How an AI agent can help caseworkers maintain the ‘human relationship’ with clients
Khanchit Khirisutchalual via Getty Images
In California, Riverside County is experimenting with an AI agent that aims to help caseworkers focus more on supporting public benefit recipients and less on administrative tasks.
Efficiency is top of mind for most government agencies, and advancements in artificial intelligence technology could help them meet that goal.
Riverside County, California, for example, is developing a new agentic artificial intelligence tool designed to assist caseworkers’ efforts to enroll residents in public benefit programs.
The Riverside County Children and Families Commission is partnering with the public benefit corporation Nava and the tech nonprofit Amplifi to build a large language model agent that can review databases and benefit portals autonomously to help fill in application forms.
The project launched in June with $1.5 million in funds from Google’s Generative AI Accelerator program, and the partners debuted a prototype of the tool during a webinar Thursday.
“During our project, we found that frontline staff spend about half their time navigating the benefit applications,” she said. “We also found that in Riverside, up to a third of [staff] aren't able to get through the process due to administrative or procedural challenges, like not having the right paperwork or having some data entry mistakes.”
The root of the tool’s mission is to ease the intake process for clients and caseworkers, said Jillian Hammer, senior designer and researcher at Nava.
“Riverside’s frontline staff and leadership are really aligned in this shared belief that clients should only have to tell their story once … because right now, typically, if a client wants to apply for a program, they complete applications one by one over multiple visits,” she said.
Every time a client returns to continue the process, they may have to repeat the same details, risking the miscommunication of data or forcing clients to go over personal information that may be difficult for them to revisit, Hammer explained.
The county and research partners hope the “agent will make it easier for caseworkers to support families submitting benefit applications by automating some of the repetitive application tasks, but still [maintaining] caseworker oversight,” Wilkes said.
To leverage the AI tool, a caseworker provides their client’s name and the application URL to the model. The agent can then screen the county’s case management system and other databases to retrieve data that matches what the application form is requesting, Wilkes explained.
The agent can, for example, interpret if a form field calls for a phone number and enter that information for the caseworker to review before submitting the application, she said. The AI tool also provides a summary to flag any missing or incomplete form fields that it was not able to address.
“The next generation of innovation is AI agents, and basically we're harnessing the power of these large language models to not just respond to questions, but to actually take actions on behalf of people,” Wilkes said.
Caseworkers are still able to manually edit benefit application forms to ensure the AI model includes human oversight and decision-making throughout the process. The human oversight aspect is also important because many application forms include CAPTCHA verification measures, she explained.
During usability testing, county caseworkers reported that the tool brought significant time savings and helped them assist clients more effectively, Hammer said. While AI agents can help streamline the administrative work of navigating data and portal systems, “the human work of discerning complex family scenarios and making sure a client feels heard definitely still belongs to people,” she said.
Indeed, “these productivity gains are really only meaningful when they’re built upon the human expertise and the human relationships that are at the core of this work,” Hammer said.