Over the past decade, I’ve watched healthcare providers invest in electronic health records (EHRs), revenue cycle management (RCM), and a broad range of health IT solution areas spanning clinical, operational, and administrative functions. Yet, one issue continues to drain resources and morale like no other: prior authorization, also known as “prior auth.” Despite being intended to ensure appropriate care and control costs, in my analysis, prior auth costs the U.S. healthcare system at least $41.4 billion to $55.8 billion annually, at least, depending on how you model and factor in labor costs, delays, and the downstream clinical impact. What is even more bothersome is that prior auth isn’t just an operational inefficiency but a symptom of a deeper failure to prioritize system redesign over entrenched inefficiencies, temporary workarounds, and conflicting incentives.
Why Prior Auth Won’t Fix Itself
What I’ve come to believe, contrary to prevailing narratives, is that the prior auth crisis is not a failure of process or technology, but people and mindset. For years, U.S. healthcare leaders, particularly within provider organizations, have largely abdicated meaningful engagement with system redesign or resistance to imposing external forces. Instead, they’ve defaulted to compliance despite any dysfunction, relying on short-term patches, manual workarounds, and narrowly scoped initiatives to ensure payment. Most efforts have been reactive, designed to navigate and endure the complexity than challenge it.
Survey data from the American Medical Association (AMA) paints a stark picture. Physicians and their teams spend 13 hours a week handling an average of 39 prior authorizations per doctor, so burdensome that 40% of practices have staff solely dedicated to it. Nearly 9 in 10 physicians report it drives burnout and inflates healthcare utilization. Even more troubling, 94% say prior auth harms clinical outcomes, 93% say it causes care delays, 82% report it leads to treatment abandonment, and 29% cite serious adverse events as a direct result.[1]
Underneath any cultural resignation is the sense that such administrative complexity is “simply healthcare,” and the sheer magnitude of it is “how U.S. healthcare works.” This has shaped decades of efforts and investments that have further baked such dysfunction into the very DNA of the system. By not challenging and, in turn, reengineering prior auth from the ground up, it was standardized. The problem has not only hardened but has also been operationalized and institutionalized, resulting in such friction and colossal costs.
The problem with prior auth isn’t that it’s only expensive to do, but also that it’s resistant to change. Unlike RCM, which has evolved over the years toward better end-to-end, front-loaded models that begin well before the claim is filed or the patient is even seen, prior auth too often gets triggered late in the care episode, after key decisions have been made. It’s still mostly payer-facing and payer-driven in the continuum of care, with misaligned, frequently conflicting incentives and inconsistent criteria across the board, except in the case of “payviders” or integrated delivery networks, where the divide is less pronounced and may promote rather than block collaboration despite it needing to be more in the patient’s or member’s interest than the system for true value-based care delivery.
U.S. healthcare providers have repeatedly mistaken digitization for modernization. Converting paper into PDFs instead of structured data, automating outdated steps, or adding a portal to a broken process. These were not transformative moves. At best, they converted paper files into EHRs or manual billing into RCM, without questioning the process design behind them. Does it drive approval any faster? Does it reduce burden or improve care quality and experiences? Rarely. Even traditional automation tools, like RPA, while helpful for repetitive administrative tasks, were never built to handle unstructured data or the dynamic, exception-heavy nature of prior authorization workflows. These tools, in essence, served more as digital band-aids over deep systemic wounds, not solutions.
Market Signals Tell Us to Move On
According to IDC survey data, 52.5% of U.S. healthcare providers are now adopting composable IT architectures to drive electronic prior auth (ePA), moving toward modular, plug-and-play systems designed for agility and continuous evolution. Meanwhile, only 6.6% remain dependent on rigid, custom-built platforms. The message is clear: the market is shifting toward flexibility, interoperability, and intelligent orchestration.
I won’t go so far as to say the tide is turning, but signals are getting louder. Across the board, I’m seeing more healthcare leaders on both the technology buyer and supplier sides acknowledging that traditional automation has reached its limits. The complexity of an area like prior auth demands something more adaptive, scalable, and intelligent.
Enter agentic AI, not just as another layer of automation, but a new class of automation. Where agents shine is that they can bridge the gap between automation with intelligence, autonomy, and context awareness, working not just faster, but smarter. As opposed to traditional rule-based systems or narrowly trained models, agentic AI can adapt, interpret, and learn on the fly. This is a significant leap from simply executing pre-coded functions.
What sets agents further apart is their ability to perform zero-shot reasoning, as well as their capability to handle new inputs or scenarios that haven’t been trained on by leveraging generalized knowledge across domains. This adaptability reassures healthcare leaders that agentic AI can function even in the face of edge cases, real-time policy updates, and unstructured clinical complexity, making it particularly well-suited for prior auth, where variability is the norm, not the exception.
Rather than following static rulesets or requiring periodic retraining, agentic AI can:
- Interpret unstructured data by leveraging NLP and LLMs to extract relevant information from free-text sources such as physician notes, discharge summaries, radiology reports, and lab results. This enables the system to understand the clinical rationale behind a treatment or diagnostic order, allowing for more accurate and context-aware authorization decisions without requiring structured, template-based documentation.
- Adapt dynamically to evolving payer rules, rather than relying on static rule engines or periodic manual updates. Agentic AI can ingest real-time payer policy feeds, API-accessible rule libraries, or even scrape payer portals (when necessary) to automatically apply the most current criteria. This eliminates the lag between policy changes and system response, reducing unnecessary denials caused by outdated logic and helping ensure compliance is maintained proactively.
- Execute complex, multi-system workflows autonomously by orchestrating interactions across tech stacks and layers, be it EHRs, eligibility verification systems, third-party prior auth platforms, and payer endpoints. It can initiate requests, validate documentation, follow up on pending statuses, and escalate exceptions without manual handoffs. This end-to-end orchestration eliminates redundant clicks, fragmented touchpoints, and disconnected workflows that slow the process.
- Continuously learn and optimize performance with built-in feedback loops. Agentic AI can analyze the outcomes of approvals, denials, resubmissions, and appeals, and use that data to fine-tune logic over time. This continuous optimization instills confidence in healthcare leaders that the loops can enhance prior auth quality, increase first-pass rates, and minimize administrative rework, leading to improved financial and operational outcomes.
The silver lining is that this isn’t a vision but is already happening.
What This Means for U.S. Healthcare Providers
For CIOs:
- A scalable, modular approach to intelligent automation that aligns with existing IT investments.
- Rather than costly rip-and-replace initiatives, agentic solutions can integrate more seamlessly (e.g., into EHRs, practice management, and RCM systems) via APIs, FHIR interfaces, and event-driven architectures.
- Agents can be further embedded within existing workflows or operate as orchestration layers on top of legacy infrastructure.
- Composability, interoperability, and accelerated ROI that support modernization without disruption, delivering improved speed, flexibility, and clinical alignment.
For CMIOs/CNIOs:
- Clinically intelligent automation that complements rather than complicates workflows.
- Agents can interpret free-text notes, align with evidence-based care protocols, and apply payer-specific criteria without forcing changes in behavior, helping to reduce ‘death by a thousand clicks’ and ‘unlimited mouse miles’ while still supporting contextual decision-making, improving accuracy and timeliness of authorization workflows without burden.
- Preservation of clinician experience while improving patient experiences and outcomes.
For RCM Leaders:
- Immediate and measurable value by dynamically aligning clinical submissions with payer policies in real time, shortening authorization turnaround times through intelligent workflow automation, and improving clean claim rates by ensuring complete and compliant documentation at the point of capture.
- Real-time visibility into authorization status, exception handling, and appeal triggers, empowering billing teams to work smarter, not harder, and to optimize reimbursement without unnecessary overhead.
Beyond these roles, the greater opportunity lies in agentic AI laying the groundwork for intelligent automation across the board, thereby elevating healthcare provider workflows and operations to be more adaptive, resilient, and scalable altogether.
A Final Thought: Don’t Automate Dysfunction
I’ll close with this: stop framing prior auth as solely a technical or workflow issue when it’s not. It’s more of a systemic and cultural issue and distinctly related to U.S. healthcare, reflecting how the experience has been deprioritized in favor of bureaucracy. If AI gets layered on top of that without redesigning the underlying processes, then it will just be scaling dysfunction and be largely counterproductive. This is not to say agentic AI is a silver bullet, no, but it offers a way forward, one that can help automate and distribute intelligence rather than dysfunction. The question isn’t whether we should adopt it, but how quickly, responsibly, and effectively we can do so. If prior auth remains a sinkhole for U.S. healthcare in five to ten years, it won’t be due to a lack of innovation or tools, but rather a lack of leadership, imagination, and willpower.
If you are a client or subscribe to our research, access the full report here: From Administrative Drain to Clinical Gain: The Case for Agentic AI in Prior Authorization for Healthcare Providers. To become a client or learn more about our research, please visit idc.com.
[1] Prior Authorization (PA) Physician Survey 2024 | AMA. Available at: https://www.ama-assn.org/system/files/prior-authorization-survey.pdf (Accessed: 17 July 2025).
