How do you train an AI service agent for the public sector? Raleigh, North Carolina, CIO Mark Wittenburg likens it to onboarding a new employee.
“You don't bring somebody on to the help desk and then immediately give them admin [permissions],” Wittenburg told Smart Cities Dive. “You slowly build trust over time.”
Since Raleigh became the first municipality to deploy ServiceNow’s L1 AI Specialist product this past spring, its service desk costs and staff hours have declined, Wittenburg said.
More public sector CIOs are exploring AI, surveys show, and providers promise big benefits for local governments. One of the tangible benefits of AI — and one government leaders seek the most — is improved efficiency. For Raleigh, the biggest time-saver has been in the service desk ticketing process, Wittenburg said.
The city developed what Wittenburg describes as a “triage agent” named Ral-E. But instead of sorting trash and having space adventures like the animated robot that inspired its name, Ral-E makes sure city help desk tickets go where they’re supposed to — a process that used to be done manually.
“For our help desk, which is very small, having those tickets go to the right place, to the right workflows, with the right templates, really is huge for us,” Wittenburg said.
Wittenburg, who is on the ServiceNow product advisory committee, said he measures the success of new AI tools based on various factors, including time saved, work diverted and accuracy. So far, Ral-E is routing tickets correctly 98% to 99% of the time, he said.
Ral-E was originally built as a chatbot for the city using generative AI with more advanced agentic AI elements incorporated later, according to Wittenburg. While generative AI can create content based on a user’s prompt, agentic AI is designed to make decisions with limited human supervision needed, according to IBM.
Alli, a separate AI agent, is Raleigh’s new help desk agent, Wittenburg said, and is capable of pulling from a knowledge base and then doing work like adding someone to a distribution list or walking someone through the process of resetting their password. The AI agent is also capable of working around the clock.
“It’s the next evolution, or the next step,” Wittenburg said.
After a testing phase that began in February, Alli is now available department-wide for city employees.
“Training Alli was truly like training a new help desk agent coming on board,” Wittenburg said.
There was a learning curve — on both sides of the screen. After Alli gained access to the city’s knowledge base and ticket knowledge base, it “started telling people to change registry settings and do things that you never do,” Wittenburg said.
“And so we realized we’d given her too much autonomy and too much information,” he said. “She wasn’t wrong; it’s just not what we typically have the users do or do for them. So, we had to actually bring her back and ground her and start over and then slowly expand the data sets and the amount of information.”
Alli made mistakes, but the mistakes also traced back to inconsistencies in the city’s data sets that could be corrected and improved, Wittenburg said.
The city has created AI training courses for staff to understand the different AI models and their capabilities. It also has a cross-functional group allowing different departments to evaluate the technology’s impact on the city, said Wittenburg, who is also involved with the GovAI Coalition, a group of public sector agencies that share AI innovations, challenges and policies.
Having a solid dataset and a strong grasp of the basics is key before launching any type of AI pilot, said Wittenburg, who came to Raleigh around three years ago after decades of working in IT in Tempe, Arizona.
“Before I got to Raleigh, there was a solid foundation that was already built around data, around ServiceNow, and a very strong practice,” he said. “Moving along in the AI journey was just almost like a natural progression of the evolution of the help desk.”
Wittenburg has big plans for AI in Raleigh, including a potential fully AI customer-facing kiosk at city hall. He also wants ServiceNow AI agents to resolve as much as 85% of the city’s IT support requests.
His philosophy around AI is the same as his approach to new technology overall: be curious, experiment and ask what value the technology provides to the community.
“I don't want to pilot the coolest, latest thing — which, ironically, I have been,” Wittenburg said. “But let's pilot these things knowing that this is going to add value and is either going to make staff more efficient and effective or have a positive influence on the community.”
“At the end of the day, we're public servants, we are here to serve the community, and we're here to make the community a better place,” he said.
