, their judgment be overridden, their humanity be deemed unnecessary.
The irony is that the AI systems are not even good enough to justify the disruption. They make mistakes, they hallucinate, they misclassify and misestimate. The 98 percent negative review rate suggests that the technology is not delivering on its promises, at least not from the perspective of the people who have to work with it every day. The insurers who have embraced AI are not saving money if the errors cost more in the long run, if the misclassified claims require expensive rerouting, if the incorrect payouts lead to litigation. The efficiency gains are real, but they are offset by the costs of the mistakes. The industry is betting that the AI will improve, that the hallucinations will become rarer, that the misclassifications will become less frequent. That bet may pay off eventually, but it is being made with the careers and livelihoods of human workers as the collateral.
The pattern is not unique to insurance. Across industries, automation has historically displaced workers before new roles emerged. The adjuster’s situation differs in one crucial respect: the BLS projections suggest outright contraction, not transformation. The traditional adjuster role, with its blend of technical assessment and human judgment, may simply cease to exist.
The deeper question is whether the goal we are pursuing is the right one. Efficiency is a virtue, but it is not the only virtue. There is something to be said for a system that treats people as individuals, that recognizes the complexity of human loss, that provides a measure of comfort along with the compensation. The adjuster who visits a burned-out home is not just gathering information; they are bearing witness. They are confirming that the loss is real, that it matters, that the person who suffered it is not alone. An algorithm cannot bear witness. It can only process data. The insurance industry is in danger of optimizing for speed and cost while sacrificing the very thing that makes insurance meaningful: the promise that a company will stand by its customers when they need it most.
The data from Glassdoor suggests that this sacrifice is not going unnoticed. The 98 percent negative review rate is a warning sign, a signal that the workforce is losing faith in the direction of the industry. When workers sense layoffs looming, Glassdoor data indicates that their reviews become increasingly anti-AI. The reviews are not just about the technology; they are about the leadership that is forcing the technology on them. The adjusters are angry at the “AI-obsessed leaders” who are making decisions from the executive suite, far removed from the realities of the claims process. They are angry at a system that values cost-cutting over quality, that treats human judgment as a liability rather than an asset, that is willing to sacrifice the trust of policyholders for a few percentage points of margin.
Jackson, the former adjuster who quit his job, has found a way to use AI that does not threaten his sense of purpose. He uses it for the administrative busywork, the tasks that no one enjoys and that take time away from the work that matters. This is the model that could work, the one where AI is a tool rather than a replacement. But it requires a discipline that the industry has not shown. It requires leaders who are willing to resist the siren song of full automation, who understand that some jobs are worth preserving even if they are not the most efficient. It requires a recognition that the human element is not a bug to be eliminated but a feature to be cherished. The industry is at a crossroads, and the choices being made now will determine whether the adjuster becomes a footnote in history or a model for how to integrate AI without sacrificing humanity.
The adjuster’s plight is a case study in what happens when AI is deployed without regard for the human consequences. The lesson is not that AI is bad, or that it should be resisted. The lesson is that automation must be done carefully, with an eye toward the human beings who are affected. Efficiency is not the only metric that matters; there are values that cannot be quantified. The adjusters who hate AI are not standing in the way of progress. They are trying to save something precious, something that the industry is in danger of losing forever.
The question that hangs over the entire discussion is whether the pursuit of efficiency is worth the human cost. The insurance industry has convinced itself that AI is the future, that the old ways are obsolete, that the adjuster is an expensive luxury it can no longer afford. But the 98 percent negative review rate suggests that the industry is making a mistake, that the cost of the AI may be higher than the savings. The adjusters who remain are demoralized, overworked, and resentful. The policyholders are being served by a system that does not understand them. The trust that is the foundation of insurance is being eroded one claim at a time. And for what? For a few seconds of processing time, a few dollars of savings, a few percentage points of margin. The trade seems profoundly unequal, a bargain that the industry will regret when the consequences become clear.
Sources
3. FDA