The New CMMI Playbook
Mandatory Risk, Prevention, Competition
Two months after terminating four payment models, CMMI released the formal architecture of what replaces them. On May 13, 2025, CMS Administrator Mehmet Oz and CMMI Director Abe Sutton hosted a webinar and published a white paper, frequently asked questions document, and updated strategy page titled “CMS Innovation Center Strategy to Make America Healthy Again.” The materials established a three-pillar framework that will govern every new model CMMI designs, every existing model it evaluates for continuation, and every certification decision it makes for the rest of this administration’s tenure.
The strategic pivot from the Biden era’s five-pillar framework, which centered on health equity, multi-payer alignment, and systemic transformation, to a three-pillar structure anchored in prevention, patient empowerment, and competition is not merely cosmetic. The foundational principle that underpins all three pillars, stated explicitly in every document CMS published that day, is taxpayer protection. That language has specific statutory meaning. Section 1115A of the Social Security Act requires the Secretary to certify that any model expanded beyond its test phase either reduces net program spending or improves quality without increasing spending. The May 2025 strategic refresh made that certification requirement the operative design standard for all CMMI activity, not just the threshold for expansion decisions.
Understanding the playbook requires understanding why the prior approach failed, what the three pillars are designed to accomplish, what mandatory risk means in practice, and what the BPCI-A episode reveals about the limits of voluntary design.
The Case Against the Prior Approach#
The May 2025 documents were released two weeks after a letter from Republican members of the House Ways and Means Committee, dated April 28, 2025, sent to Administrator Oz and Director Sutton. That letter served as the congressional framing document for the new strategy.
Committee Chairman Jason Smith and Republican members wrote that they were “concerned with the Center’s history of developing costly models that either fail to meet or are not on track to meet that standard which is rooted in statute.” The letter cited the Congressional Budget Office’s September 2023 finding that CMMI’s activities had increased direct spending by $5.4 billion between 2011 and 2020, attributing the failure in part to the Biden administration’s 2021 strategy refresh, which the letter characterized as focusing on “a politically motivated health equity agenda” that “minimized the importance of cost savings in models.” The letter called on CMMI to return to its statutory purpose, focus exclusively on payment models that save money, improve transparency with affected providers before making model changes, solicit stakeholder feedback during model design, and give renewed attention to rural and underserved communities.
The letter did not prescribe specific models or design features. But its framing aligned precisely with what Oz and Sutton presented two weeks later: a center that prioritizes certifiable savings above all other objectives, structures participation requirements to eliminate the selection bias that has historically undermined voluntary model results, and embeds prevention and competition as the positive-sum mechanism through which savings are generated rather than simply extracted from providers.
Sutton was direct on the webinar about what had changed. CMMI was departing from the Biden-era goal of having all fee-for-service Medicare beneficiaries in accountable care relationships by 2030. The goal itself was not being abandoned, but the metric was being replaced. Rather than maximizing ACO enrollment as a participation benchmark, CMMI would focus on certifiable savings and model quality. In Sutton’s words, CMMI would “work expeditiously toward the future of health” — not toward the future of enrollment counts.
The Three Pillars#
The strategic refresh organized CMMI’s forward agenda into three interrelated areas, each with distinct policy objectives and design implications for new models.
Pillar One: Promote evidence-based prevention. CMMI committed to embedding disease prevention, chronic condition detection, and functional health maintenance into every new model it tests. The emphasis is specifically on conditions that drive Medicare cost trajectories: chronic disease in older adults, cognitive decline, functional deterioration. The prevention pillar is where the MAHA political alignment is most visible. Nutrition, physical activity, lifestyle medicine, and behavioral health are all explicitly identified as relevant intervention domains. CMMI signaled that it will reimburse for outreach and engagement activities, not only for direct patient care, a departure from the traditional CMS reimbursement logic that has historically paid only for clinical services tied to provider encounters. The MAHA ELEVATE model, which is the most direct expression of this pillar, tests a reimbursement structure for lifestyle and functional medicine interventions that have no established Medicare fee schedule pathway.
Prevention is also described as a mechanism for generating certifiable savings through upstream cost avoidance rather than downstream utilization management. If Medicare can reduce the incidence of hospitalizations by funding effective chronic disease prevention at scale, the savings pathway is different from — and potentially more durable than — the pathway of restricting specific high-cost services. This is the conceptual claim behind ACCESS, BALANCE, and MAHA ELEVATE as a coherent portfolio: invest in prevention and chronic disease management, generate downstream cost avoidance, certify that avoidance as savings. Whether the certification logic holds depends on the models’ evaluation designs, which are not yet fully specified.
Pillar Two: Empower people to achieve their health goals. This pillar encompasses patient access to health data and decision-support tools, financial incentive alignment between what is good for patient health and what is reimbursed, and care delivery flexibility that allows accountable entities to use waivers and benefit design modifications in service of patient-centered outcomes. CMMI specifically flagged two potential waiver structures: allowing ACOs that assume global risk to provide durable medical equipment that bypasses National Coverage Determinations if the equipment supports transitions to or maintenance of home-based care, and enabling reduced cost-sharing for preventive or high-value services provided to caregivers managing beneficiaries with cognitive or functional decline.
The data access dimension of this pillar is where CMMI connects to the broader administration technology and AI strategy. The center committed to leveraging data sharing for both patient-facing health behavior support and provider-facing care coordination. WISeR is partly an expression of this pillar: AI-powered prior authorization as a data-driven mechanism for reducing low-value care, rather than as a utilization management tool operated by intermediaries with conflicting incentives.
Pillar Three: Drive choice and competition. The competition pillar is where the most operationally significant model design commitments appear. CMMI committed to reducing administrative burden on providers participating in models, expanding participation to provider types that have not historically entered CMMI programs — independent practices, rural clinicians, community-based providers — and promoting site-neutral payment structures across care settings. Site neutrality, which would require that identical services be reimbursed at identical rates regardless of whether they are delivered in a hospital outpatient department, an ambulatory surgery center, or a physician office, is a long-standing MedPAC recommendation and a Republican priority that has never been legislated at scale. CMMI’s framing treats it as achievable through model design: models that require or incentivize site-neutral payment as a participation condition.
The competition pillar also encompasses CMMI’s stated intent to engage providers who are currently low performers, high-cost and low-quality, in value-based models. This is where the mandatory participation logic is most explicit. Voluntary models attract high performers who expect positive outcomes. Mandatory models include the full performance distribution. Sutton has acknowledged that getting lower-performing providers into accountable care arrangements is the specific problem mandatory design is intended to solve, and that voluntary models, structurally, cannot solve it.
Why Mandatory Models: The Selection Bias Problem#
CMS’s own explanation of why mandatory models matter, published on its voluntary-vs-mandatory participation explainer page, identifies the core issue precisely. Voluntary models are susceptible to selection bias: “only groups that think they can lower costs may choose to participate. And, if they find that the financial outcome is different than expected, they may withdraw from the model. Selection bias can skew evaluation results such that a model’s total cost savings and other outcomes may reflect participant readiness more than the model’s potential impact.”
This observation has been made about CMMI’s voluntary portfolio for more than a decade. The innovation center has known the problem exists. The Obama and Biden administrations chose to live with it because mandatory models generate political resistance from provider communities that prefer optionality, and because the ACA’s design for CMMI did not assume mandatory participation as the default.
What changed is not the diagnosis. It is the appetite for the political cost. The current administration is willing to impose mandatory participation requirements because the certification imperative requires it. A model that cannot be evaluated without selection-bias-free counterfactuals cannot generate certifiable savings that survive CBO scrutiny. And a model that cannot generate certifiable savings is, under the current framework, not a model worth running.
CMMI proposed three mandatory models in 2025, which CMMI Director Sutton described as the highest number in a single year and representing one-third of all models proposed that year. The three are WISeR (mandatory AI-powered prior authorization for FFS Medicare), GLOBE (mandatory Part B drug rebate program), and GUARD (mandatory Part D drug rebate program). Two are drug-pricing models that operate through rebate mechanics rather than provider participation requirements. WISeR operates through mandatory contractor engagement in selected geographic areas rather than mandatory provider choice. None of the three requires the kind of provider-level mandatory participation that generated the most intense opposition to TEAM and earlier bundled payment proposals.
The ASM model, finalized in the CY 2026 Physician Fee Schedule final rule, adds a fourth mandatory model. ASM selects physicians treating heart failure and low back pain in ambulatory settings in specific geographic areas and requires them to participate in an episode-based payment structure that retains a portion of Part B reimbursements — 1.35 percent in year one, rising to 1.80 percent — as a built-in mechanism for generating savings regardless of individual physician performance. That design feature guarantees CMMI a return, which creates a structural pathway to certification that does not depend entirely on provider behavior change.
Sutton’s comment that “mandatory models are going to have to be part of the equation” reflects a settled administrative position, not a proposal under consideration. The question is not whether CMMI will continue to push mandatory models but whether the models it designs can achieve certification on a timeline relevant to the trust fund clock.
The BPCI-A Lesson#
The Bundled Payments for Care Improvement Advanced model is the most instructive cautionary case in CMMI’s portfolio for understanding both what voluntary models can and cannot do and what the current administration is trying to avoid replicating.
BPCI-A launched in October 2018 as a voluntary, two-sided risk episode payment model for 32 clinical episode categories, including lower extremity joint replacement, cardiac episodes, and spinal surgery. Participants included hospitals, physician groups, and post-acute care providers. Unlike many prior CMMI models, BPCI-A required downside risk from all participants: those who generated episode costs above target prices owed money back to CMS. That feature made it analytically superior to upside-only models.
The problem BPCI-A illustrated was not the absence of downside risk but the structure of voluntary participation with episode selection flexibility. When CMS modified the target price methodology and tightened the conditions under which participants could select which clinical episode categories to include in their risk portfolio, a large share of participants and their conveners exited the program. The episode categories that remained were disproportionately those where participants expected favorable outcomes under the new pricing. The result was a model that looked, from outside, like it was generating savings while participants were actually managing their portfolio to preserve the favorable episodes and drop the unfavorable ones. Net savings to Medicare were real in some analyses but substantially smaller than gross savings figures suggested, precisely because the participant population had self-selected toward episodes where they were confident of performing well.
The practical implication for current CMMI design is that voluntary participation with episode or service selection flexibility, even with downside risk, is insufficient to generate savings that are certifiable at scale. Mandatory models with mandatory episode inclusion are the structural response. TEAM, finalized in the FY 2025 IPPS rule for a January 2026 launch, applies this lesson directly: selected hospitals in randomly-chosen geographic areas are required to participate in specified surgical episode categories for the full five-year model period, with no ability to exclude unfavorable episodes from their participation. The mandatory geographic selection and mandatory episode scope eliminate the portfolio management behavior that undermined BPCI-A’s savings credibility.
Whether TEAM succeeds in generating certifiable savings will depend on execution — particularly on whether the hospitals in selected geographic areas can actually transform care delivery within the episode timeframe, or whether mandatory participation produces resentful compliance without the care redesign investments that generate genuine cost reduction. That is the central unresolved question for mandatory model design: whether the participation requirement that eliminates selection bias also eliminates the motivation for genuine transformation.
The 2030 ACO Goal: Abandoned or Refined?#
The Biden administration’s stated goal of having all FFS Medicare beneficiaries in an accountable care relationship by 2030 was an enrollment target, not a savings target. It reflected a theory that broad ACO coverage, even without downside risk, would generate coordination improvements and utilization reductions that would compound over time.
Sutton explicitly abandoned that framing in the May webinar. CMMI is not pursuing enrollment maximization. It is pursuing certification. Those are not the same objective, and the distinction matters for how MSSP growth is evaluated.
Under the current framework, MSSP’s 511 ACOs covering 12.6 million beneficiaries is not a success metric primarily because of its scale. It is evaluated on whether those ACOs are generating net savings, whether the participation rate in two-sided risk tracks is growing, and whether the program can be certified for expansion. MedPAC has found that MSSP generates net savings; CMS’s own PY2024 results showed $2.5 billion in net savings after shared savings payments. Those results are the basis on which MSSP survives in the current framework — not its enrollment share of FFS beneficiaries.
The signal embedded in abandoning the 2030 enrollment goal is directed at the ACO REACH program and the LEAD model. ACO REACH operates under global and professional risk structures that require genuine downside exposure. LEAD, the successor model being designed for a 2027 launch, extends the ACO framework with longer time horizons and stronger prevention and chronic disease management requirements. Both are positioned as the certifiable-savings pathway for accountable care, with MSSP providing the permanent statutory program infrastructure alongside them.
What is not positioned as a certifiable-savings pathway is upside-only ACO participation for providers new to value-based care. The Biden administration’s strategy of lowering the on-ramp cost for ACO entry through the ACO PC Flex model and the Advanced Investment Payment program for new low-revenue ACOs has not been reversed, but it is not the center of gravity. Those programs remain available. They are not the models around which the new CMMI playbook is organized.
The Site Neutrality Signal#
The competition pillar’s most consequential design implication for the hospital sector is the site neutrality commitment. CMMI’s May 2025 materials state explicitly that “model reviews and new model designs could reduce the role of state government in rate setting for health care services” and could “reinvest hospital capacity in outpatient and community-based care by changing uncompetitive certificate-of-need requirements, presumably in coordination with state regulators.”
Certificate of need laws, which in many states require regulatory approval before health care facilities can add services or capital equipment, have long been criticized by economists and Republican health policy advocates as barriers to competition. CMMI’s framing treats CON reform as a competition-pillar objective achievable through model design: states participating in AHEAD and Geo AHEAD may face requirements to modify their CON frameworks as a condition of model participation. The AHEAD model already includes a requirement that participating states adopt two policies from a menu that includes CON reform, any-willing-provider requirements, and antitrust enforcement enhancements.
Site neutrality as a mandatory model feature is administratively complex and legally contested. CMS has authority to test payment approaches through CMMI that differ from Medicare’s standard payment rules, but it cannot administratively impose site-neutral payment on provider categories outside a model’s mandatory participation scope without rulemaking. The CMMI pathway is to build site neutrality into mandatory model participation agreements and generate evaluation data that supports legislative or regulatory expansion. That is a multi-year process; it is not an immediate operational change. But the directional commitment, embedded in the foundational strategy document, signals that site neutrality is a CMMI design objective for every new model going forward.
What the Playbook Requires of Participants#
The shift from the prior CMMI framework to the current one has practical implications for every category of stakeholder that interacts with CMMI models.
For health systems and hospitals, the playbook signals that voluntary ACO participation without downside risk is a transitional state, not a sustainable posture. Every new and modified model will include or expand downside risk requirements. TEAM is already mandatory for selected hospitals. AHEAD imposes global budget accountability on participating health systems. The LEAD model will require genuine risk-bearing from participating ACOs in ways that ACO PC Flex does not. Health systems that have been participating in CMMI models under upside-only structures since 2012 are approaching the end of that option.
For specialist physicians, the ASM is a direct statement of intent. CMMI identified heart failure and low back pain as the initial ASM categories because they are high-cost, high-prevalence, and relatively well-understood from a care redesign perspective. If ASM generates certifiable savings in those conditions, the design can be scaled to additional specialty areas without new legislation. Specialists who have historically been outside the CMMI orbit now have a model designed for them, and the word “mandatory” is in its participation structure from day one.
For digital health companies, community health organizations, and non-traditional providers, the prevention pillar and the patient empowerment pillar open doors that have been historically closed. CMMI’s explicit commitment to reimbursing outreach and engagement activities, to including non-traditional provider types in model eligibility, and to using waivers to enable novel care delivery arrangements represents the most direct policy runway for digital health and lifestyle medicine participation in Medicare payment that has existed since the program began. ACCESS and MAHA ELEVATE are the initial expressions of this opening; they will not be the last.
For state governments, the AHEAD participation requirements and the site neutrality language represent CMMI reaching into state health policy prerogatives in ways the prior strategy did not. States that want federal AHEAD global budget payments must accept federal policy conditions that constrain their regulatory authority over health markets. That is not a condition every state will accept. The number of states that ultimately join AHEAD will be shaped partly by whether the program’s financial benefits are sufficient to offset the political cost of CON reform, antitrust enforcement, and any-willing-provider requirements that CMMI’s model design packages together.
The Certification Race#
The practical test of the new playbook is whether it generates more certified models in less time than the prior approach. CMMI has certified four models for expansion across fifteen years. Under the current framework, the standard for what counts as a certifiable model has been tightened, the timeline for achieving certification is under political pressure, and the trust fund math creates urgency that the 2011 through 2024 model portfolio never faced as directly.
Sutton acknowledged, at an industry conference in early 2026, that the bar for certification is high and that he views that as a feature rather than a problem. “For all the knocks that different people have had on the center over the years, you cannot say we have tried to certify too many models,” he said. “We have a high standard. I would like to pursue certification on more models, but I think having that standard is a good thing.”
That is a defensible position in isolation. It is harder to defend in the context of a trust fund depleting in 2033 and a model portfolio that includes ten new models announced in 2025, most of which are two to three years away from producing evaluation data that could support certification decisions. The mandatory savings approach works if the mandatory models generate savings. Whether WISeR, GLOBE, GUARD, ASM, TEAM, and ACCESS collectively certify savings before the 2028 political horizon is an empirical question with consequences that extend well beyond the model evaluations themselves.
Related Reading#
MCR-00_01 The Trust Fund Clock MCR-04_01 Is MA Still Worth It? The Strategic Recalculation for Insurers MCR-05_01 The Provider’s New Reality: Revenue, Authorization, and Accountability