Westat’s AI Governance Framework Balances Innovation, Quality, and Trust

As the use of artificial intelligence (AI)-driven methods in research continues to expand, organizations face increasing pressure to maintain high standards for quality while safeguarding data privacy and security. To meet this challenge, Westat developed a comprehensive AI governance framework designed to support the responsible, ethical, and secure use of AI technologies across the organization.

Westat’s AI Hub serves as the central body for AI governance, bringing together experts from across the organization to oversee implementation, establish policies, and provide staff education and resources. Through the AI Hub, Westat developed both its AI governance framework and a generative AI (gen AI) policy, which guide how AI tools are evaluated, deployed, and monitored throughout their use. Together, these structures create a system of technological and human oversight that supports innovation while helping ensure clients receive secure, high-quality, ethically produced, and trustworthy data and insights.

Moving Beyond High-Level AI Principles

Many organizations have adopted broad AI governance principles, but translating those principles into practical implementation remains a significant challenge. Recognizing this gap, Westat is developing structured AI review protocols modeled after review processes for human subjects research. These protocols are intended to operationalize governance principles in day-to-day project work and embed oversight throughout the full project life cycle.

“At many organizations, principles like ‘human in the loop’ sound good in theory, but implementing them consistently is much harder,” says Gizem Korkmaz, PhD, Westat Vice President, Data Science and AI, Statistics and Data Science. “Westat is focused on building governance approaches that are actionable, reviewable, and integrated into how projects are actually conducted.”

Providing a Cross-Functional Approach to AI Oversight

At the center of Westat’s AI governance effort is a cross-functional team that includes representatives from legal, IT infrastructure, cybersecurity, and the Institutional Review Board (IRB). The group evaluates AI tools to identify solutions that best meet organizational standards for efficiency, flexibility, security, and compliance.

The team continually assesses each tool’s strengths and limitations while ensuring alignment with federal guidance and responsible AI frameworks, including those developed by the National Institute of Standards and Technology. Team reviews examine a range of potential AI risks, including privacy and security concerns, hallucinations, bias, plagiarism, research misconduct, harmful outputs, and lack of transparency or replicability.

“AI is evolving rapidly, and there is tremendous potential for AI-driven technologies to improve efficiency and accuracy in federal work,” says Korkmaz. “Our goal has been to create a framework that allows us to leverage advances in AI while maintaining rigorous standards around privacy, transparency, reproducibility, and research quality.”

Embedding Oversight Throughout the AI Life Cycle

Data protection remains a central priority within Westat’s governance framework. To support this goal, the organization established a continuous oversight system that combines technological monitoring with human review. Policies guide every stage of the AI life cycle, from data collection and model development to implementation and final delivery.

Westat’s governance principles also place humans firmly in control of AI-supported work, treating AI as a tool that enhances human expertise rather than replacing it. “That’s why every AI deliverable goes through human verification,” says Korkmaz. “Subject matter experts review key outputs to ensure quality, accuracy, and responsible practices.”

At the same time, Westat recognizes that reviewing every AI-generated output manually can reduce efficiency. To address this challenge, the organization applies its longstanding expertise in sampling and statistical methodology to quality assurance processes.

According to Brian Sokol, MS, Vice President, Data Architecture and Data Engineering, Technology and Digital Solutions, Westat uses techniques such as random and stratified sampling to validate AI-classified records, identify potential bias, and strengthen quality control. Oversampling certain groups helps ensure that AI models perform accurately across different groups.

Supporting Clients With Practical AI Solutions

Westat supports clients with a range of AI-enabled activities, including proposal review, knowledge management, document summarization and classification, and the conversion of unstructured information into structured databases. “Clients also request our AI support to code medical conditions, synthesize qualitative documents, and categorize reports,” says Korkmaz.

In addition, Westat develops AI-powered chatbots for clients while implementing safeguards designed to prevent inappropriate or offensive outputs and validate the accuracy of AI-generated responses. “We develop chatbots with strong guardrails and verification processes built in,” says Sokol. “Ensuring the quality and appropriateness of AI-generated responses is essential.”

WesLytics™ Integrates AI With Subject Matter Expertise

Much of Westat’s client-facing AI work is conducted through WesLytics, the company’s enterprise-wide data intelligence platform. “We launched this integrated, AI-driven platform last year to manage, engineer, and analyze complex data of any type, volume, or origin,” says Sokol. “What distinguishes WesLytics is that it combines industry-leading Databricks AI and cloud management capabilities with Westat’s subject matter, technical, and methodological expertise to deliver efficient and adaptable solutions for clients.”

By pairing advanced AI technologies with deep research and statistical expertise, Westat aims to provide solutions that are both innovative and rigorously validated.

Addressing Emerging AI Risks

As AI technologies continue to evolve, Westat is expanding its governance processes to address emerging risks related to privacy, data protection, and sustainability. The organization continues to strengthen its software reviews, architectural assessments, and IRB frameworks to keep pace with new AI capabilities and challenges.

Sokol notes that environmental sustainability is becoming an increasingly important consideration, particularly given the substantial water and energy demands associated with cooling large-scale data centers that support AI systems. Westat experts are engaged with the research community on considerations for optimization techniques, domain-specific model development (e.g., lightweight or fine-tuned models for subject areas within health care), and human-interpretable machine learning (ML) to develop simpler models and improve processing efficiency.

Sharing Expertise Across the Research Community

Westat’s data science experts are frequently invited to share their expertise on AI governance and responsible AI implementation with external audiences.

In April, Korkmaz delivered an invited presentation during the session “Working Through AI Governance and Implementation for Statistical and Mission-Enabling Uses,” jointly sponsored by the Committee on National Statistics, the Federal Committee on Statistical Methodology, the National Institute of Statistical Sciences, and the National Academies of Sciences, Engineering, and Medicine.

For the Joint Statistical Meetings this August, Westat has organized the session “Sustainable AI Deployment and Opportunities for Statistical Practice.” The panel will discuss strategies for making AI systems environmentally and cost sustainable while maintaining performance. Experts will cover model selection, evaluation, deployment, and governance; resource optimization; and sustainable AI practices.

Advancing Responsible AI for Research and Public Service

As AI continues to reshape how data are collected, analyzed, and applied, Westat’s governance framework demonstrates that innovation and accountability can advance in tandem. By combining advanced AI capabilities with rigorous oversight, human expertise, and a strong commitment to privacy, transparency, and quality, Westat is helping establish a practical model for responsible AI use in research and public service.

This work not only strengthens client trust but also helps ensure that the insights derived from the data help guide decisions in healthcare, education, and government to serve the public good.

Multipart Series

This article is the third in a multipart series on governance at Westat, featuring insights from members of Westat’s Data Governance Committee. The series explores Westat’s framework, how data governance works in practice, and this latest article on how Westat approaches AI governance in a continually evolving technology landscape.

• Dynamic Guardrails for a Data-Driven World: Inside Westat’s Governance Frameworks – Westat
• Westat’s Data Governance in Practice – Westat