Guest Post: Before You Build for Dual-Elgibles, Understand Who You're Building For

A friend asked me earlier this year to help her family navigate her mother-in-law's care¹. The mother had progressive cognitive decline and was placed in a memory-care community by the health plan’s care management vendor. On paper, she was about as well-covered as a dual-eligible can be: state Medicaid plus a group Medicare Advantage plan inherited from her late husband's retiree benefits, which is the exact non-integrated construct D-SNPs were designed to replace. Two payers, aiming to provide coordinated coverage, with long-term services support. The kind of architecture builders point to when they argue the dual system is fundamentally intact. By some measures, she was more fortunate than most older adults in this country.

Then the eviction notice came.

The plan denied retroactive authorization for the placement on the grounds that the member didn't meet a utilization threshold — a specified number of nursing facility days, ED visits, or hospitalizations in a defined window. She didn't meet that threshold because she had never been sent to a nursing facility. She had never been sent to a nursing facility because the plan's contracted vendors had placed her in a less costly memory care facility. One health plan vendor initiated the authorization process, while another vendor completed the placement, with the authorization never crossing between them, leaving the patient and their family in the middle. The benefit she had been placed under is a community support program designed to prevent precisely that utilization by routing members to a lower-cost community-based care setting before a nursing facility stay becomes necessary. The plan was denying coverage of the prevention because the prevention had worked as designed.

By the time the family engaged me, the facility had issued a displacement notice, the arrears were in the six figures, and the family was assembling their own State Fair Hearing packet — because the legal aid organization they had approached was unable to represent them due to capacity constraints. She had two payers. She had a managed care plan. She had a placement arranged by her health plan's care management vendors, yet that same health plan ultimately denied their own designated care team’s authorization request. The system she could not get to work for her was, by every measurable metric, working as designed.

The son was coordinating across the facility, the nursing agency, the managed care plan, and the ombudsman's office for his mom’s care. The daughter-in-law, my friend, was trying to hold a job, a small business, and her own family together. The mother-in-law, in moments of clarity, repeatedly asked whether the facility that her health plan had placed her in was about to evict her, and where she would go.

I had spent the prior two decades running Medicaid, Medicare Advantage, and dual-eligible P&Ls within Fortune 50 health plans. I could explain every form to them. But I could not, with a straight face, explain to the family why the system they were navigating, with both Medicaid and a Group MA plan in place, had been built this way. The insurance coverage was adequate, but the operating model had failed them anyway.

I am writing this piece because Duncan Reece's HTN series on the Medicare Advantage bid process is the foundational reading anyone selling into MA should understand, but almost none of it survives intact when you cross over into the duals book. Every concept Duncan covers applies to D-SNPs (Dual Special Needs Plans), because D-SNPs are MA products. But D-SNPs carry a second regulatory and operational layer that most MA-focused entrants have never encountered, and the people on the other end of it look more like the family I described earlier than like the average MA member.

The audience I am writing for is founders, plan executives, ACO operators, investors, and CBO leaders who are weighing whether to build something for this population, or who have already built it and are watching it not work as promised in the deck. I am not going to walk through every actuarial mechanism. Duncan and others can do that better than I could. I am going to walk through the things I wish someone had told me before I priced my first D-SNP product more than a decade ago.

There were roughly 13 million dually enrolled Americans in the most recent joint MedPAC/MACPAC data — about 20% of the Medicare population, 36% of Medicare spending; 14% of Medicaid enrollment, and 30% of Medicaid spending, with combined program spending of approximately $548.8 billion in contract year (CY) 2022. That is a large, complex market and most plans treat it as a margin opportunity rather than the operating-model problem it actually is.

But "dual-eligible" is not a single eligibility category. It is the intersection of Medicare enrollment with one of several Medicaid assistance levels: Full Benefit Dual Eligible, Qualified Medicare Beneficiary (QMB), Specified Low-Income Medicare Beneficiary (SLMB), Qualifying Individual (QI), and Qualified Disabled and Working Individual (QDWI). Each carries a different benefit package, a different cost-sharing structure, and a different relationship between Medicare and the state Medicaid agency. The member experience is not the same. The plan economics are not the same.

Qualifying for Medicaid at 65 is not automatic. It requires income and asset tests that vary by state, an enrollment process that was not designed for people with cognitive impairment or limited English proficiency, and a redetermination cycle that occurs every six or twelve months, depending on the state, which can disrupt coverage, care plans, and the trust that makes coordination possible at all. The family I described at the start of this piece had a mother who qualified for everything she needed. She was at risk of losing her care anyway because the operational machinery among the Medicaid agency, the managed care plan, the nursing service, and the facility lacked a shared definition of the level of care she was receiving — and no timely, functional way to resolve that disagreement before the bill came due.

The product side is just as fragmented. The D-SNP family alone runs across a spectrum:

1.     Coordination-Only D-SNPs.  Medicare and Medicaid continue to operate as parallel infrastructures with a coordination wrapper. The plan sponsor coordinates information flow and some clinical management; the financial accountability stays split between the two payers. This is still roughly 65% of D-SNP benefit packages nationally. It is also, per JAMA Network Open's 2025 study of dual-eligibles with dementia, the cohort that produces no statistically significant outcome improvement on preventable hospitalizations or 30-day readmissions versus non-D-SNP comparators.

2.     Highly Integrated D-SNPs (HIDE). A single plan sponsor holds both the Medicare and Medicaid contracts, but operationally, the two products often run on separate tracks under one roof. Better than coordination-only, but also not the same thing as full integration.

3.     Fully Integrated D-SNPs (FIDE). A single plan, a single member ID, a single care plan, unified accountability for Medicare and Medicaid spending, and, in CY2027, integrated health risk assessments. This is the cohort that produced the measurable effect in the JAMA study — a 1.2 percentage-point reduction in preventable hospitalizations, a 7.2 percentage-point reduction in 30-day readmissions for people with dementia. Both are meaningful effect sizes by value-based care standards, and there is still room to do better.  And, as an industry, there is a need to move away from financial alignment and double down on clinical alignment around specific diseases and patient cohorts and experiences that are addressable, such as those in dementia and their dyadic family caregiver, as well as a tighter performance network organized around dual outcomes over spend, not just on quality improvement process measures or risk adjustment opportunities.

4.     Applicable Integrated Plans (AIP). The newer designation CMS is using to tighten unified grievance and appeal processes for HIDE and FIDE plans. If you are building anything that touches member experience, this is where the operational requirements actually live for 2026 and beyond, accelerated by new FHIR prior authorization rules for both payers and providers.

Then there is everything sitting next to D-SNPs.

·       PACE (Programs of All-Inclusive Care for the Elderly), which uses county-locked blended benchmarks with frailty-adjusted risk scores and full-risk capitation (and a different risk adjustment model, which is mostly still v22 rather than the updated v28 model). PACE organizations are simultaneously the plan and the delivery system; their architecture is fundamentally different from that of a D-SNP plan sponsor, which layers Medicare capitation on top of state Medicaid rates and supplemental benefit funding. The revenue base often sits around 55% Medicaid, leaving it exposed to state budget cycles and changes in federal matching rates.  In many states, it requires de novo centers that cost roughly $12M+ in capital per site, take 24-36 months to break even, and require significant investments in finding qualified leads.  PACE serves about 1% of California's dual-eligible population, with about 3,000 lives per organization, whereas Medi-Medi Plans (California's exclusively aligned D-SNP construct) serve roughly 19% of the population under a completely different operating model. 

·       C-SNPs (chronic condition) and I-SNPs (institutional) layer on additional condition or setting requirements.

·       Look-alike plans — non-D-SNP MA plans whose enrollment skews toward dual-eligibles — are being squeezed out by tightened CMS thresholds and additional 2027 enrollment restrictions.

·       CMMI models layer on top of all of this. The Financial Alignment Initiative, which selected states like Washington participated in, was the first federal experiment in formally aligning Medicare and Medicaid for duals. LEAD (Long-term Enhanced ACO Design) is its successor logic on the ACO side, with two states piloting a dual-track approach that can layer on GUIDE for Dementia and other Medicare PMPM programs. 

I have watched smart, well-funded health systems and ACOs, specialty-focused and primary care teams, community hubs, and start-ups treat all this as if it were one product with different stickers on the box. The rate development process, risk structure, risk adjustment methodology, and contracting counterparty differ for each line on that list. PACE's frailty-adjusted full-risk capitation does not behave like a FIDE-SNP's blended Medicare-plus-Medicaid rate. Neither one behaves like a GUIDE PMPM. Mispricing one of these because you assumed it was like the others is how seed-stage companies disappear in their second year.  It is a complex insurance product with significant financial risk, which requires specialized sub-domain expertise and experience. Beyond the bid process, the second layer is the clinical and operational model required for success. 

If Duncan's MA bid series is your starting point, here is what changes when you cross into duals.

The Medicare side of a D-SNP bid still runs through the standard MA Bid Pricing Tool — base rates, supplemental benefits, MOOP, plan benefit package design, the Stars-rebate dance. All of that applies. The thing that does not exist in a non-dual MA bid is the State Medicaid Agency Contract (SMAC), the legally required document under 42 CFR 422.107 that binds the D-SNP sponsor to the state Medicaid agency. The SMAC is where the state defines what integration actually means in that state, what data it expects to receive, what care coordination requirements apply, what the state expects in terms of model of care alignment, and increasingly (i.e., in the better-run states), what outcomes the state expects to see for the dually eligible population.

The SMAC is also where most national plan sponsors over-engineer their organizational chart and under-engineer their operating model. A plan can pass procurement with a signed SMAC, but without a functional clinical model to identify and manage the state's priority populations, the gap doesn't show up until months after go-live. In California, that means a plan contractually committed to serving older adults with dementia and their caregivers but operating with mostly remote, telephonic enhanced care management programs that have low engagement rates, unspecified community health worker strategies, and chronic care management protocols that can't distinguish early-stage cognitive decline from medication non-adherence. The failure surfaces in grievances (families reporting their loved one was placed in a setting without family notification), appeals (denials of HCBS because the authorization template required nursing facility days the prevention program was designed to avoid), quality improvement audits showing missed HRAs for members with documented dementia diagnoses, and in the worst cases, sentinel events when a member with wandering risk is discharged to an unsupervised setting. By the time the state's quarterly quality compliance report flags the pattern, the plan has already assumed full financial and clinical risk for a population it cannot actually manage.

The state side adds its own machinery. State Medicaid managed care procurement cycles run independently of the federal MA timeline; nationally, Model of Care (MOC) submissions for CY2027 are due May 29, 2026, but the state-level SMAC negotiations behind them remain open. The states with the most active dual strategies — California, Rhode Island, Massachusetts, Minnesota, and New York — are using procurement language to push contract requirements beyond coordination-only and into segment-specific outcomes. Most other states are following, driven by budget pressure, but the pace is set by state contracting capacity, and that shows up in the operating model the plan is allowed to build.

The fiscal environment and healthcare margin compression are forcing plans and ACOs to get more strategic about where they can actually deliver value. As of the most recent earnings calls, most publicly traded health plans’ Medicare MLRs, including their duals book, are in the high-80s to low-90s. The CY2026 rate process completed the three-year phase-in of the 2024 CMS-HCC risk model, with the raw model phase-in netting roughly −4.3% before normalization offsets. Plans were ultimately held harmless by an unexpectedly strong FFS growth rate (5.06% net payment increase), but the underlying model trajectory remains compressive, and the H.R. 1 Medicaid provisions amplify the squeeze. Proposed federal matching rate reductions force states to cut supplemental rates and HCBS funding, thereby reducing the Medicaid component of the FIDE-SNP blended capitation. Meanwhile, expanded work requirements create eligibility churn, disrupting care continuity and increasing administrative costs without reducing clinical acuity or mitigating the financial risk on our federal, state, and local programs.

Sixteen years after the ACA built the first federal infrastructure for dual-eligible integration, a Health Affairs Forefront retrospective published this month concluded that the field is still not where it needs to be. That is a fair reading of the evidence. It also understates the problem.  And based on market developments, the next decade of value in this space is likely to be condition-focused, person- and caregiver-centered, and clear to both clinicians and actuaries. McKinsey's specialty value-based care work has shown that condition-specific models can reduce the total cost of care by 8 to 10 percent, and adoption is still limited in most high-cost specialties because the operating models are hard to build. The economics are sharpest where clinical complexity is highest. Combined per-user spending for dual-eligible LTSS users runs roughly $125,000 annually — about four times that for non-LTSS users — and the inflection point is not the acute event. It is the long-term services transition.

AI will likely deliver real value here, but only in clearly bounded use cases. High-volume, judgment-light administrative work, such as documentation, FHIR mapping, prior authorization narratives, integrated care team note structuring, risk adjustment coding cross-checks, and schedule optimization across community partners, all of which can be automated to reduce coordination overhead for both condition-specific ACOs and managed care MOCs. What AI cannot do is make the clinical judgment call that mattered in the opening story: recognizing that the authorization denial logic was internally contradictory, that the family was coordinating across vendors who weren't coordinating with each other, and that the eligibility framework itself was failing. That kind of pattern recognition still requires a human who understands both the contract language and the operational reality behind it. Done well, AI alleviates resource constraints that currently prevent plans from building adequate care coordination capacity, freeing nurses to focus on clinical judgment and direct care, rather than documentation and endless phone calls. Done poorly, it becomes a thin-wrapper, value-based care play - investing in tools that produce algorithm-driven auto-denials that overwhelm providers, add administrative costs, and ultimately are overturned when proper review is conducted, with plausible documents at scale without changing the total cost curve or improving the member experience. The current capital environment is making that distinction clearer.

Financial markets recognized the importance of portfolio discipline before many large health plans or ACOs did. For example, nobody who runs a mutual fund accepts whatever securities their advisor happens to buy. You start with a thesis and a logic model, set performance standards you report against, and sunset the underperformers so you can move capital into what is working. This is basic portfolio management, and almost none of it reflects how the Medicare and Medicaid innovation models have historically operated.  Former CDC Director Mandy Cohen put the health-related results measurement framework standard plainly in a recent public address: adoption is not a success metric. You define what success looks like before you launch, not after.

About 40 percent of 2025 digital health funding went to workflow and data infrastructure, the category least likely to move the cost curve for complex populations without a defensible theory of change. No tool is a strategy in itself. The durability test in the current capital environment is whether you can answer Cohen’s question: Why are you doing this, and are you prepared to lead the change it implies?

For the dual-eligible population specifically, that means:

·       Which sub-segment of duals are you serving, and what problem are you solving (quality/experience, access, cost)? Is it supported by your care model, network, product design, and member experience? Is it supported by emerging research evidence?

·       Which D-SNP variant or adjacent vehicle is your delivery construct?  Do you have the right incentives within your contract?  Do you have a substantial enough population and experience to take financial risk and reward others if they enable greater success towards the quadruple aim?

·       How are you financing your operating model?  Does it align with venture economics and scale requirements, or will it push toward compliance and quality issues that lead to CMS sanctions? 

·       What's your defensible health-related process and outcome metric? How will you measure it, and how will it show in the total cost of the member’s claims?

·       What's your sunset criterion?  How do you continue to improve in your investments and evaluations?

Capital allocators have gotten noticeably better at recognizing it, and the current fiscal environment (tight MCO margins, proposed risk adjustment reductions, Medicaid budget uncertainty) is making them better at it faster. Apply this test to whatever you are building: would it have changed the outcome for the family at the start of this piece?