From Backlogs to Real-Time Decisions 

Asklepios AI Provides Scalable Clinical Support for Prior Authorization 

Abstract

Asklepios, Epidaurus Health’s clinical AI service, has consistently provided strong results for pharmacy benefit managers (PBMs) and payers, demonstrating reliability, fairness, versatility, and scalability in prior authorization (PA) support. 

PBMs and payers often face significant operational challenges when processing PA reviews. Growing case volumes, limited clinical staffing, and increasingly complex evidence-based criteria create a clear need for smarter, more scalable solutions. This case study evaluates Epidaurus Health’s Asklepios AI, with the goal of helping small to midmarket PBMs improve efficiency, accuracy, and turnaround times in their PA workflows. 

Over the final two quarters of 2025, Asklepios AI delivered measurable performance gains for new customers during limited access enterprise trials, including faster case evaluation, more consistent application of medical necessity criteria, and reduced cognitive burden for reviewers. Through a combination of generative AI, structured evidence extraction, and guideline-aligned recommendations, the Asklepios AI helped clinicians to complete reviews more efficiently while maintaining rigorous clinical quality standards. 

This study highlights the impact of generative AI-driven workflow augmentation for PBMs and payers seeking to expand their PA teams’ capacity and demonstrates the scalability and reliability of Asklepios AI as a viable solution for clinical decision support.

Problem Statement

Small to midsize PBMs frequently struggle to keep pace with the rising volume and complexity of prior authorization requests. Clinical pharmacists must manually interpret medical records, extract relevant criteria, and evaluate documentation against a growing body of FDA labeling, guideline updates, and client-specific coverage rules. This manual process results in inconsistent decision-making, prolonged turnaround times, reviewer fatigue, systemic backlog, and operational bottlenecks. 

Customers that signed up for limited access enterprise trials with Epidaurus Health were in the market for a scalable AI solution that could streamline clinical review services, improve consistency, and maintain regulatory and clinical accuracy without expanding headcount. 

The challenge was to implement a reliable, scalable clinical decision support tool that would accelerate case processing, reduce cognitive burden on reviewers, and enhance overall performance across aprior authorization program.

Background

Development on Asklepios, an AI-powered module built to transform how organizations process prescription drug prior authorizations, began in early 2025. 

Asklepios applies HIPAA-compliant generative AI to unstructured patient data extracted from image-based documents and text-based PDFs to deliver a clinical recommendation for prior authorization cases. Its curated knowledge base includes current clinical practice guidelines, peer-reviewed journal articles, and FDA prescribing information enabling the system to evaluate patient data against the latest standards of care with consistency and precision. 

Asklepios AI Evaluation Methodology

For new customers in Q3 and Q4 2025, Epidaurus Health granted access to Asklepios AI for a limited trial period. During the trials, clinicians at each client organization had the option of securely submitting their prior authorization cases to Asklepios and receiving a recommendation, rationale, and cited sources. 

Clinicians were instructed to use Asklepios as an adjunct tool to their existing prior authorization CRM, rather than as a replacement. Epidaurus Health’s platform captured user feedback on Asklepios’ accuracy through a binary response affordance (i.e., thumbs up / thumbs down to indicate agreement or disagreement with each clinical recommendation’s supporting rationale).

At the conclusion of each trial period, Epidaurus Health evaluated three primary metrics: human-AI agreement, the approval/denial distribution, and case processing time.

Results

Over a total of 1,376 cases processed through Asklepios, Epidaurus Health logged an average AI accuracy of 87% based on clinician agreement. Notably, when Asklepios approved a case with a 95% confidence level, human pharmacists concurred with the decision 97% of the time. This outcome strongly suggests that a high confidence recommendation from Asklepios, when supported by comprehensive source evidence, is highly predictive of true positive decisions.

These results spanned across 354 unique drugs and 16 therapeutic classes (see Figure 1), reflecting a broad clinical footprint across therapeutic classes, disease states, and complexity levels. Asklepios’ ability to perform consistently across such a diverse drug set demonstrates that its knowledge base and AI engine are not narrowly optimized for a small subset of therapies. This breadth is a critical indicator of scalability, suggesting Asklepios can be deployed across formularies and benefit designs without significant re-training or customization.

The overall decision distribution was 52% approvals and 48% denials (see Figure 2), indicating Asklepios does not demonstrate a bias toward either outcome. This outcome distribution is significant for PBMs and health plans, providing confidence that AI-enabled decision support maintains clinical objectivity without systematically shifting utilization patterns in either direction.

Asklepios had an average processing time of 60 seconds, synthesizing patient history and surfacing the most clinically relevant information for pharmacist review based on the medication and diagnosis. Pharmacists consistently reported that Asklepios excelled in managing clinically complex cases, which often include large amounts of unstructured clinical documentation.

Early success averages of 30–35% time savings per case, translated directly into higher operational throughput, with organizations realizing an average capacity increase of 46-54%. These efficiency gains enable organizations to scale their prior authorization volume without additional headcount, illustrating how Asklepios can reshape prior authorization processes while preserving quality and consistency (see Figure 3).

Discussion

The human-AI agreement rate observed during these limited access enterprise trials raises several important questions. What explains the approximately 13% of cases in which clinicians’ judgments diverged from AI recommendations? More importantly, what does this divergence reveal about where AI adds the most value relative to human expertise, and how should these insights inform future design decisions for Asklepios and its downstream business impact?

The divergence between human judgment and Asklepios’ recommendations can arise for several reasons:  

1. Clinical precision often depends on contextual nuance that does not lend itself to clean codification. 
2. Critical clinical context may be missing, implicit, or inconsistently documented in the underlying record. 
3. Prior authorization criteria themselves contain inherent ambiguities that require interpretive judgment or forward inference. Phrases such as “documented treatment failure” or “clinically significant response” are rarely self-defining and can reasonably yield different conclusions, even among experienced human reviewers. 

In cases where Asklepios identified robust supporting evidence and high certainty for approval, human reviewers concurred at 97%. Asklepios performs exceptionally well when documentation is complete and clinical protocols are unambiguous, the conditions under which clinical decisions should be the most straightforward. This data provides a starting point for a tiered approach to automation: identifying cases appropriate for touchless AI-driven decisions, while reserving more complex or ambiguous cases for intentional human-AI collaboration.

Looking ahead, observed divergence will help inform future iterations of Asklepios. Enhanced evidence retrieval, knowledge base augmentations, and systematic analysis of disagreement patterns will all progressively shape clinical human-AI collaboration. Asklepios is designed to surface sound clinical reasoning grounded in deep clinical context and codified criteria, while preserving human oversight for cases requiring interpretation, ethical judgment, or resolution of ambiguity. Where clinical nuance resists codification, documentation is incomplete, or criteria are inherently ambiguous, a divergence can appropriately trigger a focused human review rather than proceed with a reductive automation. This highly calibrated division of human-AI labor allows AI to improve consistency and efficiency in routine cases, while reserving human expertise for decisions where professional human judgment adds the greatest value.

Perfect human-AI concordance is neither achievable nor desirable in this use case. Rather than pursuing perfect agreement, the goal is to ensure that all of Asklepios’ recommendations fall safely and consistently within the bounds of established clinical knowledge.

Conclusion

Limited access enterprise trials during Q3 and Q4 of 2025 demonstrate that Asklepios AI delivers on its promise to enhance prior authorization processing through a powerful combination of clinical accuracy, operational efficiency, and real-world versatility. Processing 1,376 cases across 354 unique medications in just 60 seconds per case, the system tackles the two greatest challenges facing PBMs and health plans today: processing backlogs and resource constraints.

What currently sets Asklepios apart is its breadth and balance. As an AI service, Asklepios performs consistently across diverse therapeutic categories and complex clinical scenarios without the need for custom configuration, demonstrating true scalability. Moreover, a balanced 52% to 48% approval-to-denial ratio aligns with current industry expectations for utilization management decision distributions.

With 87% overall accuracy and 97% pharmacist agreement on high-confidence approvals, Asklepios proves that AI can match expert clinical judgment while dramatically accelerating decision-making. The 13% of cases where human reviewers diverged from Asklepios reflects the inherent complexity of clinical decision-making rather than systemic bias or model failure. Going forward, Epidaurus intends to leverage observed divergence to calibrate data-driven triggers for safely flagging which cases should be touchless and which require clinicians in the loop.

For organizations struggling with rising prior authorization volume, Asklepios AI offers a clear and safe path forward. Pharmacists gain back hours during their day to focus on cases requiring nuanced judgment, while patients experience faster access to necessary medications. Furthermore, complex, nuanced cases that traditionally consume the most time become more manageable with generative AI-powered synthesis and evidence retrieval. Early efficiency gains, including 30–35% time savings per case and a 46–54% increase in organizational capacity to accommodate increased caseloads, demonstrate how Asklepios meaningfully alleviates operational strain and enables organizations to scale without additional staffing.

The results speak for themselves: higher accuracy, faster decisions, and happier clinicians, all while maintaining the clinical integrity that defines quality care.

Asklepios is ready to scale across your organization. Contact Epidaurus Health to learn how we can transform your prior authorization workflow.