AI Reimbursement on Four Pillars:
Cox, Parikh, Warshawsky, and Li
The conversation about artificial intelligence (AI) in healthcare has evolved from speculative optimism to operational reality. In the last three years, four distinct yet complementary perspectives have emerged that together define the new logic of AI reimbursement. Damon Cox’s GENOSAN framework describes how companies can make AI billable and defensible in payer systems. Parikh and Helmchen’s 2022 paper in npj Digital Medicine offers a policy architecture for paying AI based on value rather than volume. Mark Warshawsky’s 2025 report for the American Enterprise Institute (AEI) translates those conceptual levers into the institutional machinery of the Centers for Medicare and Medicaid Innovation (CMMI). Finally, Li and colleagues’ 2025 analysis in npj Digital Medicine adds a critical empirical dimension—analyzing roughly 400 real-world cases of AI reimbursement worldwide to reveal the structural bottlenecks that still constrain market adoption.
Taken together, these four works outline not just a vision but an evolving system: a reimbursement ecosystem that spans implementation, incentive design, policy enforcement, and empirical validation.
Cox and the Operational Foundation: The First Mile Problem
Damon Cox’s GENOSAN Framework for AI Reimbursement (2025) begins at the granular level of billing, coverage, and documentation—the “first mile” where most AI innovations fail. Cox’s insight is pragmatic: reimbursement collapse is rarely philosophical; it happens in the trenches of misaligned codes, ambiguous coverage policies, or missing medical-necessity narratives. GENOSAN divides the problem into three phases—coverage, coding, and reimbursement—each demanding operational discipline and documentary rigor.
Coverage alignment ensures that each use case matches a payer-recognized indication and a defensible clinical rationale. Coding strategy secures the correct CPT or HCPCS alignment to avoid denial through bundling or miscoding. Reimbursement, finally, concerns the machinery of claim execution—tracking denials, iterating appeals, and building payer-by-payer issue logs. GENOSAN’s philosophy is iterative and procedural, not theoretical: reimbursement readiness is achieved through evidence of compliance, not rhetorical argument.
This framework anchors the system. Without a Cox-style foundation of administrative precision, the most sophisticated policy model remains inert. GENOSAN is therefore not a theoretical entry point but a functional prerequisite.
Parikh and Helmchen: Paying for Intelligence, Not Activity
Parikh and Helmchen’s Paying for Artificial Intelligence in Medicine (2022) remains the intellectual pivot for any serious discussion of AI payment reform. Their argument is elegant: existing fee-for-service logic pays for activity—human labor performed per instance—whereas AI creates value by removing friction and redundancy. Paying per use would replicate the moral hazard of the very system AI seeks to fix.
They proposed five pathways to align reimbursement with value:
-
Outcome-based payments, rewarding demonstrated improvements in patient results or system efficiency.
-
Advance market commitments, guaranteeing purchases of validated AI tools addressing priority needs.
-
Time-limited add-on payments (akin to Medicare’s NTAP/TDAP), bridging adoption until data justify integration into base rates.
-
Interoperability and fairness bonuses, encouraging generalizable and bias-mitigated systems.
-
Temporary separate payments with sunset provisions to support early diffusion.
Their framework is normative rather than mechanical: it defines the ethical and economic architecture for AI reimbursement. Parikh and Helmchen reframe the debate not as whether AI deserves payment, but how to pay in a way that promotes measurable system value and guards against volume-based inflation.
Warshawsky: CMMI and the Institutionalization of Incentives
By 2025, the U.S. Center for Medicare and Medicaid Innovation (CMMI) had become the logical venue for testing Parikh’s theories under real budgetary pressure. Mark Warshawsky’s AEI report reframes CMMI’s mission: from “pilot funder” to “productivity enforcer.” He argues that decades of healthcare innovation have failed to generate macro-level cost savings because they were voluntary.
Warshawsky proposes a new class of mandatory, risk-bearing models where AI adoption becomes an assumed productivity baseline. In his Tech-Enabled Prior Authorization Model, vendors are compensated for reducing inappropriate imaging and administrative waste. In the Ambulatory Specialty Model, regional providers share both gains and losses, creating incentives to use cost-saving technology—or suffer benchmark penalties.
Warshawsky’s approach completes the economic feedback loop Parikh began. The innovation pipeline (Parikh’s value design) becomes embedded in CMMI’s regulatory and actuarial infrastructure, turning “should adopt” into “must adopt.” Productivity thus becomes an enforceable policy objective rather than a moral aspiration.
Li et al. (2025): The Reality Check—400 Cases of AI Reimbursement
The missing empirical leg of this discussion arrived in 2025, when Li, Powell, and Lee published Commercialization of Medical Artificial Intelligence Technologies: Challenges and Opportunities in npj Digital Medicine. Drawing on approximately 400 cases of AI medical products navigating commercialization and reimbursement, they illuminate the gulf between conceptual policy frameworks and lived industry experience.
Their findings confirm that while over 1,000 AI medical devices have received FDA clearance, only a small fraction achieve sustainable reimbursement. Common barriers include:
-
Fragmented coding frameworks, with few AI-specific CPT codes;
-
Lack of early engagement with payers, leading to post-approval stagnation;
-
Underdeveloped evidence for cost-effectiveness, which deters investor and payer confidence;
-
Sparse integration into clinical guidelines, which impedes adoption even after approval.
The Li team identifies a handful of success stories—firms that achieved not only regulatory clearance but also payment integration. Their secret, strikingly, aligns with Cox’s operational playbook and Parikh’s economic logic: early alignment with regulatory standards (ISO, FDA 510(k)), sustained health-technology assessments to quantify savings, and active participation in payer code development. They cite a U.S. stroke-detection AI that secured both FDA clearance and a Medicare payment rate of $1,040 per case, demonstrating that deliberate engagement with CMS can turn innovation into revenue.
Li et al. thus provide the empirical confirmation that AI reimbursement is neither hypothetical nor automatic. Success depends on systematic engagement across regulatory, economic, and clinical ecosystems.
Synthesis: The Four Pillars of a Maturing Reimbursement Ecosystem
The combined message of these four works is clear: AI reimbursement is no longer a frontier of speculation but a field of engineering—policy engineering, financial engineering, and administrative engineering. Each author supplies a pillar of the structure:
| Framework | Core Focus | Key Contribution |
|---|---|---|
| Cox (2025) | Operational | Codifies reimbursement readiness—coverage, coding, documentation |
| Parikh (2022) | Conceptual | Defines incentive logic: pay for outcomes, not activity |
| Warshawsky (2025) | Institutional | Embeds value logic into mandatory CMMI payment models |
| Li et al. (2025) | Empirical | Documents real-world progress and obstacles in 400 reimbursement cases |
Cox teaches how to make an AI system billable. Parikh articulates why payers should want to pay for it. Warshawsky shows how government can enforce adoption. Li demonstrates what happens when these ideals meet reality.
When integrated, these pillars suggest an ecosystem where innovation and reimbursement are co-designed from the start. The GENOSAN approach grounds compliance; the Parikh model sets value metrics; the Warshawsky model enforces diffusion through systemic benchmarks; and the Li dataset provides continuous reality testing. The result is a self-correcting feedback loop—data from reimbursement outcomes refine the incentive models, which in turn inform both policy and operational practice.
Toward Evidence-Based Reimbursement Policy
Perhaps the deepest lesson from Li et al.’s dataset is that AI reimbursement policy must itself become evidence-based. Whereas early debates relied on analogies to drugs or devices, empirical mapping of hundreds of cases now enables comparative policy analytics. One can quantify which incentive models—add-on payments, shared-savings bonuses, or bundled integration—produce durable adoption and cost savings. This transforms AI reimbursement from conjectural policy theory into a measurable discipline.
In this sense, the field is approaching a turning point. Cox gives the compliance tools, Parikh the design principles, Warshawsky the enforcement venue, and Li the data infrastructure. Together they illustrate a maturing ecosystem in which reimbursement is no longer the “last barrier” after innovation, but a co-equal design parameter in the innovation process itself.
If Parikh’s thesis was to “pay for intelligence,” the collective contribution of these four frameworks might be summarized as “govern for intelligence.” That is, align the operational, economic, institutional, and empirical components of the health economy so that intelligent systems—human or artificial—improve care while sustaining financial integrity.
References
Cox, D. (2025). GENOSAN: Framework for AI Reimbursement. GenoSan Consulting.
Parikh, R. B., & Helmchen, L. A. (2022). Paying for Artificial Intelligence in Medicine. npj Digital Medicine, 5(63).
Warshawsky, M. (2025). Reimbursement and Incentives for AI Efficiency: The Role of CMMI. American Enterprise Institute.
Li, B., Powell, D., & Lee, R. (2025). Commercialization of Medical Artificial Intelligence Technologies: Challenges and Opportunities. npj Digital Medicine, 8(454).
