Healthcare has reached a critical inflection point. As Artificial Intelligence (AI) permeates every corner of the clinical enterprise, from diagnostic imaging to administrative billing, the industry is witnessing a paradoxical trend: while the technological capability of these tools has expanded exponentially, the realized return on investment (ROI) remains frustratingly elusive.
For many healthcare organizations, the promise of AI—relieving clinician burnout and streamlining care—is being swallowed by the complexity of legacy systems. Richard Atkin, CEO of Greenway Health, argues that the current industry struggle is not a failure of innovation, but a failure of architecture. The sector is attempting to solve 21st-century problems with 1990s foundations, leading to a dangerous "layering trap" that threatens to exacerbate, rather than solve, the administrative burdens facing modern medicine.
The Architecture of Inefficiency: Understanding the "Layering Trap"
The primary hurdle facing health systems today is a conceptual one. Decision-makers often treat AI as a modular add-on, a "plug-and-play" solution designed to sit atop existing Electronic Health Records (EHR) and legacy workflows. The logic is understandable: it is less disruptive to augment a familiar system than to overhaul a foundational one.
However, this approach is fundamentally flawed. When cutting-edge AI is layered onto outdated, inefficient operational workflows, it acts as a digital veneer. It creates a faster, more automated version of a broken process. If a clinical documentation workflow is inherently cumbersome, automating that documentation simply allows the clinician to generate more data, more quickly, without addressing the underlying friction that caused the inefficiency in the first place.
This "bolt-on" mentality shifts the burden. Instead of alleviating pressure, it often migrates it to the end-user—the clinician. Every additional digital layer adds a new interface, a new notification stream, or a new requirement for data verification, ultimately pulling the physician further away from the patient and deeper into the screen.
Chronology of the AI Integration Challenge
To understand how the industry arrived at this crossroads, one must look at the rapid evolution of healthcare IT:
- 1990s–2000s: The Digitization Era: The focus was on moving from paper to electronic records. The priority was capturing data, which led to the creation of rigid, high-input legacy systems.
- 2010s: The Optimization Phase: As EHRs became mandatory, the industry focused on "meaningful use." Workflows were built around compliance and billing, often disregarding the clinical experience.
- 2020–2024: The AI Boom: Generative AI and machine learning tools flooded the market. Healthcare systems, desperate to curb burnout and rising costs, began purchasing point solutions in rapid succession.
- 2025–Present: The Correction: As organizations look at their balance sheets, they are finding that the proliferation of "AI point solutions" has increased IT overhead and administrative complexity, leading to the current call for architectural redesign.
Supporting Data: Why "More" Isn’t Always "Better"
The disconnect between technology investment and operational success is documented in the shifting priorities of healthcare CFOs and CIOs. While traditional metrics such as "number of AI tools deployed" or "software adoption rates" were once considered markers of progress, they are increasingly viewed as vanity metrics.
Evidence suggests that when organizations prioritize the sheer number of AI applications, they encounter the following systemic issues:
- Workflow Fragmentation: When multiple AI tools operate in silos, they often lack interoperability, creating "data islands" that require manual reconciliation.
- Alert Fatigue: The layering of AI-driven diagnostic or administrative prompts often exceeds a clinician’s cognitive capacity, leading to the widespread ignoring of vital automated alerts.
- Technical Debt: Maintaining a "patchwork quilt" of disparate AI tools requires constant, costly updates and integrations, consuming budget that could otherwise be spent on patient-facing innovation.
A shift in perspective is required. Organizations that view success through the lens of "reclaimed clinical capacity"—the actual time returned to a physician to spend with a patient—consistently outperform those that measure success by the volume of software licenses purchased.
Expert Perspective: The Greenway Health Approach
Richard Atkin, whose career spans from defense industry engineering to leading healthcare technology firms, emphasizes that the path forward requires a fundamental shift in mindset. In his view, healthcare IT should not be about adding features, but about removing friction.
"In medicine, treating a symptom while ignoring the underlying cause is considered poor practice," Atkin notes. "Yet, healthcare organizations are increasingly applying software band-aids to cover workflow inefficiencies."
Atkin’s philosophy, shaped by his leadership at Greenway Health, advocates for a customer-driven approach where the product is subservient to the clinical workflow. He argues that leaders must perform a "clinical diagnostic" on their own IT infrastructure. This involves asking difficult questions:
- Does this tool eliminate a manual step, or does it add a new login requirement?
- Does the data generated by this AI flow into the decision-making process, or does it sit in a secondary dashboard?
- Are we measuring success by how much technology we own, or by how much administrative work we have eliminated?
Implications for the Future of Care Delivery
The long-term implications for the healthcare industry are profound. If organizations continue down the path of "layering," the result will likely be a continued decline in provider satisfaction and a ballooning of administrative costs that ultimately threatens the viability of independent and smaller clinical practices.
However, the "pivot" offers a path to sustainability. The organizations that thrive in the next decade will be those that prioritize "re-engineering" over "adding."
1. From Adoption to Integration
Future procurement processes must shift. Before a tool is purchased, organizations must conduct a "Workflow Impact Analysis." If a tool cannot be integrated into the existing flow without creating new tasks, it should be rejected, regardless of its technological sophistication.
2. The Rise of "Negative Metrics"
We will likely see the rise of "negative metrics" in healthcare leadership. Leaders will be incentivized by the number of steps removed from a process, the number of screens a clinician no longer has to click through, and the reduction in "pajama time"—the hours physicians spend at home finishing documentation.
3. Redesigning for the Human-AI Interface
The future of healthcare AI is not in replacing the clinician with a bot, but in creating a seamless interface where the AI operates in the background. This requires an architecture that is human-centric, where the AI acts as a silent assistant that handles the data burden, leaving the clinical reasoning to the human provider.
Conclusion: Rebuild, Don’t Bolt On
The "inflection point" mentioned by industry leaders is not just a catchphrase; it is a warning. We are at a moment where the sheer volume of digital capability is threatening to collapse under its own weight.
To realize the full promise of artificial intelligence, healthcare must stop treating technology as an accessory. It must be treated as a foundational element of a redesigned, patient-centered care delivery model. This will be difficult work. It requires a willingness to decommission legacy systems that are no longer serving the patient, a discipline to say "no" to point solutions that don’t fit the core architecture, and the courage to rethink workflows from the ground up.
The future of healthcare does not belong to the organizations with the most technology. It belongs to those who have the vision to use that technology to simplify, eliminate, and empower. It is time to stop bolting on and start building for the future of medicine.
