City of Raleigh CIO’s ‘Crawl, Walk, Run’ Approach to AI
The management question is no longer whether AI can help a city. It is how a CIO creates enough governance to move quickly without turning every use case into a risk review. Raleighโs staged approach suggests a practical operating model: start with internal workflows, prove value, then expand to resident-facing services. That sequencing matters because it creates decision rights around what gets tested, who signs off, and when a pilot becomes a service.
For public-sector leaders, the bigger issue is portfolio discipline. AI can produce visible wins in help desks, infrastructure, and service routing, but not every opportunity deserves the same level of investment. The trade-off is between broad experimentation and concentrating funding on a few high-value cases with measurable outcomes. CIOs should ask which use cases reduce backlog, improve cycle time, or lower service risk โ and which ones mainly create novelty without operational gain.
Raleigh also highlights the importance of data ownership and human validation. Closed models built on local records can reduce exposure, but they also require clean inputs, clear stewardship, and a process for correcting model output over time. Leaders should define who owns the dataset, who is accountable when recommendations are wrong, and what evidence is needed before a model influences a public decision.
The next questions are organizational, not technical. How will the city train frontline staff to trust and challenge AI-assisted workflows? Which citizen journeys should be redesigned end-to-end rather than automated step by step? And what metrics will show that AI is simplifying government rather than merely adding another layer of tools?
Breaking Down Silos to Streamline Workflows
The drive for IT innovation didn’t start with AI. Six years ago, before modern large language modelsย and AI technologies emerged, the Raleigh IT team was already making an effort to work in a more innovative way. The IT team set out to examine siloed applications and processes across various city departments to see where those processes could be streamlined, Wittenburg said. “How do we take these siloed workflows, like hiring somebody, and then what are the pieces those workflows touch along the way?” he said, as an example. “Then we start to streamline those by sharing data and sharing workflows.”Testing AI Internally Before Public Rollout
The emergence of generative AIย triggered another opportunity to improve the way things worked. The IT teams’ AI experts started using generative AI (GenAI) internally — Wittenburg didn’t want to immediately release it to the public should a major issue arise. “There’s a couple of good examples of where that’s gone bad,” he said. By starting the AI journey internally, Raleigh’s IT team was able to begin improving its own processes first and create a more efficient troubleshooting workflow. Raleigh is using ServiceNow’s AI platform; Wittenburg said that as a result, 95% of help tickets are now auto-summarized and approved, and tickets are cleared 66% faster. By addressing tickets faster, the IT team is also able to clear out service backlogs quicker.Enhancing City Services with AI
When considering how to use AI with public-facing processes, Wittenburg took a “crawl, walk, run” approach from ideation to production. That starts with asking a lot of questions about which types of AI technologies make the most sense for the city of Raleigh, what business value AI can deliver, and what the risks are should something go wrong, Wittenburg said. The first application of a public-facing AI involved Raleigh’s water system. “We digitized all the records, brought all that in, and then built a model using data analytics, AI, and generative AI to be able to start asking questions around our water infrastructure,” Wittenburg said, noting that the city’s paper records dated back to the 1700s. The AI model could then be developed and used to identify and predict potential failures in the water infrastructure. In addition, the approach allowed for the creation of a closed model using Raleigh’s own water data with an internally built AI engine, which is a low-risk method for testing AI before deploying it widely, Wittenburg explained. City employees could also verify the AI’s predictions and improve the model’s forecasts over time.Simplifying Government Complexity
Next, the IT team integrated ServiceNow’s AI platform with the city of Raleigh’s public chatbot to implement “agentic pieces to it,” Wittenburg said, referring to AI technology capable of autonomous decision making. “It was amazing, because now it started to use the knowledge articles, but it also started to dig into those closed tickets to other workflows, and to start to put all that together. It’s just amazing watching it learn things that we didn’t know we were doing,” he said. While learning internally has been critical, the ultimate goal of deploying AI is to bring value to the community, Wittenburg said. In that vein, Raleigh’s IT team plans to first use agentic AI to improve traditional city services, such as simplifying the process of identifying and fixing potholes. Fixing a pothole seems straightforward enough, but it involves a number of steps, Wittenberg said, starting with identifying if it’s on a city, county or state street — three different agencies with separate phone numbers. People just want the pothole fixed, the problem solved. “What if we could get all those AIs and those agents and all these services to hide the complexity of government from you all,” he said.Vision Zero: Agentic AI for Safer Transportation
Another use case the Raleigh IT team is working on with AI is “Vision Zero,” using machine learning at traffic intersections to create models around vehicle and pedestrian interactions. The goal is to reach zero transportation-related fatalities in the city. For Vision Zero, Wittenburg’s team partnered with Nvidia and Esri, a GIS mapping software, to build an AI model that could determine if drivers were going the wrong way or if a car narrowly missed a pedestrian, for example. The IT team is beginning to add GenAI capabilities to ask the AI model to suggest improvements at intersections based on “near misses.” The AI model has made suggestions such as adding a right turn light or implementing light change delays, Wittenburg said.Guidance for CIOs
Ultimately, when deploying AI for the community, Wittenburg said it’s important for constituents to have a consistent experience and reduce complexity when they need to connect with a local agency. His advice to other CIOs and IT professionals starting out on their IT journey is to identify the governance and “guard rails” for using AI, have a solid data foundation and “don’t let fear hold you back.”Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.