Accurate risk adjustment is not just a box to be checked; It is now a strategic lever. Hierarchical condition category (HCC) coding underpins risk scores that drive Medicare Advantage and other value-based payments. Currently, more than half of Medicare beneficiaries are enrolled in Medicare Advantage 2025 (equivalent to approximately 35.7 million people), and coding accuracy directly impacts financial performance and compliance.
Many well-intentioned organizations outsource HCC coding to third parties that promise scale and turnkey accuracy. But in practice, outsourcing can be costly, difficult, and risky. However, recent advances in generative artificial intelligence have made it easier and safer to code HCC in-house, reducing costs and improving audit readiness.
The Hidden Costs (and Risks) of Outsourcing
The business model behind outsourcing HCC coding creates misaligned incentives. Essentially, you're trading a higher payout for a softer accuracy guarantee. Health plans can spend millions of dollars under a chart-by-chart pricing model, but providers rarely provide the transparent, auditable evidence needed to show that coding accuracy is actually better.
Meanwhile, CMS estimates FY 2024 Part C (Medicare Advantage) payment errors at $19.07 billion — a reminder that if you can't see and protect every code, documentation gaps remain a systemic risk.
What's worse? The audit risk is yours, not the supplier's. While CMS has mechanisms to recover overpayments (including extrapolation), it is not a perfect system when diagnostics are not supported in the chart. If an outsourced partner “pushes” the code, you will be liable when the auditors review the records, and they will retain their fees.
Additionally, with most outsourcing models, you can send your protected health information (PHI) away and accept others' thresholds, redaction logic, and risk tolerances. This lack of control and transparency becomes a problem if CMS or a plan reviewer asks, “Why was this HCC assigned?” And you can't come up with any explainable, defensible clues.
Changing regulatory objectives
Imagine hiring a tax company to charge 20% of your deductions instead of charging an hourly rate. They have every incentive to find more deductions and push the envelope. If you are audited, you take responsibility; they keep their position. This is the risk dynamic of many outsourced HCC models: the vendor maximizes near-term revenue, while you face long-tail audit risk.
In Medicare Advantage, the risks are huge. Payments will continue to increase through 2025 as enrollment increases, increasing scrutiny of the accuracy of risk scores and coding practices. The policy update anticipates that the increase in ongoing payments will be related in part to changes in risk scores, prompting further concern from CMS and regulators.
Regulators are making the risks clearer. The Office of Inspector General (OIG) has repeatedly warned about diagnoses that come only from a health risk assessment (HRA) or chart review but are not supported anywhere else in the medical record. This type of code can increase your payment amount, but usually won't pass review. In other words, if the coding is not done right, you take a calculated regulatory risk.
internal alternatives
Thanks to advances in generative artificial intelligence, bringing HCC coding in-house can solve many of these problems at a fraction of the cost and risk profile. Your organization, not the vendor, is in control of editing logic, thresholds, evidence requirements, and escalation paths. This means audit preparation is built into the design, and every recommended and accepted code has full provenance.
Think about it: You've hired a clinical coder. When equipped with the right AI, they can pre-review charts, surface high-yield evidence, and easily expedite secondary reviews without adding staff. Perhaps most importantly, solutions that run within your environment avoid sharing PHI while providing your team with comprehensive observability.
A few years ago, “DIY” meant building a natural language processing (NLP) platform from scratch. no longer. The new AI-based generative HCC coding tool can be integrated into existing workflows to read messy, siled multi-modal data, keep pace with evolving models, run on-premises or in a private cloud environment, and allow you to customize to meet the needs of your own organization.
A safer, smarter path forward
Regulators have made their expectations clear: unsupported diagnoses will be discovered and funds recovered. The OIG will continue to focus on vulnerable coding channels such as HRAs and chart reviews when these channels are not supported elsewhere in the medical record. CMS' Part C error rate work shows billions of dollars are at risk each year.
When the skills gap is large, outsourcing makes sense. The gap has since narrowed. Today, organizations can deploy an AI-native HCC platform behind their own firewall, tailor it to their compliance posture, and operate at a predictable cost per patient while maintaining audit readiness.
Risk adjustment is too strategic to leave your four walls. The future of HCC coding is in-house coding, where through a combination of generative AI and your own clinical coders, organizations can directly address these real-world issues with control, transparency and cost savings.
Photo: Leo Wolfert, Getty Images
David Talby, Ph.D., MBA, is chief technology officer at John Snow Laboratories. He has spent his career enabling artificial intelligence, big data, and data science to solve real-world problems in healthcare, life sciences, and related fields.
This article appeared in Medical City Influencers program. Anyone can share their thoughts on healthcare business and innovation on MedCity News through MedCity Influencers. Click here to learn how.
