Power to the Patients: AI for Fighting Insurance Denials

Holden Karau had battled her health insurance company a time or two before, but when a driver in an SUV slammed into her Vespa in 2019 while trying to make a left turn across four lanes of San Francisco traffic, things were different.

“I’d had a number of denials before, but they were all sort of spaced out,” she says. “When I got hit by a car, everything just happened all at once. And it was like, ‘Oh, this paperwork is insurmountable.’”

Karau’s career in technology began with building search and recommendation tools using early versions of artificial intelligence (AI). The accident got her thinking about ways to leverage similar technology to help patients deal with insurance companies.

But it wasn’t until her next insurance fight — this time with her pet insurance company — that she had finally had enough. Karau’s beloved sheltie, Professor Timbit, needed a root canal, and the insurance company did not want to pay
for anesthesia.

“When something happens to someone that we care about, even if that’s a dog instead of a person, it’s different,” she says, “because you feel the instinct to protect them, and you’re like, ‘No, this is not OK.’”

In 2023, Karau launched Fight Health Insurance, Inc. The firm offers AI-based services that streamline interactions with health insurers. For instance, customers can use the Fight Health Insurance AI to draft prior authorization requests tailored to boost the likelihood of approval. Customers can use the product to generate appeal letters if a claim is denied. The company offers its services free of charge for individual patients. It makes its money by charging medical practices, health systems and drugmakers’ patient access teams for the professional version of its software, called Fight Paperwork. The tool is trained on databases of appeals and medical review decisions to generate documents for insurance communication. It also offers enterprise-level case management.

Gearing up for a fight

Karau’s “Fight” comes at a time when advanced technology and strong emotions are colliding in the world of health insurance.

In October 2024, ProPublica reported that the medical benefits management company EviCore by Evernorth has begun using AI to process prior authorization requests and claims using an algorithm that can be adjusted in ways that can lead to higher rates of denials. In response, the company said its use of technology allows it to accelerate claims processing and ensure that patients get appropriate care and avoid services they do not need. The company also noted that doctors have the right to appeal if a prior authorization is denied.

Some states have moved to limit or ban the use of AI in insurance denials. In June, Arizona Gov. Katie Hobbs signed a bill that does not specifically mention AI but requires that insurers have licensed (human) medical professionals review any denied claims. Several states have also adopted model language from the National Association of Insurance Commissioners that, among other things, instructs insurers to create written guidance to govern their use of AI.

The insurance industry itself has largely been quiet about the use of AI. But in a statement sent to the Senate Finance Committee in 2023, the trade group AHIP cited several potential uses of AI by health insurance companies, including using such tools to identify gaps in care, better understand healthcare utilization and reduce unnecessary spending by catching fraud and waste.

Other companies

Fight Health Insurance is one of a small but growing number of firms that see a business opportunity in helping give customers the tools and confidence to fight insurance denials.

Counterforce Health has launched a product that allows patients to upload denial letters and insurance documents and then use AI to generate appeal letters. The company is also developing an AI voice product they say will conduct phone calls with insurers on behalf of patients.

A company called Claimable is doing similar work. It started with an initial focus on rheumatology cases but recently began offering appeal assistance for rejection of glucagon-like peptide 1 drugs, such as Wegovy (semaglutide) and Zepbound (tirzepatide).

One reason these companies have formed is that insurance appeals represent an unmet need within the healthcare system. An analysis by KFF of 2023 claims data from Affordable Care Act Health Insurance Marketplace plans found that less than 1% of the 73 million insurance denials that year were appealed.

Karau says many barriers keep patients from appealing. They may not know that they can. Although insurance paperwork usually indicates that members have the right to appeal, it often does not give patients detailed instructions about how to do so. There are also technical barriers. Even as the healthcare industry begins to embrace AI, many healthcare records are still exchanged via fax machine. Karau notes that many people no longer have printers at home, much less fax machines. “I wouldn’t say that these barriers were created intentionally, but there is very little incentive to remove these barriers,” she says.

Desire for justification

John Symons, Ph.D.

John Symons, Ph.D., who directs the Center for Cyber-Social Dynamics at the University of Kansas, says decisions made by an AI algorithm are “epistemically opaque,” alluding to the “black box” phenomenon in which the factors leading to AI-generated outcomes cannot easily be traced or explained. “They are not transparent tools,” he says. But Symons says it’s wrong to think that opacity is unique to AI. “​​A person working in an office is not a transparent tool,” he says. “We don’t know what’s going on in the head of the person who would be making that decision in place of the AI.”

Symons says the issue is not so much about technology but about the need for understanding. “What is legitimate is that people want to have decisions justified,” he says.

Some of that justification may come down to Big Data and personalized medicine. In the future, insurers — and health systems — may use Big Data to determine whether a particular treatment will likely work for an individual patient. If insurers rely on nonpublic datasets to make decisions, Karau says they ought to disclose the data behind those decisions. “If you’re going to argue that this drug isn’t effective, you have to show me the data that you’re using to argue that this drug isn’t effective,” she says. “You can’t just say, ‘Trust me.’”

Karau notes that tools like Fight Paperwork also create the opportunity for clients to get insights into insurers’ decisions on an aggregate scale. Such insights can be leveraged to help patients.

“There’s always the official policy, and then there’s also what could be described as unofficial guidance,” she says, “and I think one of the things that AI is really good at is sort of inferring what this unofficial guidance might be.” By reading between the lines, AI-based tools can help practices and patient-access programs identify and challenge unwritten rules.

Keeping humans involved

Symons says it makes sense for patients to use AI tools to deal with insurance bureaucracy. “The standard large language models that people are using are great tools to level the playing field in complex, bureaucratic contexts, in legal contexts, or even in academic life,” he says.

However, they also raise the specter of insurers using AI to deny claims, patients using AI to fight back and all of this happening without much involvement by humans. Symons says it will be key to find ways to ensure that healthcare remains human-centric. Part of that means affirming human accountability, even when AI is used. “We have to make sure that liability and responsibility are clearly assigned and that people can’t offload those onto the AI,” he says.

Symons adds he does not think technology like AI is inherently contrary to the goal of human-centric medicine. If done right — which he concedes is a significant caveat — AI can empower providers to have longer interactions with patients and provide insights to improve care. “It’s a complex landscape, and it could go really badly wrong,” he says. “But overall, I think we should welcome increased efficiency.”