From 40 Coders to 2,000
In its January 2026 HPMS memo, CMS disclosed that it had scaled its RADV audit workforce from approximately 40 certified coders to plans for approximately 2,000. The agency also confirmed it is deploying AI as a “medical coder support tool” to flag coding patterns and assist human reviewers. This scaling happened alongside the shift to annual audits of all 550+ MA contracts with quarterly cadence and variable sample sizes of 35 to 200 enrollees per contract.
The math is straightforward. At 40 coders, CMS could audit a handful of contracts per year with meaningful depth. At 2,000 coders supplemented by AI pattern detection, the agency can process the volume required for universal annual audits while maintaining the clinical review quality that catches documentation failures. The bottleneck that previously limited enforcement capacity no longer exists.
For health plans evaluating their coding vendors, this scaling changes the probability calculation. The question is no longer “will our codes be audited?” It’s “when our codes are audited, will they hold up?” Every vendor’s output is now subject to review by a workforce 50 times larger than the one that existed three years ago, augmented by AI that identifies the patterns manual review would miss.
What CMS’s AI Looks For
CMS hasn’t published the specific algorithms its AI uses for audit targeting, but the agency’s public guidance and enforcement actions reveal the patterns it monitors. Population-level coding intensity that rises without corresponding changes in clinical outcomes. Concentration of submitted codes in high-value HCC categories, particularly acute conditions like stroke, MI, and cancer that OIG audits found failing at 100% rates. Asymmetric coding patterns where additions far exceed deletions across review cycles. And chart review diagnoses that can’t be linked to submitted encounters.
These are the exact patterns that add-only, volume-focused retrospective programs produce. CMS built its AI to find the statistical signatures of the practices its enforcement actions have targeted. Plans whose vendors produce these signatures are generating the data trail that CMS’s technology is designed to detect.
Vendors that understand this don’t just avoid producing these patterns. They actively monitor their own output for them. They track population-level coding distribution, deletion rates, encounter linkage rates, and category concentration across their client portfolios. They catch the signals before CMS does, because they know what CMS is looking for.
The Vendor Evaluation Through the CMS Lens
When evaluating coding vendors, apply the CMS perspective. If CMS’s AI analyzed your vendor’s output across your entire membership, what would it find? Would it see balanced two-way coding or one-directional additions? Would it see coding distributions that reflect clinical reality or concentrations in high-value categories? Would it see evidence trails for every code or recommendation lists without supporting documentation?
Ask your vendor for their population-level coding metrics across all clients. Deletion rate. Category distribution. Encounter linkage rate. If they can’t produce these numbers, they’re not monitoring for the patterns CMS tracks. If the numbers show zero or near-zero deletions, heavy concentration in high-value categories, or low encounter linkage, the vendor’s methodology is producing the exact output CMS’s expanded workforce is built to find.
Selecting for the New Enforcement Scale
Risk Adjustment Coding Companies that operated successfully when CMS had 40 auditors may not survive scrutiny from 2,000 auditors supplemented by AI. The margin for methodological weakness contracted by a factor of 50. Plans selecting vendors in this environment need partners whose output is designed to withstand the scale and sophistication of the current audit operation, not partners whose methodology predates it. The vendors that monitor their own patterns, track defensibility metrics, and proactively flag risks are the ones aligned with what CMS built its workforce to examine.
