Artificial intelligence (AI) helps cut down repetitive work, but lower operating costs do not necessarily translate into lower premiums, lower bills, or better access for patients. The biggest savings opportunity is in administration, by eliminating duplicated work, reducing documentation burdens, improving claims workflows, and freeing up clinician and staff time.

 

Why is healthcare so expensive before AI enters the picture

A letter looking at AI systems in healthcare delivery notes that healthcare costs “comprise nearly one-fifth” of US gross domestic product (GDP) and that the last 25 years have been marked by rising administrative costs and lack of labor productivity growth. Much of that overhead has nothing to do with clinical care. According to the U.S. Centers for Medicare & Medicaid Services (CMS), US health spending reached $5.3 trillion in 2024, or 18.0% of GDP.

Much of the problem is administrative overhead. JAMA estimates that administrative costs account for about 15% to 25% of total US health spending, from billing and coding to physician administration and insurance-related costs. The same drag applies to nurses, too. A literature review in Pub Med Central in 2023 found that regulatory and administrative duties account for roughly a quarter of US nurses’ working hours, while documentation, coding, and prior authorization eat up huge amounts of physicians’ time. The American Medical Association estimates that prior authorization alone takes up about 12 hours of physicians’ and staff time each week.

 

How AI could realistically reduce healthcare costs

A Health Affairs Scholar analysis on the use of AI to improve Medicaid administrative processes indicates that administrative spending is often estimated to be 15% to 30% of total healthcare spending and cites estimates that current AI technologies could save $200 billion to $360 billion in healthcare spending within five years by reducing repetitive administrative work.

It is confirmed in narrower settings by early results. In a 2025 NCBI Bookshelf review, 2025 Watch List: Artificial Intelligence in Health Care, AI scribes cut documentation time by 69.5% in laboratory settings and saved primary care providers an estimated three hours a week in an Ontario pilot.

According to Paubox’s research report, Shadow AI is outpacing healthcare email security; 95% of healthcare organizations report staff already using AI tools in email, while 62% have seen employees experimenting with unsanctioned tools such as ChatGPT. The same report found 16% admitted compliance was never consulted before enabling AI email tools, and 75% believe employees assume tools like Microsoft Copilot are automatically HIPAA compliant. AI can remove administrative friction, but savings from unapproved tools, soft oversight, or unclear HIPAA controls can quickly become deferred risk rather than real efficiency, muddying the cost question.

 

Who benefits first from healthcare AI?

Healthcare AI is being built for providers, payers, health tech companies, and startups, which means institutional buyers are positioned as the most immediate customers rather than individual patients.

Adoption data supports this conclusion. An Office of the National Coordinator for Health Information Technology data brief, Hospital Trends in the Use, Evaluation, and Governance of Predictive AI, 2023–2024, found that system-affiliated hospitals used predictive AI at much higher rates than independent hospitals in 2024, 86% versus 37%. Scale buys negotiating power, IT staff, and data volume, all of which make AI cheaper to deploy and easier to justify.

Patients tend to benefit last, and only indirectly, through fewer billing errors or faster claims decisions. Whether any of that shows up as lower premiums or smaller bills is a separate question, one that depends heavily on what a health system or payer decides to do with the money it saves.

 

The difference between good AI savings and bad AI savings

Good savings

Good savings come from removing genuinely wasteful work like re-entering data, chasing down missing documentation, or manually reviewing routine claims. They free up clinician or staff time that is redirected to patient care. They can also survive a full cost-effectiveness review.

The analysis, Systematic review of cost effectiveness and budget impact of artificial intelligence in healthcare, examined the cost-effectiveness, utility, and budget impact of clinical AI interventions, showing why AI’s economic impact needs the same scrutiny as its clinical performance.

 

Bad savings

Bad savings come from denying or delaying care. A study of AI-assisted Medicare Advantage denials and automated review practices describes how AI-assisted coverage denials have come under fire, the subject of class-action lawsuits, congressional investigations, and provider pushback.

They also pass on the cost to patients and providers rather than removing it. Claims that are denied do not just disappear. They become appeals, delayed treatment, and unpaid bills. Poor savings may look good on a payer’s balance sheet while making care harder to access.

 

The payment model decides whether AI lowers costs

In fee-for-service, more volume means more revenue, so AI that speeds up documentation or coding can just as easily support billing more, not less. In an analysis on paying for AI in medicine, they note that as CMS moves to value-based arrangements like MIPS and the Medicare Shared Savings Program, the payment architecture increasingly dictates whether efficiency gains translate into real savings or simply increased output.

 

Red flags that AI is increasing profits more than lowering costs

Processing times are falling while claim denial or prior authorization rates are rising, a pattern that fits the concerns about AI-assisted Medicare Advantage denials and automated review practices. Savings estimates are only from the vendor and no independent data on cost-effectiveness exist to support these. The gap was identified in the Systematic review of cost effectiveness and budget impact of artificial intelligence in healthcare. Paubox’s shadow AI research revealed that staff are already using generative AI tools at scale in patient communications or billing correspondence without a business associate agreement in place.

The assumption that everyday tools are automatically compliant is usually wrong. Compliance depends on the service relationship and required safeguards, including whether a cloud service provider is acting as a business associate and whether appropriate agreements and protections are in place. In other words, HIPAA compliance depends on the signed business associate agreement.

See also: HIPAA Compliant Email: The Definitive Guide (2026 Update)

 

FAQs

What parts of healthcare are most likely to be automated first?

Administrative and operational workflows are the most likely near-term targets.

 

Can healthcare staff use ChatGPT or other public AI tools for patient information?

They should not put protected health information into public or unapproved AI tools unless the organization has formally approved the tool, completed the required risk review, and has the right contractual safeguards in place.

 

Does the FDA regulate healthcare AI?

The FDA regulates some healthcare AI, especially AI-enabled medical devices and software used for medical purposes. But not every AI tool used in healthcare is an FDA-regulated device.

 

Should patients be told when AI is used in their care?

In many cases, yes, especially when AI meaningfully affects clinical decision-making, patient communication, diagnosis, treatment recommendations, or coverage decisions.