RALEIGH — North Carolina State Treasurer Brad Briner issued an independent study report on his agency’s OpenAI pilot program.
The report was rolled out at a joint press conference Aug. 1 with NC Central’s Siobahn Day Grady, founding director of the university’s Institute for Artificial Intelligence and Emerging Research (IAIER), which conducted the study.
“Adopting this innovative technology has helped us deliver improved results to our constituents and to taxpayers,” Briner said in a press release. “This important analysis clearly shows that adding the power and speed of artificial intelligence to the talent, experience and judgment of our state employees is the key to unlocking greater workplace achievements.”
The pilot program began in March, and the 48-page report covered various user experiences with ChatGPT. It included how AI saved workers time but also illustrated some of AI’s limitations and how it did not replace human judgment but was used to refine the work being done.
ChatGPT was used in the areas of unclaimed property and state and local government finance.
“The 2025 ChatGPT Pilot engaged a diverse group of public sector employees to explore how generative AI could support day-to-day operations,” the report states. “Feedback from pre- and post-pilot surveys showed broad enthusiasm, with 85% of participants reporting a positive experience.”
While 85% of participants reported a positive experience using ChatGPT, some users found it sometimes gave inaccurate information. In particular, the program didn’t handle phrasing well, there were issues with data comparisons, or ChatGPT had “hallucinations,” a situation where the AI produced nonsensical or misleading information as fact.
“The technology is not perfect,” Grady noted. “We always need to do our due diligence, and when using the technology, making sure that things are factual and not rely 100 percent on it to do our jobs, because we are the subject matter experts.”
According to the findings, time savings averaged 30 to 60 minutes per day, and “real-world usage evolved significantly beyond initial expectations.”
“We estimate that it improved the productivity of our employees by about 10 percent in the early phases of the trial, and it kept improving,” Briner said during the press conference.
The treasurer also said the partnership on the program did not cross the “bright red-line” and “we did not touch personal and private data.”
There were 26 participants divided into two cohorts and four user types: sporadic users who made up 48% of participants, super users (37%), specific users (12%), and nonusers (4%) who had limited use due to constraints or discomfort.
Some users found the training adequate, but some reported a steep learning curve that will need to be addressed.
“Some pilot participants found it difficult to get started with ChatGPT because they were overwhelmed by the new technology, hesitant to change established work habits, or simply too busy to experiment,” the report says.
