Policy & Practice | Spring 2026

4. Prioritize Use Cases With feasibility and risk understood, agencies should prioritize use cases using a structured scoring approach. At a minimum, this should include: n Expected business value and impact on service delivery outcomes n Technical feasibility within the existing environment n Level of risk, including compliance and operational considerations This allows agencies to compare opportunities consistently and focus on those that offer the greatest benefit with acceptable risk. Starting with smaller, lower risk use cases that maintain human oversight can help build confidence and dem onstrate early value before expanding into more complex applications. Selected use cases should be imple mented as pilot projects in controlled environments before being intro duced into high-volume operational workflows. Pilots allow agencies to validate assumptions, assess performance, and gather feedback from users who will interact with the system on a daily basis. They should include mechanisms for users to report issues and provide input, as well as clearly defined metrics to evaluate success. Considerations such as transpar ency of outputs, data security, and privacy protections should be evalu ated throughout this phase. The goal is to test, learn, and refine solutions before scaling them more broadly across programs. Full implementation should include extensive user testing and ongoing engagement with staff to ensure align ment with real-world workflows. AI systems require continuous oversight. Agencies should monitor per formance, evaluate user satisfaction, and conduct regular audits to ensure that systems remain accurate, fair, and aligned with policy requirements. 5. Pilot and Test AI Solutions 6. Implement and Continually Monitor

This includes ongoing assessment of potential bias, adherence to estab lished principles, and measurement of impact on service delivery outcomes. Continuous monitoring allows agencies to refine models over time, address issues proactively, and maintain trust in AI-supported processes.

Agencies must assess whether the necessary data exist and are suitable for training models. n Technology: Infrastructure readi ness, integration requirements, and model maturity all influence feasi bility, particularly when connecting to existing eligibility and case man agement systems. n Skill Set: Successful adoption requires training and potentially new capabilities, whether developed internally or through partnerships. n Cost: Total cost should be consid ered, including implementation, training, and ongoing maintenance. This step ensures that agencies focus on use cases that are both prac tical and sustainable, rather than investing in solutions that cannot be effectively supported. 3. Assess Risk of Incorporating AI to Address the Business Problem Before moving forward, agencies must explicitly evaluate the risks associated with each use case. This is especially important in human services environments, where decisions directly impact access to essential benefits. Key considerations include: n Potential bias: AI systems can reflect or amplify biases present in under lying data. Understanding this risk is critical to developing mitigation strategies, particularly in eligibility and benefits determination contexts. n Traceability: Systems must allow decisions to be tracked, traced, and explained to ensure transparency and trust. n Security and privacy: Protecting sensitive personal information is essential, particularly given evolving standards around AI and data usage. n Regulatory compliance: Requirements related to AI are still developing. Agencies must account for uncertainty and ensure align ment with existing legal frameworks and program requirements. Making risk explicit at this stage allows agencies to make informed deci sions about where and how to proceed, rather than addressing these concerns after implementation.

Moving Forward With Confidence AI has the potential to play a meaningful role in the future of

human services delivery. It can help agencies manage complexity, reduce administrative burden, and improve responsiveness to constituent needs. At the same time, successful adoption requires discipline. Agencies must balance innovation with responsibility, ensuring that new capabilities align with policy requirements, protect the populations they serve, and integrate effectively into existing operations. A structured, risk-aware and human centered approach provides a path forward. By focusing on outcomes, evaluating feasibility, assessing risk, and implementing thoughtfully, agencies can move beyond experimen tation and begin applying AI in ways that are practical, sustainable, and aligned with their mission.

Eamonn Moriarty is the Chief Technology Officer and Vice President of Engineering at Cúram by Merative.

Reference Notes 1. National Association of Medicaid

Directors. Medicaid agency workforce challenges and unwinding. https://medicaiddirectors.org/resource/ medicaid-agency-workforce-challenges and-unwinding 2. National Association of State Chief Information Officers. The 2025 state CIO survey. https://www.nascio.org/ resource-center/resources/ the-2025-state-cio-survey 3. U.S. Department of Health and Human Services. Public benefits and AI. https://www.hhs.gov/sites/default/files/ public-benefits-and-ai.pdf 4. U.S. Government Accountability Office. Fraud and improper payments: Data quality

and a skilled workforce are Essential for unlocking the benefits of artificial intelligence. https://www.gao.gov/ products/gao-25-108412

Spring 2026 Policy & Practice 27