Administrative complexity is the single most expensive category of healthcare waste in the United States. The landmark 2019 Journal of the American Medical Association (JAMA) study led by Dr. William Shrank attributed $265.6 billion in annual waste to this category alone. 1 Adjusted for 2025 national health expenditure data from the Centers for Medicare and Medicaid Services (CMS), that figure now projects to approximately $352 billion per year. 2
That $352 billion does not fund patient care. It funds paperwork. It funds prior authorization requests, denial management cycles, billing disputes, duplicative data entry, and the administrative overhead that every provider, payer, and employer in the system absorbs. The U.S. spends over $1,000 per capita on healthcare administration, roughly five times what comparable countries spend. 3
For self-insured employers providing health benefits to more than 160 million Americans, administrative waste is embedded in the entire claims process, every vendor invoice, and every renewal negotiation. 4 It is measurable. It is growing. And artificial intelligence is now giving plan sponsors the tools to address it at scale.
This article examines how administrative waste accumulates, why traditional approaches have failed to control it, and where AI is producing measurable results today.
The $352 Billion Tax That Treats No One
No other developed nation generates administrative costs at the scale of the U.S. healthcare system. The multi-payer structure, with its overlapping coverage rules, competing billing requirements, and fragmented data standards, creates bureaucratic overhead that single-payer systems simply do not experience. Canada, for example, spends roughly $550 per capita on healthcare administration. The U.S. spends nearly double that figure.
Much of this burden falls directly on physicians. The American Medical Association’s 2023 Prior Authorization Survey found that physician practices complete an average of 39 prior authorization requests per physician per week, spending 13 hours of staff time on the process. 5 Ninety-three percent of surveyed physicians reported that prior authorization delays necessary patient care. Nearly 80% said it leads patients to abandon recommended treatments.
These delays carry real clinical and financial consequences. A patient who abandons a recommended treatment today often returns with a more advanced condition later. The resulting emergency visit, hospitalization, or surgical intervention costs the health plan multiples of what the original treatment would have required.
On the payer side, the costs are equally staggering. A recent MedCity News report documented that payers and providers collectively spend $25.7 billion annually on denial management alone. 6 That figure represents the cost of disputing, appealing, and reprocessing claims that the system itself generates. It does not include the underlying clinical or financial harm those delays produce.
The AI Arms Race Between Payers and Providers
Payers and providers are both deploying AI, but often in opposing directions. A Healthcare Dive investigation in 2024 described an emerging “AI arms race” in which payers use machine learning to identify and deny claims faster, while providers deploy AI to predict denials, automate appeals, and accelerate claims resubmissions. 7 Both sides are investing in technology designed to win billing disputes rather than eliminate them.
Bill Fera, Principal Consultant at Deloitte, described the shift in concrete terms: “We’re taking the mystery away. And there is a fact base. There’s a core piece of information that can be interrogated. It’s just now, it can be interrogated very quickly.”7 The speed advantage is real. But when both sides use AI to fight faster rather than to reduce unnecessary disputes, the net effect on administrative cost can be neutral or even negative.
This dynamic creates a structural problem for self-insured employers. When payers automate denials and providers automate appeals, employers fund both sides of the dispute cycle. Every denied claim that is later overturned represents administrative cost that the plan absorbs twice: once in the denial processing and again in the appeal resolution.
The 70% appeal overturn rate documented in recent industry research reinforces the point. 6 Seven out of ten denied claims are eventually paid. This means a majority of initial denials lack sufficient clinical justification. The administrative cost of cycling those claims through denial and appeal adds expense without adding value.
Beyond Front-Office Automation
Recent coverage of AI in healthcare has focused heavily on front-office automation: scheduling, appointments and reminders, patient intake, and documentation support. A March 2026 MedCity News article, for example, argued that AI-driven administrative automation represents “one of the most realistic paths to reducing healthcare costs without sacrificing quality.” 8 The Commonwealth Fund has estimated that 15% of excess healthcare spending comes from operational inefficiency in areas like billing, scheduling, and documentation. 9
These improvements matter. Reducing no-shows, streamlining patient intake, and automating routine documentation all contribute to lower operational costs at the facility or practice level. Studies published in JAMA have shown that AI-powered ambient scribes reduced physician burnout from 51.9% to 38.8% and cut documentation time by 30 minutes per day.
But the larger opportunity sits deeper in the system. The $352 billion in annual administrative waste is concentrated in claims processing, payment disputes, coding errors, prior authorization bottlenecks, and the adversarial billing dynamics that front-office automation does not touch. Automating the waiting room is useful. Automating the detection and prevention of claims-level waste is transformative.
Where AI Is Producing Measurable Results
The most financially significant application of AI in administrative waste reduction is claims auditing. AI-powered platforms can review every line item on every claim, compare billed amounts against set benchmarks and Medicare reimbursement rates, and flag duplicate charges, unbundled services, upcoded procedures, and charges with no corresponding clinical record. This analysis can happen in near real time, covering 100% of claims rather than the 5% to 10% that traditional manual audits tend to cover.
The American Medical Association has documented a 20% claims-processing error rate among commercial health insurers, representing an estimated $17 billion in annual waste. 10 Industry analyses suggest that up to 80% of hospital bills contain at least one error. When AI reviews every claim rather than a statistical sample, it uncovers patterns that periodic auditing would never identify.
Natural language processing is reducing one of the most time-intensive administrative tasks in healthcare: prior authorization. AI systems can process routine authorization requests automatically, apply pre-set clinical criteria, and flag only complex cases for human review. For high-performing providers, “gold-carding” programs that waive prior authorization requirements can eliminate hundreds of thousands to millions of hours of unnecessary administrative work across the system.
Pharmacy analytics represent another area where AI is targeting administrative overhead. AI systems can identify members on branded medications where clinically equivalent generics exist, flag therapeutic alternatives that cost a fraction of brand-name drugs, and detect formulary patterns that reflect pharmacy benefit manager (PBM) incentive misalignment rather than clinical best practice. Organizations moving to transparent, pass-through pharmacy pricing models are reporting savings of 15% to 30% on pharmacy spend. 11
The Self-Insured Employer’s Unique Position
Self-insured employers bear a disproportionate share of administrative waste because they directly fund their health plans. Unlike fully insured arrangements where premiums are set by the carrier, self-insured organizations pay claims as they are incurred. Every billing error, every unnecessary denial-and-appeal cycle, and every inflated administrative charge comes out of the plan’s operating budget.
Under the Employee Retirement Income Security Act (ERISA) and the Consolidated Appropriations Act (CAA), plan sponsors have fiduciary obligations to ensure that plan assets are used prudently and in the exclusive interest of participants. 12 A National Alliance of Healthcare Purchaser Coalitions survey found that 65% of employers expressed growing concern about potential litigation related to fiduciary duties. 13 Administrative waste that goes undetected does not stay invisible. It accumulates in the plan’s financial performance and, increasingly, in its legal exposure.
Christine Akers of SmartLight Analytics captured the challenge precisely: “The buck stops with the plan sponsor when it comes to fiduciary responsibilities involving health plan administration and operations.” 14 Employers cannot outsource this responsibility to third-party administrators and assume compliance. They must actively measure, audit, and manage administrative spending.
A 14% payment inaccuracy rate on a $50 million health plan equates to $7 million in annual losses. That money could fund better benefits, lower employee premiums, or a variety of wellness programs. Every dollar absorbed by administrative waste is a dollar that is otherwise unavailable for employee care.
From Detection to Prevention
The first wave of AI in healthcare cost management has focused largely on detection: finding errors, flagging anomalies, and identifying waste after it occurred. The next wave should shift toward prevention. AI systems trained on multi-year claims datasets can identify emerging cost patterns, predict high-risk member trajectories, and flag administrative bottlenecks before they generate downstream waste.
Former CMS Administrator Dr. Donald Berwick estimated that as much as $800 billion in recoverable waste sits untapped across the U.S. healthcare system. 15 A significant portion of that waste originates in administrative processes that AI is uniquely equipped to address: repetitive data entry, redundant clinical reviews, payment disputes caused by coding errors, and prior authorization requests that could be automated with appropriate clinical decision support.
The most effective deployments should pair AI detection with human decision-making. Claims audit results would lead to contract renegotiations. Predictive risk flags would activate care management outreach. Pharmacy optimization analytics would drive formulary redesign. The technology presents its findings, people act on them, and do so much quicker than before.
What Plan Sponsors Should Do Now
Employers do not need to wait for the healthcare delivery system to solve its own administrative problems. Several concrete steps are available today.
First, demand 100% claims auditing. Periodic manual audits that sample 5% to 10% of claims are insufficient for plans processing hundreds of thousands of transactions annually. AI-powered systems that review every claim in near real time are commercially available and increasingly affordable to build.
Second, evaluate prior authorization processes. Sixty-nine percent of employers surveyed by the Business Group on Health are considering changes to their prior authorization processes. For plans with high denial rates, AI pre-submission validation can catch errors before they trigger the denial-and-appeal cycle.
Third, benchmark vendor performance with data. Measure third-party administrator accuracy rates, PBM pricing against independent benchmarks, and provider billing patterns against regional and specialty norms. AI analytics can automate these comparisons and flag outliers for review.
Fourth, establish fiduciary governance structures. Appoint a fiduciary committee responsible for reviewing plan performance metrics, vendor relationships, and claims accuracy data. Document decisions and the data supporting them.
Administrative Complexity Is Quantifiable. The Response Should Be Too.
Administrative waste in U.S. healthcare is not a hidden problem. It is documented in peer-reviewed research, measured by industry surveys, and visible in the claims data of every self-insured health plan in the country. The JAMA research confirmed that proven interventions could save $191 billion to $282 billion annually across the system. 1 For individual plan sponsors, the savings potential is proportionally significant.
AI does not eliminate the structural incentives that produce administrative waste. What it does, however, is make waste visible, measurable, and actionable at a speed and scale that manual processes simply cannot match. The technology is ready. The fiduciary mandate is clear. The question is which organizations will act on the data and which will continue absorbing costs they could prevent.
