Avoid ‘analysis paralysis’: Utah AI director says time-saved is best measure of AI’s worth

Utah’s AI director on this week warned that officials examining a technology as complex as artificial intelligence can fall victim to “analysis paralysis,” but that he has a simple solution. Christian Napier, director of AI at the Utah Division of Technology Services, said officials can most clearly demonstrate AI’s value to lawmakers by measuring how much time it saves state employees.

Since Utah rolled out Google Gemini for Workspace in February 2025, state employees have collectively saved more than 60,000 work hours per month, Napier shared Wednesday during a webinar hosted by the Government Risk and Authorization Management Program. (GovRAMP is a nonprofit, previously known as StateRAMP, that develops security standards for cloud computing providers).

Napier said that because government is not driven by profit, defining AI’s return on investment can be challenging in the public sector. Analysis paralysis sometimes arrives because while AI makes it possible to measure nearly everything, trying to measure too much can slow decision-making and stall action.

“So we have to look at returns in different ways, and those returns can be different for different agencies and different projects, and that can be confusing for people, especially when trying to approach the legislature for funding,” Napier said during the webinar. “What we tried to do is really distill it down. Is there one metric that we can focus on and not get lost in the weeds, because you can measure so many things.”

Napier described Utah’s process of rolling out AI as including “a lot of bumps in the road,” and admitted that the state has learned some “hard lessons.” To avoid getting lost in the weeds, he said, his office has focused on showcasing the technology’s value through surveys. He said roughly 5,500 of the state’s 22,000 employees are at least weekly users of Google Gemini, and a first survey of those active users conducted last July showed that the technology was saving employees about 11,000 hours of work per week, or 40,000 hours per month — which comes out to about 500,000 hours per year.

Napier said the state is working on another survey, which 40% of the state’s active users have responded to, showing employees are now saving around 60,000 hours of work per month. He said the state is using this data to plot a route to its next goal, which is that by end of the year, employees will be self-reporting at least one million hours per year saved through AI tools.

“Now, that’s not a perfect metric, and I don’t know if there is a perfect metric, but what’s important for us to settle on one, and kind of shut out all of the others,” he said. “I think these surveys have been very important for us to help understand the mindset of employees, both those who are adopting these tools and those who have yet to adopt these tools, so that we can engage more people and meet them where they’re at.”

Napier did mention a few pitfalls regarding a focus on the time-saved metric, such as that it can be misread by the legislature as a potential opening to cut staff. He said his office has been “very, very careful” to not tie time savings to an argument for a reduction in headcount, and that this shift in narrative has led employees away from worrying about AI taking their jobs to learning how they can use AI to perform tasks more efficiently.

The state is also piloting Claude Code, Anthropic’s AI-powered coding tool, and as part of the pilot, the state conducted another survey. It asked participants to estimate the percentage change in their efficiency or task completion speed using the tool, and 55% reported being at least 20% more efficient after just one week of using it, Napier claimed.

“These technologies are significant. They’re not going away,” he said. “I do believe they are once-in-a-generation tools. And the question then for us is, OK, how can we take the lessons from this group of people who are seeing these kinds of results in the work that they’re doing, and apply them to other groups of people who are doing similar things in the government?“

Napier said Utah is in the “advanced stages” of citizen-facing AI prototypes, but they are not yet fully rolled out to all residents. These efforts follow the “pro-human” AI initiative that Gov. Spencer Cox announced in December that will see $10 million spent on ensuring Utah’s workforce is “AI ready.”

Under Cox, measuring time-saved though AI use has been a recurring method of tracking success for the last several years. Alan Fuller, Utah’s former chief information officer, said in 2023 that modernizing applications and services to advance the governor’s goals was a top priority for the state’s Division of Technology Services. Fuller also used time-saved as a core metric, with the aim of saving residents one million hours per year.  

A number of other states have also shared ways AI is saving their employees time. Pennsylvania’s officials have said the state’s pilot with ChatGPT showed employees individually saved about eight hours per week, freeing them up to do more complex work. They, too, highlighted that generative AI acted as a “job enhancer,” rather than a replacement. New Jersey’s innovation report last November highlighted how scaling up an in-house AI assistant tool across its agencies helped cut staff time on manual processes.