Policy brief: AI-first Medicaid: how CMS can build a smarter safety net with Precision Benefits

Abstract

Medicaid serves over 70 million Americans, yet barriers to consistent, high-quality care endure due to workforce shortages, fragmented service delivery, and administrative burden. Artificial intelligence (AI) offers not just operational efficiency but the potential to transform the Medicaid care experience. AI-powered digital assistants can deliver 24/7 multilingual voice or text support, expanding access to personalized, emotionally-intelligent assistance. Under existing workforce supervision, these agents can bridge critical gaps in behavioral health and community coordination through tools like therapy chatbots that reduce loneliness and improve engagement. As “embedded staff” in provider offices and community organizations, digital assistants can create a unified infrastructure for whole-person care. We introduce the concept of Precision Benefits: delivering the right support to the right person at the right time to prevent avoidable health and social deterioration. This aligns with administrative and eligibility reforms in H.R.1, which require states to improve efficiency and verification while fostering innovation and preserving state authority over AI regulation. Realizing this vision demands responsible AI development – addressing safety, bias, privacy, and trust – and modernization of infrastructure and payment models. Yet the opportunity is clear: AI can power a smarter and more equitable Medicaid system, one that puts everyone on an upward life trajectory.

Medicaid serves over 70 million Americans but faces rising costs, workforce shortages, and fragmented delivery¹. Beneficiaries often struggle with language barriers, behavioral health needs, and unmet social needs. Meanwhile, artificial intelligence (AI)’s ability to process language, reason, and respond is rapidly maturing and becoming a native part of how people navigate the world.

One of AI’s clearest opportunities lies in reducing administrative overhead. With 15% of every healthcare dollar spent on administration, AI could lower U.S. healthcare spending by 5–10% – up to $360 billion annually. Scaled proportionally, Medicaid could save $37–$67 billion each year². Beyond administrative automation, the adoption of AI systems in clinical, operational, and population-health contexts can also yield substantial economic benefits such as reducing redundant testing, optimizing resource allocation, and improving care coordination – which in turn may generate downstream savings.

As AI reshapes industries from retail to finance, Medicaid must keep pace. Beneficiaries will expect on-demand, personalized, broadly useful, and multimodal (text, voice, video) experiences. Meeting this expectation will require investments in innovation, payment models, regulatory frameworks, and data infrastructure.

Digital assistants

We use the term AI-powered digital assistants to describe a class of agentic tools that accomplish tasks on behalf of users. These systems are designed to act with a degree of autonomy – initiating workflows or interacting with multiple systems to support care navigation based on user input or internal reasoning. They differ from rules-based chatbots by leveraging advances in large-language-model (LLM) reasoning, natural conversation, and general-purpose tool use. Emerging frameworks further expand agents to take real-world actions — for example, retrieval-augmented generation (RAG) for memory and knowledge grounding, and the Model Context Protocol (MCP) for connecting AI tools to external systems via application programming interfaces (APIs)³. Together, these technologies form the foundation for next-generation digital assistants capable of integrating across data sources and supporting more dynamic, human-centered care experiences.

AI-powered digital assistants offer Medicaid a scalable way to expand workforce capacity and enhance member engagement. These agents can operate 24/7 across voice and text, in any language, handling tasks like appointment scheduling and eligibility verification⁴.

The value of AI-powered digital assistants extends beyond administration. With real-time translation and culturally responsive design, they can help non-English-speaking members navigate care and connect with resources like food or housing⁵.

These tools also show emotional intelligence. In a randomized trial by Dartmouth researchers, users of “Therabot” – a generative AI mental health chatbot – saw a 51% drop in depression, 31% reduction in anxiety, and 19% drop in eating disorder risk⁶. Participants reported meaningful connections, showing AI’s potential to close behavioral health gaps in Medicaid.

Digital assistants substantially enhance access as they bolster the broader Medicaid ecosystem – community organizations, benefits agencies, and crisis lines like 211. A food bank employee, for instance, could use an AI assistant to help a client enroll in Medicaid, schedule an appointment, and arrange transportation – all in one interaction.

Precision Benefits

Digital assistants are not the end goal – they are the infrastructure that enables a more ambitious vision: Precision Benefits.

Predictive analytics may flag someone at risk but without real-time, tailored follow-up, these insights often go unused. That’s where digital assistants come in: available 24/7 and connected to benefits systems, they can send a personalized message, arrange transportation, alert a community organization, or help someone enroll in aid programs – all in the precise moment when action can prevent a crisis.

Consider a single mother living with high blood pressure and diabetes. She juggles caregiving and hourly work without paid leave or reliable transit. When her own mother falls ill, she misses shifts to help, loses income, and begins skipping appointments and rationing medication to afford rent. As her blood sugar and pressure climb, warning signs go unnoticed. Soon, she lands in the emergency room, her health destabilized, her job at risk, and eviction looming. But a missed prescription refill or appointment could have prompted an AI assistant to schedule a telehealth visit or offer other assistance. A few hundred dollars of support, offered at the right moment, could have prevented thousands in emergency care.

This is the promise of Precision Benefits: delivering the right resource, in the right dose, at the right time – before breakdowns become emergencies.

In prevention, these tools can flag early risks like missed screenings, rising Body Mass Index (BMI), or social isolation – then offer nudges or referrals to local programs. For chronic conditions, AI can support daily stability: reminding a member with asthma to refill inhalers, track symptoms, and stay indoors during high-pollution days. Just as clinicians adjust care based on patient needs, Medicaid can use AI to titrate support, shifting from reactive to responsive care.

Challenges to AI adoption in Medicaid

Despite its promise, AI adoption in Medicaid faces significant barriers – ranging from structural financing constraints to data limitations and workforce concerns. Historically, Medicaid has been hesitant to fund emerging innovations, even when cost-saving potential is evident⁷. The lack of clear reimbursement pathways and financial incentives discourages states from investing in AI tools. This challenge is exacerbated by Medicaid’s regulatory complexity and state-by-state variation in delivery models – from managed care to fee-for-service – requiring coordination across multiple levels of government^8,9.

Fairness and safety also remain major concerns. AI models trained on commercial insurance data may not generalize to Medicaid’s socioeconomically and racially diverse populations, risking biased predictions and unequal care¹. Generative AI introduces further risks – hallucination, inappropriate tone, and overgeneralization – if not properly constrained. Mitigation strategies include RAG to ground outputs in verified data, the use of synthetic data to augment underrepresented populations, and reinforcement learning from human or expert feedback to align model behavior with fairness and empathy standards¹⁰. Rigorous pre-deployment testing – supported by LLM-as-a-judge evaluation frameworks – can ensure transparency, reliability, and fairness before integration into care.

Concerns about AI displacing human jobs deserve attention, but Medicaid already faces a structural mismatch between service demand and workforce supply, especially in long-term services and supports (LTSS) and home- and community-based services (HCBS). AI should be viewed not as a replacement but as a tool to augment human capacity, by automating routine tasks, triaging cases, and freeing professionals to focus on hands-on, relationship-based care.

Technology capacity poses another constraint. Many Medicaid agencies lack the in-house expertise and infrastructure to deploy and govern AI tools. Fragmented data systems and limited interoperability hinder integration with healthcare and social service networks. Trust is also a barrier, due to fear that data sharing may lead to surveillance. Addressing these challenges will require investment in secure, interoperable systems, alongside transparent governance and community engagement.

Finally, AI systems in Medicaid may fall under FDA oversight, depending on their intended use. Software that diagnoses, predicts, or recommends treatment for an individual is generally regulated as Software as a Medical Device (SaMD) and must follow a premarket review pathway (510(k), De Novo, or PMA). In contrast, the 21st Century Cures Act excludes several non-clinical functions from the medical-device definition, and FDA’s Clinical Decision Support (CDS) guidance recognizes a category of non-device CDS when four criteria are met: it is healthcare provider-facing, shows its inputs, supports rather than replaces clinical judgment, and allows clinicians to independently review the rationale for a recommendation. As a result, digital assistants limited to navigation, eligibility checks, scheduling, messaging, reminders, or care-coordination logistics – without patient-specific diagnostic or treatment inference – will typically fall outside FDA jurisdiction¹¹. If an assistant generates or prioritizes clinical diagnoses or therapies in ways clinicians cannot independently verify, it becomes a regulated device¹².

For AI functions that do qualify as devices, the FDA emphasizes a total product lifecycle approach that pairs premarket review with ongoing real-world performance monitoring. To accommodate iterative model updates, FDA uses Predetermined Change Control Plans (PCCPs), submitted at the time of review, to define allowable changes, validation methods, and monitoring requirements. This enables adaptive AI to evolve safely without repeated submissions while maintaining assurances of performance.

Accordingly, Medicaid digital assistants should be intentionally scoped to non-device functions such as administration, navigation, and logistics. When a use case requires the assistant to support a clinical decision, design should aim to meet CDS criteria. If future versions incorporate patient-specific diagnostic or therapeutic recommendations, they should be treated as SaMD and include a PCCP and commit to local postmarket performance monitoring. This approach reflects current FDA direction and the agency’s emphasis on trustworthy AI that improves patient outcomes¹³.

Solutions: a strategy for AI-driven Medicaid transformation

To advance AI adoption in Medicaid, the federal Centers for Medicare and Medicaid Services (CMS) must take a leadership role by fostering innovation, ensuring regulatory clarity, and promoting state-level experimentation. CMS could launch an AI Innovation Challenge to fund pilot projects that enhance administrative efficiency, streamline eligibility, and improve care coordination. A State Medicaid Director’s Letter outlining best practices would offer implementation guidance, e.g., payment for digital assistants under Medical Loss Ratio versus Administrative Loss Ratio, while CMS provides the necessary financial and technical support.

States should create AI Centers of Excellence¹⁴, modeled in part after the Centers of Excellence in Regulatory Science and Innovation (CERSI) for medical devices and pharmaceuticals, to serve as dedicated hubs bringing together policymakers, technology developers, healthcare providers, and health plans to evaluate AI in real-world Medicaid settings. Priorities could include predictive analytics for early detection of health and social risk, AI-assisted care navigation and outreach, and public-private partnerships for scaling innovation.

Concurrently, AI tools should undergo bias audits and benchmarking to ensure fairness in access and outcomes. Training models on Medicaid-representative data improves accuracy, but for large language models, fairness also depends on safeguards like human-in-the-loop review, output constraints, and prospective monitoring for differential impacts across populations. We propose that Medicaid AI systems be designed with clear fallback to human support pathways, transparent explanation of reasoning, user feedback loops (e.g. “I’m not satisfied, take me to human”), and ongoing user trust monitoring (e.g. periodic surveys, trust metrics). We also suggest pilot user testing emphasizing acceptability and trust measures before scaling.

CMS leadership will be critical in realizing this vision. Beyond pilot programs and operational improvements, CMS can signal a bold ambition: to establish Medicaid as a leading example of an AI-first public service.

Conclusion

With targeted investments and strategic leadership, Medicaid stands poised to lead the nation into a new era of AI-powered, Precision Benefits – where care is not only more efficient but more human, proactive, and accessible. By harnessing AI to deliver personalized, timely, and appropriately scaled support, Medicaid can break down longstanding barriers to access and improve health outcomes for millions of vulnerable Americans. This transformation requires thoughtful stewardship – ensuring safety, fairness, transparency, and collaboration across federal, state, and private sectors. In embracing this vision, CMS and its partners have a rare opportunity to set a global example for how public programs can leverage intelligent technologies to better meet the needs of every individual.

Author contributions

N.F. and N.B. conceived the idea for the article and drafted the initial version. M.M. and R.D. contributed to revising and refining subsequent versions. All authors have approved the final version of the manuscript.

Data availability

No datasets were generated or analyzed during the current study.

Competing interests

The authors declare no competing interests.