Reimagining AI Reimbursement:
From GENOSAN to Parikh to Warshawsky
Artificial intelligence in healthcare has advanced more quickly than the reimbursement systems meant to sustain it. Damon Cox’s GENOSAN Framework provides an on-the-ground, operational lens for making AI reimbursable. Parikh and Helmchen’s 2022 npj Digital Medicine paper supplies the conceptual blueprint for aligning AI payment to value rather than volume. And in 2025, Mark Warshawsky at the American Enterprise Institute updates those ideas within the specific machinery of the Centers for Medicare and Medicaid Innovation (CMMI), arguing that AI reimbursement must evolve to drive genuine productivity growth rather than merely add costs. Read together, these works trace a coherent intellectual arc—from practical triage, to policy design, to macroeconomic reform.
Damon Cox and the GENOSAN Framework: Getting Paid in the Real World
Damon Cox’s GENOSAN framework is not an abstract policy essay but an operations manual. Its goal is to help early-stage AI and digital diagnostic companies navigate the messy intersection of coverage, coding, and reimbursement. Cox begins with a deceptively simple insight: before debating outcomes or cost-effectiveness, a company must ensure its product can be billed, adjudicated, and paid through existing payer systems. GENOSAN is, in essence, a pragmatic playbook for reimbursement readiness.
The framework divides the problem into three phases. The first is coverage, which means mapping every intended indication to existing payer policies and ensuring the clinical narrative of “medical necessity” matches the top use cases that payers already recognize. The second phase is coding, which focuses on identifying the right CPT, HCPCS, or PLA codes, aligning them precisely to what the product actually does, and ensuring that bundling rules do not inadvertently erase payment. The final phase is reimbursement, which entails aligning site-of-service rules, documenting denials and appeals, and building payer-by-payer issue logs that can be systematically resolved.
Cox’s method is as much about project management as it is about policy. He recommends short, focused sprints: a two-week audit to identify gaps, followed by a month-long implementation period to fix them. The GENOSAN checklist emphasizes documentation—template appeals, validation dossiers, and claim-level artifacts—so that AI companies can survive payer audits and scrutiny.
GENOSAN’s spirit is anti-theoretical. It exists because reimbursement failure for AI products often happens in the first mile: mismatched coding, ambiguous coverage language, and missing clinical documentation. Its goal is to build an operational foundation upon which the broader policy innovations of Parikh or Warshawsky can rest.
Parikh and Helmchen: Paying for Intelligence, Not Activity
In their 2022 paper, Paying for Artificial Intelligence in Medicine, Parikh and Helmchen argued that traditional “per-use” payment models are poorly suited to AI. Paying for every invocation of an algorithm—every image read, every prediction made—risks the same moral hazard that plagued fee-for-service medicine. Conversely, refusing to pay at all stifles innovation. Their challenge was to find middle ground: ways to pay for value rather than volume.
The authors proposed five complementary approaches. First, they recommended paying for outcomes, not clicks—for example, rewarding AI that shortens stroke door-to-puncture times or improves cancer detection rates. Second, they envisioned advance market commitments—government or payer pledges to purchase effective AI solutions to high-priority problems, analogous to vaccine incentives. Third, they suggested time-limited add-on payments, such as the New Technology Add-On Payment (NTAP) or its transitional cousin, TDAP, to encourage initial adoption while collecting data for long-term incorporation into bundled rates. Fourth, they called for interoperability and bias mitigation bonuses, rewarding tools that generalize across populations and institutions. Finally, they acknowledged that some separate payments may be necessary early on, but only with sunset provisions and rigorous evaluation.
Parikh’s framework operates at the level of policy imagination—it is a blueprint for incentive alignment, agnostic to clinical domain. But its implementation demands both operational rigor and institutional power. GENOSAN offers the first; Warshawsky, writing from the American Enterprise Institute, provides the second.
Mark Warshawsky: Updating Parikh through the Lens of CMMI
By 2025, the Center for Medicare and Medicaid Innovation had accumulated more than a decade of mixed results. Many pilots failed to deliver savings, and productivity growth in healthcare remained stagnant. In his AEI paper, Warshawsky argues that CMMI should explicitly use its authority to mandate adoption of cost-saving technologies—AI among them—and to capture the resulting savings for taxpayers. His focus is narrower but more forceful: turning CMMI from a passive pilot funder into an active productivity engine.
Warshawsky points to two CMMI models as levers for reform. The Ambulatory Specialty Model introduces mandatory two-sided risk for selected regions, compelling providers to share in both savings and losses. The Tech-Enabled Prior Authorization Model experiments with vendor compensation tied to avoided costs, targeting high-overuse services such as imaging. Across both, CMMI’s mantra is “risk required”: models must impose downside accountability, not just bonus opportunities.
Within this framework, AI adoption is not merely encouraged but expected. Warshawsky envisions a system where payment schedules and benchmark updates assume the use of productivity-enhancing tools, with providers who lag behind facing financial disadvantage. The shift is philosophical: innovation becomes a duty rather than an option. In his words, without explicit expectations to adopt cost-saving technology, the system will drift toward rationing and political intervention.
Bridging the Frameworks: From Ideas to Implementation
When viewed together, these three approaches form a continuous logic chain. Parikh supplies the why—the rationale for paying AI based on value. Warshawsky supplies the where—a real policy venue (CMMI) with the power to institutionalize those principles. And Cox supplies the how—a method for converting policy theory into billable claims.
Under a Warshawsky-style CMMI model, Parikh’s ideas could be implemented concretely. Outcome-based payments become part of model benchmarks, with shared savings tied to AI-sensitive measures such as sepsis mortality or cancer time-to-treatment. Time-limited add-ons evolve into internal, model-year carveouts that sunset after data collection. Interoperability and bias mitigation bonuses become eligibility requirements for participation, not just optional incentives. And advance market commitments turn into CMMI challenges focused on specific cost drivers, where winning tools are guaranteed pilot deployment.
Even Parikh’s concern about “AI overuse” finds a counterpart in Warshawsky’s vision: tech-enhanced prior authorization models can curb low-value AI use while rewarding algorithms that reduce unnecessary denials and expenditures.
At the ground level, however, all of this depends on GENOSAN-like discipline. CMMI models succeed only if participants can produce valid claims, consistent coding, and defensible documentation. Cox’s insistence on payer-specific logs, validation artifacts, and template appeals thus becomes the practical backbone of what Warshawsky and Parikh describe at the policy level. Without it, no model—however elegant—can function.
Conclusion: A Convergent Future
The GENOSAN framework, the Parikh-Helmchen model, and Warshawsky’s CMMI vision represent three tiers of the same emerging architecture for AI reimbursement. GENOSAN teaches companies to navigate the existing system with precision; Parikh reimagines how that system should reward intelligence rather than activity; and Warshawsky proposes how government can enforce those rewards through structural reform.
Together, they suggest a mature ecosystem in which AI reimbursement is not an add-on or novelty, but a disciplined process connecting micro-level operations to macro-level incentives. The future of AI payment, in this vision, depends as much on denial logs and CPT stacks as on economic theory and model design. It is a fusion of ground-level competence and high-level policy—exactly the combination the healthcare system has long needed to translate technological promise into lasting productivity gains.
Would you like this revised into a 1,200-word draft formatted for publication (e.g., for Discoveries in Health Policy), with citations and subheads styled in APA/house format?
##
PS. Too late to include, see Li et al. 2025 NJP Dig Med, Commercialization of medical AI here.
