The New Jersey Department of Labor and Workforce Development (NJDOL) has significantly enhanced its unemployment insurance (UI) system, focusing on ease of use and improving access for workers who are at higher risk of falling through the cracks. Working with U.S. Digital Response (USDR), NJDOL aimed to reduce language barriers by simplifying the application process and, for the first time, making their initial UI application form available in Spanish.
Through this effort, USDR and NJDOL developed a human-centered approach to language access that produced dramatic results. Launched in April 2024, the new form takes 80% less time to complete than the previous version, and NJDOL has successfully achieved parity in completion time between English and Spanish applicants. With the success of this project, NJDOL and USDR sought to use generative AI (gen AI) technology to optimize and expand this approach as they prepare to upgrade other parts of the UI system.
This work aimed to help New Jersey's estimated 700,000 Spanish-speaking workers better access UI benefits. NJDOL's efforts represent a significant step towards a more inclusive and effective UI system, potentially setting a new standard for benefits access nationwide.
Pandemic exposes systemic inefficiencies
When COVID-19 struck, NJDOL faced unprecedented unemployment claims, exposing long-standing issues in the state's UI system:
- Complex terminology: many non-English, and even native English speakers, found the UI jargon confusing.
- Burdensome processes: the application process was time-consuming and error-prone.
- Overwhelmed support systems: claimants struggled with long wait times and limited language support.
NJDOL's existing language access approach fell short. State-approved translation vendors lacked UI-specific vocabulary and didn't reflect New Jersey's diverse Spanish dialects. Bilingual call center agents helped with translations, but this was an unsustainable add-on to their regular duties. These stopgap measures proved inadequate for the complex task of making UI truly accessible.
These issues disproportionately affected vulnerable groups, including lower-wage workers, people with limited English proficiency, and those with non-traditional employment situations.
Gillian Gutierrez, senior advisor/director at the New Jersey Office of Unemployment Insurance Modernization, saw this as an opportunity for transformation. “During the pandemic, we recognized a chance to reinvent the UI experience. We decided to use human-centered design, drawing on experiences from healthcare.gov and VA benefits, to improve how people interact with the system," she said.
Developing a human-centered approach to language access
NJDOL partnered with USDR to develop a human-centered approach to language access.
“USDR does everything possible to make the team functional with minimal burden on us. They find ways to present options and help us make decisions efficiently. This is crucial given how understaffed and overworked state employees often are,” Gillian said.
Over eight months, they co-created content to serve New Jersey's diverse Spanish-speaking population. The framework was grounded in three pillars:
- Co-creating content with policy experts and bilingual front-line staff
- Validating translations with 15 Spanish speakers representing 10 different countries of origin
- Staffing the team with Spanish-bilingual UX researchers and content designers
The project resulted in:
- A new Spanish version of the initial UI application - the first time in New Jersey that the form has been made available in a language other than English
- An initial Plain Spanish Glossary of Unemployment Insurance Terms
- A comprehensive Spanish Translation Guide for Unemployment Insurance
These resources are grounded in plain language principles and incorporate cultural nuances and regional variations to ensure clarity and readability for New Jersey's diverse Spanish-speaking population.
Quantifiable improvements in benefit accessibility
New Jersey’s broad modernization efforts have yielded significant improvements:
- Average application time reduced from over 3 hours to 28 minutes for all users
- Call center wait times cut from 40 minutes to 2 minutes
- 14% reduction in claims requiring manual review, decreasing administrative burden
- Parity achieved in completion time between English and Spanish speakers
- 20% increase in follow-through to the next step after submitting the initial application for people who filled out the form in Spanish
These improvements have had a particularly significant impact on marginalized workers, including seasonal workers, individuals with multiple part-time jobs, and Spanish speakers. The application call center wait time reductions mean less lost wages for those who can least afford it.
"It was simple and the verification was very quick. The agent was very friendly and clear with the information," one Spanish-speaking New Jersey claimant shared. At the same time, these enhancements streamlined the process for all users, regardless of language preference, improving efficiency and reducing administrative burden across the board.
The project has created a ripple effect beyond New Jersey. The Illinois Department of Employment Security integrated the Spanish Glossary into their machine translation tool to translate their unemployment insurance website, demonstrating the initial scalability of this approach.
Leveraging generative AI to scale Language Access
Building on this human-led initiative, USDR and NJDOL explored the potential of gen AI to optimize and scale language access efforts. Thanks to support from Google.org, USDR was able to deploy a team of Google.org Fellows and Spanish bilingual content experts to further develop NJ’s unemployment insurance glossary in plain English and Spanish using genAI. Together, the team developed an approach to validate genAI-assisted translation:
- Fine-tuning large language models (LLMs) such as ChatGPT, Claude, Gemini, etc. using the human-created glossary, translation guide, and relevant UI policy guidance
- Developing a prompt library (a list of commands for the LLM) with specific instructions for writing accessible, culturally relevant Spanish
- Creating an evaluation rubric to measure translation quality quantitatively
Preliminary results show promise:
- Translations were produced three times faster than traditional methods
- In readability tests, Spanish-speaking workers could barely tell the difference between AI and human translations
- The genAI-generated text was 90% similar in overall quality when compared to expert human expert translations
While the gen AI-assisted translations show promise, NJDOL and USDR emphasize the continued importance of human oversight, testing, and validation. The goal is to use genAI to augment, not replace, human expertise in the translation process.
A model for inclusive government services nationwide
NJDOL and USDR are setting a new standard for inclusive, effective public service delivery. “It's a continual journey to make the product meaningful. USDR doesn't just deliver and walk away. They're invested in creating real change, helping us navigate roadblocks and think through complex issues,” Kristin Mateo, policy officer at the Office of the Commissioner, said.
By addressing language barriers and modernizing systems, New Jersey has created a model that enhances equity, improves claimant and administrator experience, and potentially reduces long-term costs. This approach offers a roadmap for other states and government agencies looking to update their services and better serve their diverse populations in an increasingly complex employment landscape.
“By making these training materials free and available to all governments, we’re providing public agencies nationwide with the tools they need to improve language access to ensure even more people can receive the support they need and deserve," said New Jersey Department of Labor Commissioner Robert Asaro-Angelo.
As NJDOL continues to refine and expand this innovative approach, they're exploring opportunities to:
- Expand to additional languages, especially those that are lower-resourced
- Apply these lessons to other government services beyond unemployment
- Share this approach with other states
The long-term goal is to train LLMs to speak "public benefits" in plain language in any language spoken in the US, ensuring usability and access for all, across every state, and every benefit program.
Photo by Andrew Neel on Unsplash
