While coding accuracy has always been vital in Medicare Advantage, recent changes—such as expanded RADV audits and the rollout of CMS-HCC V28—have changed how risk adjustment teams gauge success.
Universal RADV audits mean that success no longer depends solely on how completely plans capture risk, but also on how confidently they can defend every submitted diagnosis. And with the significant reduction in payment-eligible codes under V28, risk adjustment teams must now evaluate vendor and coding team performance not only on ROI, but also in terms of how much value went uncaptured.
Redefining the vendor scorecard
Historically, health plans assessed risk adjustment coding vendors by answering three core questions:
- Did the vendor identify additional supported HCCs?
- Were the recommended codes accurate?
- Were unsupported diagnoses flagged for deletion?
While these metrics remain relevant, they are no longer sufficient. Increasingly, risk adjustment leaders are refocusing evaluation frameworks around two critical questions designed to protect against audit exposure and maximize the value of every retro coding project:
- How much unsupported RAF exposure was avoided?
- How much defensible value went undetected?
In the past, answering these questions required costly and time-consuming QA that most teams weren’t ready to support. Today, new technological solutions have emerged to make answering them faster and easier.
From retrospective QA to continuous performance management
Traditional quality assurance in HCC coding relied heavily on retrospective manual sampling of charts after coding was completed. But this process always had shortcomings, including blind spots, delayed feedback, and the inability to offer complete visibility across entire populations. Today, with advancements in AI applications, QA is transitioning from a delayed audit function into a powerful tool for continuous performance management.
Rather than asking reviewers to manually review thousands of charts, AI can rapidly organize medical records by date of service, surface evidence supporting HCCs, identify potential missed diagnoses, and flag unsupported codes for targeted review. This enables experienced coders to focus on validating documentation and coding decisions rather than manually searching thousands of pages. The cost and time savings generated by AI efficiencies allow organizations to evaluate a much larger percentage of their coding population than traditional QA methods permit—often 100%.
The same cannot be achieved by staffing up: The operational rigors of the new risk adjustment landscape cannot realistically be met by layering more headcount onto traditional retrospective audit processes. Manual chart sampling, while useful for identifying broad trends, inherently lacks the granular visibility required to address the precision mandates of CMS-HCC V28 and universal RADV audits. Sampling provides a snapshot, but what organizations require today is comprehensive, detailed analysis.
By integrating AI into the coding workflow, organizations gain the ability to shift their QA strategy along three strategic pillars:
- Operationalizing population-wide visibility: Traditional QA relies on statistical samples, which inevitably miss the "unknown unknowns"—systemic coding drifts that occur across the broader patient population. AI-driven QA allows for the evaluation of entire populations, providing the comprehensive data necessary to confirm that coding accuracy is consistent across every contract, not just the selected charts. This moves the goalpost from statistical confidence to total audit readiness.
- Reclaiming expert human capital: The most significant bottleneck in any QA program is the time expert coders spend hunting and gathering evidence. AI automates the mechanical organization of medical records by date of service and links HCCs directly to evidence in clinical documentation. This effectively redefines the role of the auditor: rather than spending most of their time searching for relevance, they are freed to focus entirely on complex clinical judgment and coding validation.
- Shifting from post-mortem to real-time calibration: One of the most significant liabilities of legacy QA is the feedback loop. By the time errors are identified in a post-project audit, the damage is often done. An AI-augmented approach provides real-time identification of variation. This allows for course correction and fosters a tighter, more collaborative feedback loop with coding teams and vendors. It transforms the vendor relationship from one of quality control to quality assurance, and ensures alignment with regulatory standards happens during the project, not after it.
Accurate coding has always been the primary objective, but in the current regulatory environment, the ability to simply report results is not enough. Organizations must now demonstrate—at scale—that their risk adjustment operations are resilient, defensible, and proactive. The shift toward continuous performance management is a strategic necessity for any Medicare Advantage organization looking to move beyond ROI tracking, achieve lasting audit integrity, and definitively prove the accuracy of their submissions.
To learn how AI can evaluate your vendor performance within days of project completion, schedule a consultation with one of Charta’s risk adjustment AI deployment specialists.