Arizona's Department of Economic Security (DES) serves one of the nation's most linguistically diverse populations, with 8.1% of the State’s population reporting that they speak English less than “very well." DES has 8,000+ employees across multiple divisions, and faced the challenge to provide consistent, high-quality translations across 60+ programs in 15 threshold languages.
Even with skilled in-house translators, the sheer volume of requests outpaced available capacity. Meanwhile, separate vendor relationships across divisions made it challenging to maintain consistent terminology and manage costs across 60+ programs. When policies or programs changed, updating translated materials became a resource-intensive challenge.
"It just wasn't working. We needed to look at a more scalable approach to this, not only for capacity building, but also because all of the divisions within our large agency were doing things differently," Crystal Newman, Creative Bureau Chief in DES's Office of Strategic Communications, said.
DES was launching several technology modernization initiatives, and needed a translation solution that could scale across the agency while maintaining quality and reducing vendor dependency.
Why USDR?
DES initially considered acquiring an online translation platform; however, their governance and innovation team introduced them to USDR after identifying the need for a more holistic strategy. USDR's approach was particularly impressive: not only would they prototype a solution, but the team would also engage in user research, assess current workflows, and offer recommendations based on data insights.
“The prospect of conducting research and obtaining actual data to inform our decisions was particularly significant. Bringing in USDR to analyze all that information and offer us guidance on the optimal course of action seemed like the most logical step for us," - Crystal Newman, Creative Bureau Chief in DES's Office of Strategic Communications
Our approach
USDR deployed a cross-functional volunteer team over a 14 week period, combining product management, engineering, content strategy, user research, and design expertise. The engagement focused on the Supplemental Nutrition Assistance Program (SNAP) as a pilot, working closely with DES staff across communications, policy, governance, IT, and frontline services.
The team started by developing core components for effective AI-assisted translation:
A "Golden" translation dataset: Working with DES subject matter experts, USDR co-created a SNAP-specific lexicon of 100+ terms and content examples, user-tested with both English and Spanish-speaking residents. This ensured AI-generated translations would maintain contextual accuracy, consistency, and cultural relevance.
Configurable prompt library: The team developed prompts for use in any Large Language Model (LLM) tool that could be easily adapted across multiple languages, DES programs, and use cases, empowering staff to generate contextually appropriate content for specific scenarios.
Quality evaluation framework: USDR established assessment tools and explored automated evaluation methods to measure and maintain translation quality over time, including rubrics for accuracy, cultural relevance, and readability.
Empowering staff
Rather than creating a "black box" solution, which is what many vendors in the AI space are offering, USDR prioritized building internal capacity:
AI literacy workshops: The team facilitated prompt engineering training for DES communications staff, teaching them to create and refine translation prompts independently. "The assistance of a project manager and someone willing to help, along with the coaching and training they provided, was essential,” Crystal said.
Human-led, AI-assisted workflow: Using real SNAP content, the team co-designed a translation workflow that demonstrated how AI could increase staff capacity and reduce translation times by over 50%, all while maintaining human oversight and quality control.
Creating flexible implementation pathways
Understanding DES's technical landscape, USDR developed a dual-track approach:
Immediate deployment: A low-code translation workflow using Google Workspace and Apps Script that leverages the glossary, prompts, and evaluation tools with minimal IT support. This gave non-technical staff immediate plain language and translation capabilities while the agency built its longer-term solution.
Long-term scalability: Technical documentation and implementation-ready components that DES's IT team could integrate into their Azure AI infrastructure, ensuring the solution could expand across programs and languages with minimal overhead.
"USDR laid the groundwork, and we’re excited to implement it.," Crystal said. “The process has been incredibly smooth."
Technologies & practices used
Technologies:
- Google Workspace & Apps Script
- Azure AI Services (OpenAI API)
- Automated LLM evaluation tools
Practices:
- User research with residents and staff
- Co-design with subject matter experts
- Prompt engineering and LLM evaluation
- Content strategy and plain language development
- Service design
- Human-centered AI implementation
The Impact
The prototype is currently in pilot phase, with DES gathering performance data and preparing for agency-wide deployment in 2026. Early results demonstrate the solution's potential:
For staff:
- Pilot testing indicates reductions in translation time by over 50% compared to manual processes
- Maintained plain language translation quality above 90% accuracy
- Enabled rapid response to urgent requests with minimal revisions
For the agency:
- Scalable infrastructure designed to support 15 languages
- Projected 80%+ reduction in translation costs
- Consistent terminology standards across 60+ programs
- Reduced dependency on outside vendors through built internal capacity
For SNAP recipients: The solution aims to maintain or improve outcomes for the 923,400 individuals receiving SNAP benefits in Arizona by:
- Reducing language barriers to critical benefits information
- Improving clarity and accessibility of program communications
- Supporting parity in language-based outcomes across English and Spanish speakers
Word of the pilot's success has spread rapidly across DES. "We continually get hit up by divisions. I heard you have this, and I gotta say, yes, just wait. We're almost there," Crystal said. "It has spread rapidly."
Perhaps most significantly, the project demonstrated how communications teams can effectively partner with IT on AI tool development. "The beauty of it," said Marcie Chin, USDR Engagement Advisor, "is that you don't have to be a computer scientist or a software engineer or any kind of AI expert to really use these tools effectively. A lot of it is really around content design, content strategy, user research... empowering Crystal's team with the skills they need to maintain a tool like this over time."
The Team
USDR Volunteers:
- Christopher Fan – Product Manager
- Kim Springston – Project Manager
- Karla Fernández – Lead Content Strategist
- Felipe Gacharná – Content Strategist
- Sofia Sandoval – UX Researcher
- Gabby Pascual – UX Researcher
- Laurel Korwin – Front End Developer
- Donna Miller Watts – UX Designer
Looking Ahead
DES is on track to deploy the translation tool agency-wide in March 2026. The comprehensive glossary will also support the agency's broader plain language initiatives, helping meet new ADA Section 504 accessibility requirements across web, graphic and design, and video content.
"I see the solution being agency-wide, something that all divisions can utilize," Crystal said. "Really expanding that glossary is going to be important... Anything that we can do to be efficient is going to not only help us, but help our clients. And that's ultimately who we're doing all this hard work for."
The project demonstrates that with the right approach, combining human expertise, user research, and practical AI implementation, government agencies can dramatically expand language access while building sustainable internal capacity. As DES continues to refine and scale the solution, they're creating a model for other agencies seeking to serve linguistically diverse communities more effectively. This can also serve as a model for states across the country seeking to improve access to benefits and services to their communities.
"I don't see any reason you would not want to work with USDR... There's something about volunteers too, they were as dedicated as we were. Along for the ride. Ultimately, no matter where the solution leads, you will gain valuable information that can inform your next steps and decisions. This knowledge will help ensure you are prepared and ready to move forward." – Crystal Newman, Creative Bureau Chief, Arizona Department of Economic Security
Thumbnail photo by: Photo by Centre for Ageing Better on Unsplash
