Why AI may be making your administrative burden worse

The promise of artificial intelligence (AI) in the modern clinical setting was supposed to be a reprieve from the "death by a thousand clicks.” For the average physician, the prospect of an ambient scribe capturing every nuance of a patient encounter or an automated system handling the Byzantine maze of prior authorization felt like a long-overdue technological rescue. However, a series of new industry insights suggests that instead of thinning the administrative thicket, AI may be accelerating a "bot war" that threatens to inflate costs and trigger blunt reimbursement cuts across the board.

According to a report from the Peterson Health Technology Institute (PHTI), the U.S. health care system loses an estimated $350 billion annually to administrative waste. Of that staggering sum, $266 billion is attributed to administrative complexity alone. While health systems and payers are rushing to deploy AI tools to bridge this gap, the early results indicate that the technology is often being layered on top of fundamentally broken processes, leading to what experts call an efficiency paradox.

The rise of the bot wars

For many doctors, the most frustrating barrier to care is the prior authorization process. It is a system characterized by a lack of transparency; of more than 5,000 procedure codes requiring prior authorization across four major U.S. insurers, only 3% require it across all four. This fragmentation creates a friction-filled environment in which providers and payers increasingly use AI to fight one another in automated loops.

Caroline Pearson, executive director, PHTI, describes this phenomenon as a digital escalation. “The frustration physicians feel about prior authorization is well documented, and AI is not yet solving it,” she says. “When providers deploy AI to automate submissions and health plans deploy AI to triage and evaluate those requests, the result can be an escalation of back-and-forth with more rounds of appeals and denials, rather than fewer. Workshop participants described ‘bot wars’ — automated exchanges that multiply the volume of communications per prior authorization without resolving the underlying clinical or administrative questions.”

This back-and-forth is not just a nuisance; it is a financial drain. Each prior authorization submission cycle is estimated to cost providers between $20 and $30, while health plans spend $40 to $50 per cycle. When AI is used to simply automate the existing friction rather than remove it, these costs compound without improving patient outcomes.

“Furthermore, plans impose different, unpublished requirements for how providers can meet these requirements and what documentation is needed,” says Pearson. “This means AI is being applied on top of a fragmented, inconsistent process rather than a standardized one.”

The ambient scribe dilemma

While prior authorization remains a battlefield, many physicians have found immediate relief in ambient AI scribes. These tools, which automate the translation of patient visits into medical claims, are now nearly standard in large health systems. They offer modest time savings — roughly two hours per physician annually — but their primary impact has been on billing accuracy and intensity.

By capturing a more complete picture of patient complexity, AI documentation has led to a sharp increase in higher-complexity coding. One multihospital system found that after deploying an AI scribe, level 5 encounters for established patients increased by 5%, while level 4 encounters rose by 7%. This translated to an average revenue increase of $1,004 per provider per month.

However, this newfound efficiency in charge capture is already triggering a defensive response from payers. Major insurers are citing “aggressive” provider coding as a driver of higher medical expenditures in 2025 and 2026, leading to across-the-board “downcoding” strategies.

Pearson warns that the widespread adoption of these tools could lead to a volatile cycle of inflation and reimbursement cuts. “Adoption is already widespread and growing. The current evidence suggests that the majority of health systems and providers will have these tools in place in the future, and the industry should expect a sharp acceleration in higher-complexity coding,” she says. “The question of what comes next is important. In the near term, payers are responding with across-the-board downcoding, and some states — including Missouri and Indiana — have responded by introducing bills to restrict AI-enabled downcoding by health plans. Unfortunately, blunt downcoding policies may disproportionately harm certain providers and their patients, including those practicing in rural areas or community health centers, that have been slower to adopt scribe technology.”

However, the impact on physician well-being has been among the most frequently cited benefits of the technology. “There is literally no other intervention in our field that impacts burnout to this extent," Rebecca Mishuris, M.D., M.P.H., told Medical Economics. At Mass General Brigham, where Mishuris serves as chief medical information officer, the rate of burnout fell from 52.6% to 30.7% over an 84-day period following the introduction of the tools.

Other experts agree that the primary value of these tools is the gift of time. “If you’re working five days a week, [20 minutes per day] is giving you back a couple hours of your life, which you can spend in any way you want,” Deepti Pandita, M.D., FACP, FAMIA, told Medical Economics. Pandita, who serves as vice president of clinical informatics and chief medical information officer at the University of California, Irvine, presented data showing that AI-assisted documentation saves physicians anywhere from 12 to 20 minutes per day

The widening digital divide

This “blunt” response from payers creates a significant equity concern. Physicians who have not yet adopted AI documentation tools — often those in smaller, independent or rural practices — face a double penalty. They do not benefit from the increased revenue capture of AI-assisted coding, yet they are subjected to the same reimbursement reductions designed to curb AI-driven inflation.

Data from the Office of the National Coordinator for Health Information Technology confirms that the providers lagging in AI adoption are the ones operating on the thinnest margins. “This is a significant concern, and it does not yet have a clean policy answer,” Pearson says.

Without targeted policy safeguards, the “efficiency” of AI in large systems could inadvertently bankrupt the providers who are already the most vulnerable in the American health care landscape.

Toward real-time adjudication

The solution to the efficiency paradox may lie in moving away from the “submit and wait” model toward real-time adjudication at the point of care. Emerging models, such as the collaboration between Optum Rx and Cleveland Clinic, have shown promising results, including an 88% reduction in appeals and a 68% reduction in denials caused by missing information.

In these systems, AI doesn’t just draft a request; it flags documentation gaps or standard-of-care discrepancies while the patient is still in the room. If a provider requests 18 physical therapy sessions when the guideline supports 12, the system can immediately approve 12 and outline the specific requirements needed for the additional six.

However, the leap from small-scale pilots to a national standard is massive. Pearson highlights the remaining technical and structural hurdles. “The early results from real-time adjudication pilots are encouraging, but the gap between these proofs of concept and a production-ready, scalable solution is substantial,” she says. “Real-time adjudication requires reconciling fragmented medical policies across health plans, navigating proprietary medical necessity criteria and building infrastructure for real-time bidirectional data exchange.”

Federal policy is beginning to catch up. The CMS Interoperability and Prior Authorization Rule (CMS-0057-F) will require many plans to implement standardized application programming interfaces by 2027, enabling more electronic, real-time communication. While this helps with straightforward requests, it does not yet address the variation in medical necessity criteria that drives the most contentious cases.

A need for structural redesign

True reform may require moving beyond prior authorization entirely, exploring alternatives like postpayment review or allowing providers to bypass up-front reviews in exchange for discounted reimbursement rates.

Pearson says that current incentives are simply not aligned for systemwide savings. “A central finding is that under current incentive structures, no stakeholder is positioned to deploy AI in ways that actively reduce friction across the system — efficiency gains tend to accrue within organizations rather than flowing through to lower costs systemwide," she says.

For the physician on the front lines, the message is clear: AI is a powerful tool, but it is not a panacea for the administrative burden. As the bot wars escalate, the focus must shift from how fast we can automate the current bureaucracy to how we can fundamentally redesign the way health care is billed and authorized.

The trajectory of AI in administration is currently set toward higher costs and more frequent denials. Reversing that course will require more than just better software; it will require a coordinated policy response that establishes guardrails for coding, ensures equity for rural practices and pushes for the standardization of medical policies.

Until the underlying workflows are simplified, the efficiency paradox will continue to haunt the modern clinic.

“AI applied on top of a flawed process makes the flaws more visible and more costly. It does not fix them,” Pearson says.