Nevada turns to AI to speed up unemployment appeals
Leinani Shak Photography via Getty Images
By
Chris Teale
The state partnered with Google Public Sector on a system that analyzes appeals hearings against state and federal law to help adjudicators reach a decision. They say it’s made the process four times quicker.
When the COVID-19 pandemic hit in 2020, every state struggled under a massive influx of unemployment insurance claims, but few were hit harder in that area than Nevada.
In a state reliant on hospitality and entertainment, the shutdown of public spaces and events led to the state being faced with 200,000 unemployment insurance claims a week, up from a normal weekly average of 20,000. It inundated the state’s legacy system, caused an enormous backlog, delayed payouts, and impacted the appeals process where claimants are interviewed to determine if they are eligible.
The appeals process alone had a backlog of more than 40,000 cases and was conducted through an outdated system that relied on an adjudicator interviewing an applicant, analyzing artifacts and evidence supporting their claim and checking the relevant sections of Nevada state law. It could take months to serve residents in a timely manner, said Carl Stanfield, chief information officer at Nevada’s Department of Employment, Training and Rehabilitation.
“It's not necessarily manpower or scaling up, because in this area with the appeals, you're talking about very specialized individuals who have knowledge of the law,” Stanfield said in a recent interview. “There's a lot of laws and obligations in there that also have to match up with federal law and state law in order to provide that service. When you have that many backlogged cases and you've only got so many appeals referees that can hear them, our search was to try to find a way to find technology to help with this.”
That search led the state to partner with Google Public Sector to find a solution that uses artificial intelligence to adjudicate those appeals more quickly. Nevada started working with the company in September 2023.
Now, appeals hearings are held virtually via Zoom, and those interviews are transcribed then ingested by the AI, which compares it to relevant cases and state and federal law, then provides a recommended decision to an appeals referee. That referee then verifies the software’s findings and the relevant sections of law it has cited before issuing a final decision. The state estimates that this small change to how it processes appeals has made approvals four times faster.
“What would take an appeals referee hours, now takes five minutes,” Stanfield said. “It literally takes two minutes for the decision to be put forth by AI, and the referee can spend another three to five [minutes] going back through it, checking and validating for errors. They were there. They participated in that hearing, so they know what they're looking for.”
The initiative means the state’s appeals backlog is “pretty much caught up now,” Stanfield said Adding that the time savings are crucial, especially given how long an appeal can take for complex cases.
“An actual hearing itself can take up to five hours, and that's a lot of transcripts that the referee will have to go back through, and then they have to write the decision letter from scratch, find the right law, apply the right law to the right case that they're on,” Stanfield said. “And it's rather lengthy. It could take 15 minutes to write a very simple determination, but it could take up to a day if they’ve got one that has multiple witnesses, multiple artifacts that need to be included.”
And it means that referees’ jobs have become easier, as the AI can help them find relevant information more efficiently.
“The referee won't have to research law as intensively,” Stanfield said. “It's somewhat of a training tool as well. The AI can find the right applicable law and bring it back much sooner than the referee can go out and find something that's brand new and something they may never have seen before. Then they simply verify that that's the right law to use for this case.”
A number of states and localities are turning to AI to help them better navigate what can be a lengthy and challenging process for approving public benefits. That includes jurisdictions like New Hampshire and Los Angeles, and more could follow given the public’s interest in it.
A Salesforce survey last month found that 87% of respondents said they would use an AI agent, especially as many of the public feel that interactions with government agencies and the paperwork they must complete can be too challenging.
While the COVID-19 pandemic was a wake-up call for Nevada, especially for its impact on the entertainment and hospitality sectors, Stanfield said this system can help the state be prepared for the next catastrophic event, whenever it comes.
“We don't want to be caught flat footed again,” he said.