Medicaid reimagined: Scaling trusted AI for 2030

Session overview

AI is moving quickly in Medicaid. For most states, the question is no longer whether AI can help, but how to put it to work in ways that are practical, governed and defensible.

At the Medicaid Enterprise Systems Conference (MESC) 2026, Karl Schelhammer, PhD, and James Lukenbill, PhD, of Optum State Government Solutions joined Martin Gavin, MS, MBA, of Carahsoft to explore real-world AI applications and what it takes to scale them responsibly as states plan for Medicaid in 2030. Here are the highlights.

Why AI matters for state health programs

The session started with pressures states already know well: thousands of rules, policies and data points to manage; staff stretched thin by manual compliance and reporting; language and access barriers for vulnerable populations; and growing pressure to show better health outcomes.

The vision presented was not AI for its own sake. It was AI that’s accessible and purpose-driven: natural language interfaces that open up data to more people, agents that assist teams with repetitive workflows and insights that update as conditions change.

Using AI to make modernization more attainable

The session showed how AI is changing delivery work. AI tools help turn business and technical context into requirements, user stories and work plans, and coding assistants generate code, tests and documentation, with humans guiding and verifying at every step.

One application is legacy data migration. Manually rewriting legacy ETL code for modern cloud platforms has delayed access to advanced analytics for many state data warehouses.

An AI-assisted migration accelerator translates that code and produces the artifacts teams need to validate the work, including mapping documents, unit tests and target-system schemas. This approach can lower migration project costs by up to 50%.

Agentic analytics: Putting data in the hands of program experts

Nearly every major data platform now includes an AI assistant. The session made the case that for Medicaid, what sets a tool apart is the governed layer around it: auditable query trails, role-based access and findings that can hold up in a fair hearing or audit.

Two agentic capabilities were demonstrated using illustrative data, both designed to put data access in the hands of the subject matter experts who know the program:

• Agentic Investigator: Program integrity staff ask a question in plain English. Coordinated agents turn policy into detection logic, screen the full claim population rather than a sample, and assemble a case narrative. Every finding includes its query trail, source data and written rationale, and a named investigator makes the decision.
• Agentic Research Analyst: Policy and quality teams pose a research question and agents build the cohort, run the statistics and draft a reproducible study in hours instead of months. A named analyst validates the conclusions.

Voice AI for assessments and contact centers

The session also looked at how voice-enabled AI can help reduce administrative burden while keeping people in control.

• Voice-based assessments: With the beneficiary’s permission, AI captures the assessment conversation and fills in the right form fields so the assessor can focus on the person. The assessor reviews and validates every entry before submitting, and the collected information is deleted afterward.
• Real-time language translation: When a bilingual contact center agent is not available, an AI translator can join the call and translate in real time, helping members and providers get answers without waiting for an interpreter.

Building a trusted AI strategy

Capability is only half the story. The session laid out five foundations for responsible AI in state health programs:

• Governance first: Establish a framework aligned to the NIST AI Risk Management Framework, with clear roles, accountability and risk controls before scaling.
• Transparency and explainability: Every AI-supported decision should produce a human-readable rationale and audit-ready evidence.
• Bias and equity safeguards: Test for disparate impact across populations and provide language access and ADA/WCAG compliance.
• Interoperability and standards: Design for CMS FHIR APIs and T-MSIS data quality from day one and integrate with MES modules rather than building silos.
• Lifecycle risk management: Monitor continuously for drift, security and performance, with playbooks for rollback and remediation.

Turning principles into procurement

The session also tied governance to procurement, an often overlooked step. States can use the buying process to evaluate security, data use and model risk; set expectations for transparency, audit access, performance and human decision rights; and require monitoring, remediation and exit provisions. Carahsoft helps agencies map those needs to available contract paths and align the right technology partners.

For states just getting started, Martin’s advice is to begin with clear AI use policies. They give business units across an agency consistent guidelines for purchasing, which can help secure better pricing, and they set guardrails so any AI acquired complies with the agency’s IT security policies.

The bottom line

Scaling AI in Medicaid depends less on the model and more on the foundation around it. States that pair practical use cases with strong governance, transparency and human accountability will be well positioned to turn AI into lasting value by 2030.