The following is a guest article by Gigi Yuen, PhD, Chief Data & AI Officer at Cohere
Prior authorization is entering a moment of major transition. New federal requirements and AHIP-backed commitments taking effect in 2026 are pushing health plans to modernize legacy workflows that have long frustrated providers and delayed patient care. In response, many organizations are exploring AI-driven automation to meet the new standards and relieve administrative burden.
The challenge, however, is that not all AI is built for healthcare. Yes, some of the general-purpose tools can speed up routine tasks, but they are often built without the clinical nuance needed to make accurate decisions. To build a system that is truly faster, safer, and more consistent, health plans need technology that understands medicine like a clinician and that supports transparent, evidence-based decision-making.
For decades, prior authorization has relied on faxes, phone calls, and duplicative paperwork–driving provider burden and patient care delays. Physicians consistently link prior authorization to treatment abandonment, and many practices employ staff whose sole job is managing authorization requests. With new Centers for Medicare & Medicaid Services (CMS) requirements and industry commitments taking effect in January 2026, the status quo is no longer sustainable for staying compliant and competitive.
The CMS Prior Authorization Final Rule raises the bar, requiring faster decisions and greater transparency and interoperability, with CMS estimating the reforms could save providers $16 billion over the next decade. Meeting these standards will take more than digitizing legacy workflows; health plans need transformative technology that understands clinical context.
Imagine a process where clinically trained AI agents can gather physician notes and claims history, evaluate requests against evidence-based guidelines, and return real-time approvals for appropriate care, giving submitting providers quicker answers while enabling clinicians to focus on complex cases that truly require human judgment. With advances in clinical-grade AI built with physician oversight and transparent governance, this vision is now achievable. But before investing in new technology, health plans should consider four reasons why this approach is the most effective path to modernizing prior authorization and meeting compliance.
1. Superficial Automation Falls Short
Many organizations are adopting automation ahead of 2026, yet digitizing existing workflows rarely solves the underlying problems. General-purpose large language models and agents can speed up routine tasks more efficiently, but it doesn’t resolve the deeper issues of siloed data, varying health plan policies, and a lack of clinical context.
Generic, task-based AI can further inefficiency and introduce new risks, including the faster processing of incomplete information, inconsistent policy interpretation, and automation errors that amplify provider abrasion.
That is why attention is shifting to clinical-led agentic AI that is precision-trained for specific tasks, using large-scale, comprehensive clinical datasets and designed to act rather than merely suggest. These clinically intelligent systems can:
- Find and interpret a patient’s longitudinal health history from unstructured and structured data
- Apply clinical criteria with proven consistency
- Provide real-time, full visibility into AI decision-making
- Ensure effectiveness, responsibility, and trust
Built with responsible governance and continuous clinical oversight at their core, they show every decision path clearly, allowing reviewers and regulators to understand how conclusions were reached. This level of visibility prevents new errors from being introduced and ensures automation strengthens, rather than destabilizes, clinical workflows.
2. Building Trust and Safety Into the Workflow
Skepticism about AI in healthcare is understandable. Clinicians worry that technology could be used to deny care, replace their expert judgment, or drive inappropriate utilization. Trust in AI is not assumed; it must be earned through deliberate design and transparency. Clinical-grade AI supports this by:
- Incorporating clinician insights in development, not just in validation of models
- Maintaining fully traceable audit trails
- Applying calibrated risk tiers that match clinical complexity
- Building decision pathways that reflect how clinicians think and act, and reflecting the latest evidence and policies
- Routing any requests that cannot be automatically determined to the same-specialty clinicians for review
Several states have explored legislation on AI use and guardrails in utilization management, while a recent federal executive order calls for the creation of a national AI framework. Even in the absence of legislation, embedding trust by design–through transparency, accountability, and risk calibration–directly into clinical workflows supports health plans with near-term compliance and provides a flexible foundation for the evolving regulatory landscape.
Clinical-grade AI also strengthens patient safety by identifying clinical risks and contraindications that may be overlooked in traditional workflows. Models built and continually validated on evidence-based best practices, coverage policy, and clinical documentation can surface the most relevant clinical factors for a determination, so that patients receive appropriate care as quickly as possible. By flagging concerns up front and routing complex cases to expert clinicians, clinically led AI reduces the likelihood of inappropriate or delayed treatment. Ultimately, this combination of precision and oversight supports better patient outcomes and reinforces trust among providers.
3. Reducing Burnout Through Intelligent Delegation
For many clinicians, the most compelling case for clinical-grade AI is efficiency. The AMA survey found that practices complete around 39 prior authorizations per physician per week, spending about 13 hours weekly on authorization requests. A clinically trained AI platform can automate routine or repetitive tasks, escalating cases to a clinician only when a human determination is needed. Providers experience less paperwork and abrasion, faster turnaround times, and more time spent practicing at the top of their licenses.
The financial and operational stakes are significant. In IDC’s recent analysis, prior authorization costs the U.S. health system between $41.4 and $55.8 billion annually, factoring in labor, delays, and downstream impacts. Early deployments show that machine-learning systems can compress portions of these workflows from days to minutes. As scale increases, those gains compound into meaningful operational savings.
4. Reform Requires Measurable Outcomes
As plans scale their automation strategies to meet 2026 requirements, the central question becomes whether AI meaningfully improves the clarity, consistency, and speed of authorization decisions. The new CMS rule and AHIP-HHS commitments bring this into focus by tying modernization to observable performance across the ecosystem:
- CMS has mandated shorter turnaround times, decision transparency, and metrics reporting; plus, CMS is piloting new prior authorization programs with turnaround times even shorter than current requirements
- Health plans that have pledged to meet AHIP-HHS commitments will be accountable for improving appropriateness, achieving real-time answers for 80% of electronic prior authorizations, and strengthening communication
- Providers will measure outcomes in staff time saved, reductions in avoidable denials, and improvements in member experience
Clinical-grade AI makes these metrics observable. Every decision is explainable and traceable, giving stakeholders a shared view of performance. Because agentic AI executes work rather than suggesting steps, its contribution can be measured directly: fewer manual interventions, faster approvals, and fewer errors.
Each AI agent is built with defined training plans, supervision, and measurable performance goals, functioning as an extension of the clinical team rather than a replacement. This structure creates shared governance between plans and providers, giving both sides clear visibility into how decisions are made, how criteria are applied, and how the AI’s performance evolves over time.
The Opportunity Ahead
The 2026 CMS rule defines the baseline. Prior authorization must become faster, more transparent, and prioritize patient safety, yet generic automation will not reach that goal. Clinical-grade agentic AI, built with physician expertise, governed responsibly, and measured by outcomes, offers the path forward.
The forthcoming standards create the foundation for progress, but they also present an opportunity to go further. Health plans recognize the need for AI and automation to meet emerging standards like the AHIP pledge. The real differentiator is using that foundation to ensure solutions are safe, clinically validated, and capable of modernizing the member and provider experience. By embedding clinical reasoning and explainability into each decision, health plans can improve accuracy, strengthen provider relationships, and set new benchmarks for member satisfaction and operational efficiency.
The challenge is substantial, but the opportunity is larger. Applied thoughtfully, clinical-grade AI can transform prior authorization from one of healthcare’s most persistent bottlenecks into a process that supports patients, providers, and health plans alike. The coming mandate is more than a deadline. It is a chance to get prior authorization right.
Gigi Yuen serves as Cohere’s Chief Data & AI Officer, leading the data organization, which includes data management, data science, machine learning/AI, and business intelligence.
Gigi has more than 20 years of experience leading cross-functional teams in data solution innovation. Prior to joining Cohere, she established the first-ever AI and Analytics function in Availity. At IBM, Gigi was recognized as a Distinguished Engineer. She became one of the first data scientists to receive this designation that is reserved for only the top 0.2% of technical staff in the global firm. Additionally, she led R&D for Watson Health’s $100M+ analytic portfolio, where she drove advancements in patient care, real-world evidence research, and actuarial modeling. Gigi is an author of more than 15 peer-reviewed publications and has nearly 20 patents granted or pending.
Gigi holds a bachelor’s, master’s, and Ph.D. in Engineering from Northwestern University.
No comments:
Post a Comment