Transforming HEDIS Abstraction with Agentic AI: What Health Plans Need to Know

The $1 Billion Wake-Up Call

Health plan quality bonuses dropped $1 billion in 2024—from $13 billion to $12 billion. Not because care quality declined, but because health plans couldn't efficiently prove it with data.

The culprit? Manual HEDIS abstraction processes that consume months of clinical staff time, cost hundreds of thousands of dollars, and still leave care gaps open until it's too late to close them.

Meanwhile, a new category of AI technology is emerging that could fundamentally change how quality measurement works. It's called agentic AI, and forward-thinking health plans are beginning to understand its implications.

The HEDIS Abstraction Challenge

Every quality director knows this reality: HEDIS season means mobilizing teams of clinical abstractors for 3-4 months, manually reviewing thousands of charts, racing against NCQA deadlines.

HEDIS abstraction is labor-intensive, prone to human error and inconsistencies, and the evolving nature of HEDIS measures requires constant updates and training. Quality improvement abstractors struggle to efficiently extract critical medical evidence from unstructured clinical notes, creating significant operational bottlenecks.

But the real cost isn't just operational—it's strategic. Manual processes mean most abstraction happens retrospectively during "HEDIS season" instead of continuously throughout the year. By the time you identify care gaps, your window to close them has narrowed dramatically.

The financial stakes make this more than an efficiency problem. High Star Ratings (4+ stars) earn significant bonus payments, while a drop in ratings can cost plans hundreds of millions—one major insurer fell from 4.5 to 3.5 stars and saw an $800 million reduction in operating income.

Why Current AI Solutions Address Only Half the Problem

The healthcare AI market has made real progress. AI-assisted abstraction tools can reduce manual effort by 60-75%, highlighting relevant evidence and guiding abstractors to critical information faster.

This is valuable. But it's not transformational.

These solutions are assistants, not agents. They still require:

• Human review of every finding

• Manual validation and data entry

• Separate processes to close identified gaps

• Limited learning from corrections

• Seasonal rather than continuous operation

They make your team faster. But they don't change what's fundamentally possible.

The Agentic AI Difference: From Tool to Partner

Agentic AI represents a different paradigm—systems that don't just assist with tasks, but reason through complex workflows, make decisions within defined parameters, and take action autonomously while learning continuously.

In the context of quality measurement, this means moving from:

• Sample-based abstraction → Population-wide surveillance

• Retrospective gap identification → Real-time detection

• Seasonal operations → Year-round quality monitoring

• Human-intensive validation → Autonomous processing with human oversight

• Static systems → Continuously learning platforms

The distinction matters: AI-assisted tools make existing processes faster. Agentic AI makes entirely new approaches possible.

What Becomes Possible: HEDIS Abstraction

When AI can abstract, reason, and act autonomously—with appropriate oversight—health plans can reimagine their entire approach to quality measurement:

• Year-Round Quality OperationsInstead of mobilizing for HEDIS season, quality teams monitor continuous AI-driven surveillance, intervening when the system identifies gaps that require human judgment or complex care coordination.

• Population-Wide CoverageRather than statistical sampling, the technology enables review of every member's complete medical record, identifying opportunities that would never appear in a 411-chart sample.

• Proactive Gap ClosureCare gaps get identified in real-time as documentation occurs, creating 12-month windows for intervention instead of 2-month scrambles after abstraction is complete.

• Intelligent PrioritizationThe system doesn't just find gaps—it helps triage them based on likelihood of closure, member engagement patterns, and ROI potential.

• Continuous LearningEvery human correction, every new clinical guideline, every NCQA technical specification update trains the system to improve.

• This isn't speculation about distant future technology. The foundational capabilities exist today. The question is implementation readiness.

The Economics Are Compelling—But Complex

Health plans exploring agentic AI typically model scenarios like this:

Current State:

• 5,000 charts requiring manual abstraction

• 15 minutes per chart average

• $75,000 in direct labor costs

• 3-4 month timeline

• Sample-based coverage

Agentic AI Future State:

• Entire population under continuous surveillance

• Minutes of human oversight per day instead of hours of abstraction

• Estimated 60-70% reduction in manual effort

• Year-round operation

• 10x coverage expansion

The potential ROI isn't just cost reduction—it's the revenue impact of improved Star Ratings, better member outcomes, and competitive positioning as the industry shifts toward digital quality measures.

But these benefits require significant organizational change, technical infrastructure investment, and realistic timelines. Organizations expecting immediate transformation typically struggle.

Three Forces Making This Urgent

1. NCQA's Digital Transition

NCQA is moving toward digital quality measures using HL7 FHIR and Clinical Quality Language. Health plans building advanced data capabilities now will be prepared. Those waiting will face crisis when manual processes become obsolete.

2. Economic Pressure Intensifying

The $1 billion decline in quality bonuses signals tightening margins across the industry. Manual processes costing $300K-$500K annually for mid-size plans are becoming economically unsustainable.

3. Technology Maturity Arriving

Large language models now demonstrate clinical reasoning capabilities that were impossible two years ago. Agentic frameworks enable reliable autonomous action. The risk profile has fundamentally shifted.

Early adopters moving now will have 24-36 months of learning and optimization before this becomes table stakes.

The Implementation Reality

Success with agentic AI requires more than technology procurement. It requires:

1. Strong Data Foundation: Clean, accessible, well-structured clinical data in standardized formats. Many health plans discover their data isn't ready.

2. Organizational Readiness: Clinical staff transitioning from abstraction to oversight roles. Change management for transformed workflows. Leadership commitment to multi-year transformation.

3. Regulatory Navigation: NCQA audit compliance for automated processes. Documentation standards. Member privacy requirements. Liability frameworks.

4. Realistic Expectations: Benefits typically materialize over 18-36 months, not 3-6 months. Pilot phases are essential. Learning curves are real.

   The organizations succeeding are those treating this as strategic transformation, not technology implementation.

   
   

What SmartSigma AI Believes

We see health plans at an inflection point. The technology enabling agentic approaches to quality measurement has arrived. The economic and competitive pressures demanding new approaches are intensifying. The regulatory environment is evolving to support digital measurement.

But potential isn't automatic. Success requires:

• Strategic vision beyond operational efficiency

• Technical foundation in data infrastructure

• Organizational commitment to change

• Expert guidance through complex transitions

The health plans that move thoughtfully now—assessing readiness, piloting carefully, learning continuously—will have decisive advantages in 2026-2027 when these capabilities mature.

Those that wait for certainty or perfect solutions will find themselves scrambling to catch up when digital quality measurement becomes the standard.

The Question Facing Quality Leaders

The technology discussion is settled—agentic AI will transform HEDIS abstraction. The only questions are timing and readiness:

• Is your data infrastructure prepared?

• Is your organization ready for this change?

• Do you have a roadmap that balances ambition with realism?

• Are you positioned to learn now, or play catch-up later?

These aren't technology questions. They're strategic leadership questions.