AI in action: Practical use cases for state health programs

Today, state health programs face serious challenges:

• Rising complexity in Medicaid operations and compliance
• Workforce strain and administrative burden
• Urgent need for equity and better patient/provider experiences

To address these challenges and transform how state health programs operate, AI is critical. Three pillars form a foundation for applying AI to achieve smarter health care.

1. Democratized data access: Natural language interfaces make accessing insights and completing tasks simple for everyone, from policy leaders to clinicians
2. Autonomous workflows: AI agents reduce risk and manual effort by automating workflows for compliance, fraud detection and reporting
3. Dynamic insights: Dashboards adapt in real-time to user preferences, surfacing actionable intelligence to proactively help improve quality of care

This foundation facilitates accessible, purpose-driven AI to deliver better outcomes for patients and providers.

Present methods for finding and intercepting fraud, waste and abuse (FWA) are often manual, siloed and reactive. AI can augment the work of human analysts by:

• Automating FWA detection workflows
• Fusing rules, machine learning and graph analytics for higher precision
• Using large language models (LLM) to create explainable evidence cards and draft outreach messages

With an intelligent payment integrity platform, specialized fraud analytics agents link data sources to analytics results, allowing human FWA analysts to migrate from reactive to proactive oversight.

Capabilities like the tools below make it easier for states to leverage AI, regardless of cloud provider:

• Query tool: Translates natural language questions to structured query language (SQL)
• AI help: Retrieves data using natural language questions
• Policy to SQL: Generates SQL for uploaded policy documents
• AI analytics & backstory: Provides background and summarizes an analytic using LLM

Today’s health care assessment interviews are manual, time-consuming processes. AI enhances the human role by automating tasks to reduce clinician burnout and improve productivity.

1. (Human): Assessor begins assessment & beneficiary grants permission to record.
2. (AI): AI tool records assessment. Appointment. AI speech-to-text tool is prompted by the assessor to record the conversation between assessor and beneficiary for each section.
3. (Human): Assessor begins dialogue around specific domain
4. (Human): Beneficiary responds to dialogue.
5. (AI): AI tool inputs information to system form fields. AI uses determining logic to identify the correct field or text box(es) for the beneficiary response to be entered.
6. (Human): Assessor reviews and validates AI-populated form.
7. (AI): AI updates system based on validated information. AI reviews assessor’s updates and reflects changes within the form.
8. (Human): Assessor closes out and submits assessment.
9. (AI): AI permanently deletes beneficiary information. After submission, AI discards all collected information from assessment.

For contact centers, AI removes common friction points like language barriers, long hold times and manual routing to achieve more equitable health care communication.

Responsible AI is more than a compliance checkbox — it underpins trust, fairness and sustainability in health programs.

By framing governance and safeguards as part of a broader commitment to ethical innovation, states can ensure that AI solutions not only meet technical standards but also uphold public confidence and deliver equitable outcomes.

As Karl Schelhammer notes, “Strong governance is critical. It lays the foundation for successful AI — aligned to frameworks like NIST AI RMF — so decisions remain transparent, explainable and fair.”

These key considerations help ensure responsible AI development within your state health programs.

• Goverance first: Establish an AI governance framework aligned to NIST AI RMF. Define roles, accountability and risk controls before scaling.
• Transparency and explainability: Every AI decision should produce human-readable rationale and audit-ready evidence. Supports fair hearings, public records and compliance review.
• Bias and equality safeguards: Test for disparate impact across populations. Provide language access and ADA/WCAG compliance for all outputs.
• Interoperability and standards: Design for CMS FHIR APIs and T-MSIS data quality from day one. Avoid siloed solutions — integrate with MES modules and state data ecosystems.
• Lifecycle risk management: Continuous monitoring for drift, security and performance. Incident response playbooks for rollback and remediation.