Los Angeles turns to AI to give public benefits enrollment a boost
Teera Konakan via Getty Images
A chatbot tool launched in Los Angeles County looks to help case workers more quickly and efficiently get vulnerable residents linked to social services.
A pilot program in Los Angeles County looks to streamline workflows for case workers tasked with linking people to public benefits programs by leveraging an artificial intelligence-enabled chatbot.
Imagine LA, a nonprofit dedicated to reducing homelessness and poverty in Los Angeles, is launching the chatbot for benefit navigators with the help of Nava, a public benefits corporation that works with federal, state and local government agencies to address technology challenges.
The tool is designed to offer users quick information on public benefits programs and tax credits to help caseworkers streamline eligibility determinations for their clients, said Diana Griffin, senior product manager at Nava, speaking during a webinar on the pilot program last week.
“The goal is that caseworker experience and expertise, combined with the AI solutions that support them, will ultimately result in better enrollment and referral outcomes,” she said.
The assistive chatbot builds upon Imagine LA’s previous work developing its Benefits Navigator software-as-a-service solution, which includes an information hub where case navigators can review the scope and requirements of federal, state and local public benefits and tax credits and a tool to project how people’s benefits will change if their income is adjusted.
A pilot program from September 2023 to November 2024 showed that the Benefits Navigator solution helped more than 500 case managers across government agencies, community-based organizations and other social services groups assist more than 10,000 beneficiaries, securing, on average, an additional $10,869 in benefits per household.
The pilot is slated to operate until June, but user feedback throughout its duration has already suggested a need for a chatbot, said Brit Moore Gilmore, chief product officer at Imagine LA and co-founder of Benefits Navigator.
Case navigators have reported, for instance, that sometimes a client has a question that the navigator would like to answer quickly without having to dig through the appropriate documents first, she explained. Another obstacle includes case managers knowing they could proceed with eligibility determination more efficiently, if they had a baseline understanding of program requirements that a chatbot could offer in real time.
The chatbot is not an “ask-me-anything tool,” Griffin said. It draws responses from a predefined set of documents, such as information from government websites and other resources approved by Imagine LA’s expertise and research Nava conducted with caseworkers.
Chatbot responses include two parts, including responses to users’ inquiries that are summarized from the large language model and citations that identify the source content, she explained.
The citations “are not touched by the LLM, which is really important for allowing caseworkers to review where the chatbot got its information and verify its summarized response,” Griffin said. “If the caseworker wants to dig deeper, they can follow the link to go directly to the webpage that the citation is quoting.”
Users can also direct the chatbot to use simpler language or translate content to other languages like Spanish, and the chatbot can flag when outputs contain potentially out-of-date content, such as recent policy changes.
The organizations are also partnering with researchers at Cornell University and Georgetown University’s Better Government Lab to study how helpful the chatbot is with answering users’ queries accurately and impacting benefit navigators’ workflow and administrative burden, among other metrics.
“The big picture idea here is that there are a lot of applications of LLM-based tools to public services right now, and there aren't a lot of general ways to evaluate these tools,” said Eric Giannella, associate research professor at the Better Government Lab.
By collecting and evaluating data to determine the accuracy of human versus human-and-chatbot responses, he said, program leaders will be able to better refine the chatbot’s outputs before scaling and launching the tool to be available to more users.
The Benefits Navigator tool and the chatbot pilot program, Gilmore said, is in the works to expand to other areas in California and other cities like Chicago, D.C. and New York City.