Medicare Advantage risk adjustment is one of the most directly measurable connections between data quality and revenue in the healthcare insurance industry. CMS pays health plans a capitated rate for each enrolled member based on the member’s Risk Adjustment Factor score, which is calculated from the Hierarchical Condition Categories submitted through encounter data. A member with a higher RAF score — reflecting a more complex clinical profile — generates higher capitation revenue. A member whose clinical complexity is not fully represented in submitted HCC codes generates lower revenue than their expected cost profile would justify.
The economics are specific. Each 0.01 RAF score point represents approximately $80 to $100 per member per year in capitation revenue. On a Medicare Advantage population of 100,000 members, a 5 percent improvement in coding gap closure translates to $4 to $5 million in annual revenue recovery. That revenue is not speculative — it is the legitimate revenue for care that was provided, documented, and clinically supported, but not fully captured in the encounter data submitted to CMS.
The distinction between undercoding and overcoding
The risk adjustment programme must operate within a compliance boundary that makes the distinction between undercoding and overcoding critical. Undercoding — failing to submit HCC codes that are supported by clinical documentation — leaves legitimate revenue on the table. Overcoding — submitting HCC codes that are not adequately supported by clinical documentation — is a compliance risk that has resulted in significant CMS audit findings, OIG investigations, and in extreme cases Department of Justice enforcement action against Medicare Advantage plans and their partners.
The objective of an AI-assisted risk adjustment programme is complete and accurate coding: every condition that is clinically documented and supported should be coded, and no condition should be coded that is not. A model that identifies supported but unsubmitted codes is a revenue recovery tool. A model that suggests codes based on inferred or probabilistic clinical inference rather than documented clinical evidence is a compliance liability. The design of the programme determines which it is.
This distinction is where many risk adjustment analytics programmes introduce risk. A chart review vendor that suggests HCC codes based on statistical inference from pharmacy claims rather than clinical documentation review is not improving coding completeness. It is generating compliance exposure. The programme that uses AI to identify which charts are most likely to contain undocumented supported HCCs — and then directs clinical review to those charts — is adding value through prioritisation rather than through inference. The clinical reviewer makes the coding determination based on the documentation, not the model.
The two components of coding gap closure
Retrospective coding gap closure is the review of prior year clinical documentation to identify conditions that were treated and documented but not submitted as HCC codes in the encounter data for that year. At scale, this requires reviewing tens of thousands of charts for a large Medicare Advantage population. Without prioritisation, the chart review cost per identified gap is high enough to erode the revenue benefit. With prioritisation — a model that identifies which members are statistically most likely to have coding gaps based on their pharmacy utilisation, diagnosis codes, and encounter patterns — the chart review resources are directed to the highest-yield cases.
Prospective risk stratification addresses the problem earlier. A model that identifies members likely to have undocumented chronic conditions based on pharmacy claims, laboratory results, and care utilisation patterns — but whose diagnostic codes do not fully reflect the clinical profile those signals imply — enables proactive outreach to the member and their treating physician before the encounter. A health plan that reaches out to a member who is filling medications consistent with a chronic condition that is not appearing in their diagnostic record, and facilitates a clinical encounter where the condition is assessed and documented, improves coding completeness through clinical engagement rather than retrospective chart review.
Encounter data completeness monitoring adds a third dimension. A large percentage of coding gaps are not failures to document or code the condition at the clinical level — they are failures to submit the encounter data completely to CMS. Encounter data completeness models that identify submission gaps before the CMS deadline enable correction within the submission window rather than through costly retroactive adjustment.
The compliance architecture
The programme must maintain a complete audit trail from submitted HCC to clinical documentation. Every code submitted must be traceable to a specific clinical encounter where the condition was documented by the treating provider. CMS audit readiness requires that this traceability is preserved for every submitted code for the applicable lookback period. A programme that generates revenue through coding improvement but cannot demonstrate the documentation chain that supports the submitted codes is creating compliance exposure that offsets the revenue benefit.
What success looks like
The metrics are RAF score completeness versus expected based on the clinical profile of the population, encounter data completeness rate, CMS audit finding rate, coding gap closure rate per chart reviewed, and prospective outreach conversion rate for identified risk stratification targets. The programme should track these metrics separately to distinguish between retrospective recovery — finding codes already supported by existing documentation — and prospective improvement — improving the clinical documentation and encounter submission process so that gaps are smaller in the next coding year than in the current one.
