With a new virtual assistant that uses artificial intelligence to answer Californian’s tax questions, the California Department of Tax and Fee Administration hopes to eliminate the need to shift workers and speed up the call center’s work.
Brandon Taylor, a call center employee who began using the tool two weeks ago, had a positive review. The program made it easier to narrow the caller’s question, he said, and decreased the amount of time he needed to spend on the phone with customers.
“It’s great to have more backup,” Taylor added. “Anything we can do to get (callers) off the phone quicker is a good thing.”
The call center assistant is one of three AI-related initiatives receiving state funding as a result of Gov. Gavin Newsom’s broader effort to use the now-widespread technology in state government. Evidently, the Newsom administration sees benefit in generative AI — a technology popularized by OpenAI’s ChatGPT, a model that uses training from massive data sets to produce text and other content.
“GenAI is here to stay, and the state is excited to learn more about GenAI’s capabilities through these projects,” said Roy Kennedy, the deputy secretary of communications for the state’s Government Operations Agency, where the Newsom administration referred the Bee’s questions about these AI projects.
As California pushes into an AI-infused future, departments that are incorporating this new-to-government technology, like CDTFA, have to juggle maintaining the public’s trust and addressing labor concerns — without disrupting the critical services, like collecting taxes, needed to keep the state government functioning.
AI'S ARRIVAL BRINGS LABOR CONCERNS
In early public announcements about the generative AI push, the Governor’s Office made it clear that the state would not be investing a significant amount of taxpayer dollars into the projects in their nascent stage.
Now, three of those projects have moved into the next phase, which involved departments signing contracts with vendors to provide these technologies to the state. In total, two departments have signed deals that will cost the state $3.9 million.
CDFTA is paying $445,000 for a year-long deal with SymSoft Solutions to use the AI assistant. The other two initiatives are run by the California Department of Transportation: One involves analyzing frequent crash locations with the goal of creating safer roads for bikers and pedestrians; the other will attempt to reduce highway congestion by studying traffic patterns and weather.
Departments using AI models to support services must also consider their impact on the workforce. Labor leaders representing public employees want these new technologies to support workers, not push them out.
Anica Walls , the president of SEIU Local 1000, which represents CDFTA employees, expressed broad concerns about how AI technology could impact the workload of employees and the quality of public services if there is an expectation that workers’ output increases.
In regards to the CDTFA project, Walls said that the union had meetings with the department about the generative AI technology. But to the extent SEIU Local 1000 has been involved in ensuring the technology helps employees do their jobs, “I’m not sure we have been engaged to the level that I would like to see us engaged in,” Walls said.
Those comments come five months after the union signed on to a statement from the governor’s office celebrating the AI initiatives. In that statement, the union advocated for “strong safeguards that promote equity, transparency, and ensure that workers — especially those most impacted — have a real voice in how these tools are developed and used.”
But, Walls told the Bee, union involvement has fallen short of her expectations. For projects like these, she said, it’s crucial for employees’ perspectives to be included from the beginning because they are most familiar with the work that generative AI is designed to assist with.
“Noticing us is great and we’ll tell you how it’s going to affect the working conditions,” Walls said. “But sometimes at that point, it’s a little too late.”
Kennedy with GovOps said the agency was surprised to hear this assertion.
“Our labor partners were engaged from the earliest point and met with our teams more than a dozen times to work with us on the GenAI executive order,” Kennedy said.
Asked about this claim from Kennedy, Walls reaffirmed her stance that invitations from the Newsom administration to be part of the conversation have come too late and have not sufficiently included the voices of workers.
“Oftentimes, management is out of touch with how or what they consider involvement,” Walls said. “They probably do think an invitation is enough.”
Though Walls did not provide specific examples of ways in which SEIU Local 1000 has been excluded in the development of AI models, she maintained that the Newsom administration has not included the union in conversations to ensure departments have guardrails to protect workers as they incorporate AI programs.
Artificial intelligence is going to affect the workforce, said Carlos Ramos, a former chief information officer with the Brown administration who cofounded a company in recent years to help local governments to adopt AI tools.
The government needs to be honest about that, he added. Though the net loss or gain in jobs and how it will impact public employees’ work isn’t clear yet.
Often in government, the first instinct is to avoid disrupting systems even if something is cumbersome or inefficient, Ramos said, but the technology can make government information more accessible and easier to understand.
“As with anything new, you have to start incrementally, right?” he added. “A lot of people don’t trust the government to begin with. You don’t come out the gate.”
INSIDE CDTFA'S AI ASSISTANT
Each year, CDTFA generates over $90 billion in revenue for California by collecting sales and use, alcohol, fuel, tobacco and cannabis taxes.
To help collect that revenue, 120 state workers answer roughly 800,000 taxpayers’ questions each year, said Dunn, who oversees the center.
Call center employees have to sort through thousands of web pages and documents to provide guidance on how to correctly file taxes, Dunn said. For complex questions, it takes time to provide accurate answers. Extended calls can create longer wait times or result in abandoned calls, which can have an impact on revenue generation for the state.
Dunn said the AI assistant has led to a more efficient call center: “Anything we can do to create more efficiencies is a good thing and that’s what we’re looking to do.”
The generative AI tool used by CDTFA — built by SymSoft Solutions, a Sacramento-based technology company — works as a sort of assistant to agents fielding taxpayers’ questions. When a customer calls, the system transcribes the conversation and redacts personal information before the information is shared with the AI assistant, Dunn said.
Three steps follow: The AI assistant determines the customer’s question based on the conversation; then the technology sorts through the relevant information to find the answer; and finally, generative AI is used to produce a response to the original query.
The responding agent then reviews the AI-supplied answer, decides if it’s correct and then can provide that information to the customer, Dunn said. He said roughly a quarter of the center’s agents started using the tool late last month.
Even saving 1.5% of the agents’ time on calls — which is what initial analysis found that SymSoft’s program does — represents a major improvement, Dunn said. Given the number of queries CDTFA receives annually, the department estimated it could result in being able to answer over 10,000 more calls each year.
The state can learn from concrete, thoughtful pilot projects to determine these technologies’ potential and learn from their mistakes, said Meredith Lee, the head of strategic partnerships at the UC Berkeley College of Computing, Data Science, and Society.
Recruiting talented technology experts has always been a challenge for the government because it’s difficult to compete with the private sector, Lee said. That’s not likely to change, and these public-private collaborations are crucial to incorporate new technologies into government services.
For California’s nearly 40 million residents, there’s a range of familiarity with AI. To manage those varying degrees of trust, Lee said it’s critical for these projects to have a human-centered design.
“If you don’t think about human beings in the loop in the beginning, it’s going to be quite challenging to build and sustain trust,” Lee said. “People don’t use technologies they don’t trust.”
©2025 The Sacramento Bee, Distributed by Tribune Content Agency, LLC.
