AI-enhanced document review streamlines Medicare Advantage workflows at national plan

For Medicare Advantage plan providers, the annual enrollment period (AEP) is a sprint—not a marathon—but the race doesn’t end when that window closes. One large national health plan with a significant Medicare Advantage footprint wanted to position its teams to finish stronger, with less strain and burnout during the most demanding stretches of the review cycle. The answer kept pointing back to the evidence of coverage (EOC) and annual notice of change (ANOC) documents the Centers for Medicare & Medicaid Services (CMS) requires to communicate approved benefits to beneficiaries.

The challenge

EOC documents can run several hundred pages, often encompassing more than one plan and incorporating benefit structures, CMS model language and plan-specific details. Across a review cycle, teams work through dozens of these files, each requiring independent validation.

The review process itself is methodical, largely manual and resource intensive. Draft documents are authored using structured benefit data, then reviewed by a separate quality team responsible for validating benefits, cost sharing and regulatory requirements before submission. Reviewers compare draft language against source materials line by line to confirm compliance. As documents grew and became more complex, the organization began examining how to optimize the process and keep it repeatable.

Building the expertise to do that work reliably takes time. Developing the needed level of proficiency requires a sustained investment in training across multiple review cycles. Reviewers must internalize CMS model language, plan-specific benefit structures and the logic used to validate consistency across hundreds of pages. When experienced staff rotate off the work or leave the organization, the learning curve resets and the organization must rebuild that expertise.

“Working with independent reviewers means we look for ways to increase what we call muscle memory,” a senior manager in national Medicare business operations said. “We train once on each part of the ANOC or EOC, explain how the model maps to the plan benefit package (PBP), demonstrate how to analyze or research discrepancies and continue building on that foundation. Generative AI supports reliability by allowing us to train and refine in a shorter time frame, and it doesn’t forget what it learned before. We get reliable, repeatable results.”

The solution

ZS partnered with the client to develop a gen AI-powered quality review accelerator that integrates into established workflows. While independent human oversight remains critical, the tool conducts initial comparison checks using existing review logic and terminology and applying the same structured checks across documents and plan years. By formalizing core comparison logic, the accelerator helps stabilize the process across review cycles and reduce dependence on rebuilding expertise each year. Reviewers can then focus their attention where it’s required.

ZS began by listening. The team worked with business and compliance stakeholders to understand how document reviews functioned in practice and where manual comparison absorbed the most time. An initial first pass was introduced within an active review cycle, allowing the client to refine the approach before broader deployment.

The impact

Once implemented across the full set of EOC and ANOC documents, the AI-assisted review approach materially reduced the time required for manual validation. Documents that previously required roughly 10 hours of review were completed in three to four hours—a 60%-70% reduction in manual review time.

That shift changed the cadence of the work. Reviewers spent less time moving line by line through repetitive comparisons and more time focusing on areas that required interpretation and judgment. Automating the initial comparison checks reduced the risk of inconsistencies surfacing under deadline pressure.

The initiative also marked the organization’s first business-facing use of gen AI. Applied within a compliance-driven process, it demonstrated automation can ease operational strain, preserve critical standards that reduce the chances of errors and improve future scalability.

“By leveraging some of the advantages of working with generative AI,” the senior manager said, “we could quickly develop a connection among the various sources such as PBPs and models, refine checks and results and easily repeat the results across a lot of disparate documents.”