The Future of Risk Adjustment: AI, Automation, and What Comes After V28

Where Risk Adjustment Stands Today

The future of risk adjustment refers to the convergence of AI-driven HCC coding, natural language processing for clinical documentation, predictive risk modeling, and next-generation CMS model development that will fundamentally transform how Medicare Advantage plans and providers identify, document, validate, and optimize member risk scores beyond the CMS-HCC V28 framework.

The risk adjustment landscape in 2026 is defined by a single reality: CMS-HCC V28 is now the sole model governing Medicare Advantage payments. The three-year transition from V24 is complete, and every plan in the country is operating under a fundamentally restructured framework with 115 HCCs, constrained coefficients, and a reduced ICD-10 mapping universe of 7,770 codes.

But V28 is not the end of the story. It is a waypoint in a broader CMS strategy to modernize how risk is measured, validated, and compensated across the Medicare Advantage program. The forces reshaping risk adjustment extend well beyond model version updates.

• Regulatory Pressure: RADV audits are intensifying with extrapolation now applied to payment years 2018 forward, creating unprecedented financial exposure for plans with documentation gaps
• Data Volume: The average MA plan now processes millions of encounter records annually, far exceeding what manual review teams can meaningfully audit
• Workforce Constraints: Certified coding professionals remain in short supply, with industry estimates showing a 25-30% shortage of qualified HCC coders nationwide
• Accuracy Demands: CMS estimates that 9.5% of MA payments are improper, translating to billions in overpayments that invite enforcement action

These pressures are converging to make the status quo unsustainable. Organizations that continue relying on manual-first workflows, annual retrospective reviews, and spreadsheet-based analytics will fall behind those embracing technology-driven approaches.

AI in Risk Adjustment

Artificial intelligence is not a future concept for risk adjustment — it is already reshaping operational workflows across the industry. The question is no longer whether AI will play a role, but how quickly organizations can integrate it effectively.

• Automated Code Validation: AI systems can cross-reference ICD-10 codes against clinical documentation in real time, flagging unsupported diagnoses before submission rather than after audit selection
• Pattern Recognition: Machine learning algorithms identify coding patterns that deviate from expected distributions, highlighting both undercoding and overcoding at the provider, facility, and plan level
• Workflow Prioritization: AI-driven triage systems rank charts by revenue impact and audit risk, ensuring that limited coding resources focus on the highest-value opportunities first
• Continuous Learning: Unlike static rule engines, AI models improve with each audit cycle, incorporating new CMS guidance and OIG findings into their validation logic
• Scalability: A single AI platform can process volumes that would require hundreds of manual coders, making comprehensive population-level review feasible for the first time

The automation of RAF score workflows represents one of the most immediate applications. Plans that have deployed AI-assisted coding validation report 40-60% reductions in manual review time while simultaneously improving HCC capture rates by 8-15%.

Natural Language Processing for CDI

Clinical Documentation Improvement has historically depended on trained specialists reviewing individual charts and querying providers about documentation gaps. Natural Language Processing is transforming this labor-intensive process into a scalable, technology-assisted workflow.

NLP systems parse unstructured clinical text — progress notes, discharge summaries, specialist consultations, and operative reports — to identify conditions that are clinically present but not coded. This capability is particularly valuable under V28, where the narrowed ICD-10 mapping universe means every valid code carries greater significance.

• Condition Extraction: NLP algorithms identify clinical language patterns that correspond to HCC-mappable diagnoses, even when providers use non-standard terminology or abbreviations
• Specificity Enhancement: Systems detect when documentation supports a more specific diagnosis code than what was captured, such as identifying laterality, severity, or complication status from narrative text
• Query Generation: Automated CDI queries are generated and routed to the appropriate provider when NLP identifies documentation that is suggestive but not definitive enough for coding
• Retrospective Mining: NLP can scan years of historical clinical notes to identify conditions that were consistently documented but never coded, revealing systematic capture gaps
• Compliance Safeguards: Unlike aggressive upcoding approaches, NLP-driven CDI focuses on capturing what is already clinically present — aligning documentation with reality rather than inflating it

Current NLP systems achieve 85-92% accuracy in identifying HCC-relevant conditions from unstructured notes. While this does not eliminate the need for human review, it dramatically reduces the volume of charts requiring manual assessment and ensures that the most impactful opportunities are never missed.

Predictive Risk Modeling

The shift from retrospective to prospective risk adjustment depends on the ability to predict which members are likely to have undocumented conditions before they present for care. Predictive risk modeling makes this possible.

• Suspect Condition Identification: Models analyze historical claims patterns, pharmacy data, lab results, and utilization trends to generate member-level suspect lists with probability scores for specific HCCs
• Provider Visit Optimization: Predictive scores help plans prioritize which members need targeted outreach for annual wellness visits or specialist referrals, maximizing the yield of each provider encounter
• Risk Stratification Integration: Predictive models feed directly into RAF score forecasting, enabling plans to project revenue impact before the payment year begins
• Medication-Condition Correlation: Pharmacy claims revealing chronic disease medications in the absence of corresponding diagnosis codes serve as high-confidence signals for suspect conditions
• Social Determinant Overlays: Emerging models incorporate neighborhood-level socioeconomic data, access barriers, and health literacy indicators to improve prediction accuracy for underserved populations

Plans deploying predictive risk models report 20-35% improvements in prospective HCC capture compared to traditional chart chase methods. The financial impact compounds: conditions identified prospectively in the current year generate full-year RAF value rather than partial-year value from late retrospective captures.

What Comes After V28

V28 is the most significant CMS-HCC model revision in over a decade, but it is not the final word. CMS has signaled several directions for future risk adjustment evolution that organizations should monitor closely.

• Encounter-Level Risk Adjustment: CMS has explored moving from annual aggregated risk scores to encounter-level payment adjustments that tie compensation more directly to individual service events
• Social Determinant Integration: The current model excludes SDOH factors entirely. CMS has funded research into incorporating area-level deprivation indices, dual-eligibility granularity, and disability status into future model iterations
• Expanded Constraining: The constraining methodology introduced in V28 may extend to additional disease families, further flattening severity-based coefficient differences within condition groups
• Behavioral Health Expansion: Mental health and substance use disorder HCCs are widely viewed as underweighted relative to their cost impact. Future models may introduce more granular behavioral health condition categories
• Real-Time Validation: CMS is investing in infrastructure to support real-time encounter data submission and validation, potentially replacing the current 12-18 month lag between service delivery and RAF finalization

Industry analysts anticipate a V29 proposal could surface in the 2028 or 2029 Advance Notice, though CMS has not committed to a timeline. What is clear is that the direction of travel favors greater precision, more data requirements, and tighter audit enforcement.

Preparing Your Organization

Future-proofing a risk adjustment program requires strategic investment across technology, talent, and process design. Organizations that wait for CMS to finalize changes before responding will consistently lag behind.

• Flexible Technology Architecture: Invest in platforms built on API-driven architectures that can adapt to model changes without requiring complete system rebuilds. When V29 arrives, your infrastructure should need configuration updates, not replacement
• Data Integration First: Build robust data pipelines that consolidate claims, clinical, pharmacy, and demographic data into unified member profiles. AI and predictive models are only as good as the data they consume
• Hybrid Workforce Strategy: Train coding professionals to work alongside AI tools rather than compete with them. The highest-performing teams use AI for volume screening and human expertise for complex clinical judgment
• Compliance by Design: Embed audit readiness into every workflow rather than treating it as a separate function. Every HCC captured should be audit-defensible from the moment of submission
• Continuous Monitoring: Replace annual retrospective reviews with continuous risk monitoring that identifies gaps, validates documentation, and triggers provider outreach in real time

The organizations that thrive in the next era of risk adjustment will be those that treat technology investment not as a cost center but as a core revenue strategy. The gap between technology leaders and laggards in risk adjustment performance is widening, and it will only accelerate as AI capabilities mature.