Around $5 trillion is spent on U.S. healthcare each year, a system that many Americans rate poorly, and one which will confront even greater challenges in the years ahead as annual expenses continue to rise, the population ages, and clinical workforce shortages persist.
Pharmaceutical companies, hospital systems, insurers, and pharmacy benefits managers operate in a complex industry that’s earned a reputation for siloing data and badly lagging other sectors like banking and telecom when it comes to deploying new technological advancements. Greater investments in artificial intelligence are lauded as a key treatment for the healthcare system’s most serious problems, though most in the industry report they still lack sufficient resources and planning in AI and other digital investments.
Sandeep Dadlani, the chief digital and technology officer at insurance giant UnitedHealth Group, says he understands that the system is broken. “I empathize with everybody who has a grievance towards healthcare,” says Dadlani. “The system has to be fixed by many players, not just by us.”
UnitedHealth, ranked fourth on the Fortune 500, says it is seeing some big productivity gains and adoption from the company’s 1,000 AI use cases already in production. These applications, a mix of generative and traditional forms of AI, include more than 60 million lines of AI-written software code that has already been accepted by around 20,000 engineers, 65 million customer calls that were initially answered by an AI chatbot in 2024, and 18 million AI-enabled searches to help customers find a doctor that were conducted in the first quarter of 2025 alone.
Another generative AI tool expected to launch later this year in the UnitedHealthcare and Optum subsidiary apps is a consumer-facing conversational bot that will help customers find a doctor, schedule appointments, or review their lab results. That tool was initially tested with UnitedHealth’s employees before it will debut externally.
“It’s a maze,” says Dadlani, of the labyrinth that customers face when dealing with healthcare providers, insurers, and laboratory services. “So something that helps navigate through that and provides great end-to-end access, that’s the dream.”
Dadlani spoke to Fortune in early May, before the surprise announcement that CEO Andrew Witty had resigned. Dadlani’s efforts overseeing the company’s AI evolution come as UnitedHealth, which serves more than 52 million consumers, undergoes a difficult period. In December, UnitedHealthcare CEO Brian Thompson was murdered in Manhattan and in April, the company’s first financial results since the killing, UnitedHealth reported earnings that missed Wall Street expectations and cut its bottom-line target for the full year. The company is also reportedly facing a Justice Department probe for possible criminal Medicare fraud.
There is also a class-action lawsuit in the courts that alleges UnitedHealth has used AI to illegally deny claims. Dadlani refutes this. “AI is never used to deny a claim,” he says. “If a claim is not eligible to be approved, it goes up to a human agent,” who, he says, then makes the final determination.
Dadlani says around 90% of claims are auto adjudicated, the process in which software is able to manually review a claim. Of the 10% of claims that go through an extra step for review, most of the issues are clerical, meaning there are missing details or the information wasn’t input into the system properly. After that manual review, Dadlani says 98% of claims are approved and for the remaining 2%, denials tend to be attributed to either ineligible benefits or because of clinical or medical safety determinations.
Later this year, Dadlani says he intends to launch new AI products that will make auto approval for claims even higher. “We are already seeing in our early experimentation that AI can help fill some of the missing information for these claims,” he says.
UnitedHealth’s vast AI portfolio is reviewed on a monthly basis by every business unit. On a quarterly basis, chief information officers come together to monitor how AI is being used across the enterprise. The company has also established a responsible AI board that is a mix of internal and external technologists, clinicians, legal experts, and others who review hundreds of AI use cases each month before authorizing what can go into production.
Dadlani says UnitedHealth monitors AI use cases for safety, bias, fairness, and legal compliance. He stresses that AI is not being used to perform a clinical diagnosis. “We don’t see AI replacing doctors or clinicians,” he says. “We want AI to be a tool.”
To bring the workforce along, UnitedHealth launched an advanced AI learning course in early March that saw more than 10,000 enrollees. The company has also set up a dedicated platform, called United AI Studio, that allows employees to securely access the large language models offered by large AI hyperscalers, as well as the small language models that UnitedHealth developed on its own based on proprietary datasets.
The company also had about half-a-dozen use cases in production based on agentic AI, which is designed to more autonomously complete complex tasks—ideally with limited or no human supervision. Most of the agentic AI use cases have been for repetitive, administrative tasks. The technology will need to mature more before it expands to other parts of the business.
“We will be very cautious when we get into more clinical use cases and the use of agents,” says Dadlani.
John Kell
Send thoughts or suggestions to CIO Intelligence here.
This story was originally featured on Fortune.com
