- Sruthi Madhavan, Vice President, Strategy & Transformation
AI has stormed into healthcare headlines. Adoption among U.S. physicians nearly doubled in the last year, with two-thirds now reporting they use AI in some form. There is a popular myth that is emerging: that intelligent automation alone can resolve fragmented care, inefficiency, and clinician burnout. Yet the outcomes remain flat. Despite billions invested and thousands of pilots initiated, fundamental gaps in healthcare delivery continue to exist.
Why? Because the workflows remain unchanged to keep up with the innovation.
A recent MIT study found that 95% of generative AI initiatives fail to deliver measurable outcomes. This is because they are bolted on existing bad processes without really designing for adaptability, accountability, or trust. In short: AI is not failing. The system is.
This raises a hard but necessary question: What exactly will AI not fix in healthcare?
AI Cannot Fix Broken Operating Models
Even the most advanced agentic AI systems cannot fix a fundamentally flawed operating model. Layering automation on top of siloed care delivery, unclear accountability, and linear processes that do not reflect the complex needs of patients, will only accelerate dysfunction.
Yes – AI can route tasks, analyze risks, or suggest next steps. But it cannot redefine how organizations make decisions, coordinate across functions, or hold teams accountable for outcomes. For example, an AI agent may identify high-risk patients for outreach, but if care teams remain siloed with unclear escalation paths, those insights will rarely translate into action.
Before AI: Healthcare organizations must redesign workflows with end-to-end value streams in mind, applying “AI-first” thinking to how work should flow. Only then can automation drive measurable efficiency and cost savings.
AI Will Not Enforce Governance Discipline
As AI becomes a critical component in real-time decision-making across healthcare functions like utilization management, prior authorization, or member outreach, the central question shifts from “Does the AI model work?” to “Who owns its behavior and decisions?”
It is not enough to deploy agentic AI workflows that optimize processes. Organizations must also define who approves AI decision rights, who overrides outputs, who interprets AI-driven decisions for stakeholders, and who is accountable when the system fails.
Healthcare cannot outsource accountability to algorithms. Without clearly defined governance, AI enablement risks eroding patient trust and exposing organizations to compliance challenges.
Before AI: Rearchitect governance models to include shared accountability between humans and AI. Define rules for what AI can and cannot do, escalation paths and override protocols, and the post-deployment learning loops. This will be the trust layer for intelligent systems.
AI Cannot Earn Trust
Healthcare decisions are life-altering and personal. When a claim is denied, a referral is delayed, or a treatment gets deprioritized, members and providers expect to understand who made the call, and why.
An AI agent might flag something as low priority. But it cannot explain that decision to a member or physician. It cannot hold responsibility when trust is broken. Evidence shows that when there is a lack of transparency, provider adoption and patient confidence fall sharply.
Before AI: Any AI-driven denial or clinical decision must be validated by a human. Workflows should prioritize explainability, document override rationale, and make accountability clear. A recent study by Yu et al. (Mayo Clinic and Zyter|TruCare) showed that trust only improved when AI confidence was calibrated and explanations were transparent, cutting override rates from 87% to 33%.
AI Cannot Replace Strategic Continuity
Even the most innovative large-scale transformation initiatives often fail because they lacked the follow-through. AI cannot champion the long arc of change management.
New workflows, roles, and behaviors do not embed themselves. If no one owns the continuity between strategy, implementation, and adoption, momentum gets lost and so does value.
Before AI: Assign transformation owners who span from vision to execution. Align technology rollouts with change management that supports teams, reinforces new behaviors, upskills employees, and keeps the focus on outcomes beyond go-live.
AI Cannot Overcome Structural Bias
AI will only ever be as fair as the data and rules it is built on. If risk scoring models are trained on incomplete claims data, they may consistently under-identify vulnerable populations. If prior authorization agents rely on narrow clinical guidelines, they may systematically disadvantage patients outside those baselines. Bias does not disappear with automation; it scales.
Before AI: Embed equity reviews into model design and governance. Validate performance across subgroups and require transparent reporting that makes disparities visible and correctable.
Redesign First, Automate Second
AI is not the transformation. Redesigning the work is.
The real value lies in rethinking how the decisions are made, how value flows, and how teams operate before any automation is leveraged. So instead of starting with the question, “Where can we use AI?”, ask: “What should this work even look like in the first place?”
At Zyter|TruCare, that is exactly where we start. Our RECODE™ methodology provides the blueprint for redesign, ensuring governance, accountability, and trust are built in from the beginning. By embedding AI into reimagined workflows through RECODE, we help payers and care networks reduce inefficiency, restore clinician focus, and deliver sustained measurable outcomes.
👉 Ready to explore what AI can (and cannot) fix in your organization? Connect with the Zyter|TruCare team today to see how we can help you redesign care for the future.
