Sullivan County debuts generative AI chatbot, Saige, to answer constituent FAQs
Dan Hust
Director of Communications, Sullivan County
The public sector guide to getting started with generative AI
10-step, 30-day plan to get started with your first GenAI use case.
DownloadSullivan County, New York, is home to the Catskills Mountains. It’s a great place to live, play, and raise a family, and we have a robust group of visitors and tourists. We are also a fairly small county. Being on the cutting edge of technology isn’t typically something people associate with local governments, and I’m proud of Sullivan County for consistently innovating.
Traditionally, in offices like ours, our teams are routinely answering rote questions about operating hours or document filing processes. These calls take up hours of their day they’d like to spend helping solve more complex problems. In 2023, we introduced a virtual agent powered by Dialogflow that has helped us quickly provide our constituents with answers to simple questions. We embedded the agent into our chatbot on the county website to support those seeking information about our County Clerk’s and Treasurer’s offices.
With the chatbot’s support, we’ve seen a 62% reduction in call volume. This has allowed our teams to focus on resolving more complex issues. The success of that rollout attracted the attention of the other offices in the county, and we saw a chance to expand our chatbot’s capabilities.
Simplifying generative AI with Vertex AI to overhaul a chatbot in 10 weeks
Although more than 40 offices in Sullivan County were excited to implement a chatbot, we were initially concerned about the time and effort it would take to tackle a project of that scale. Some leaders assumed that many chatbots would require manually combing through information on office websites and building chatbot workflows for each of them. Then, I was introduced to Vertex AI Agent Builder. Vertex AI Agent Builder makes it easy, even for our nontechnical teams, to train machine learning models that grow and change over time. Since it can “learn,” we can create chatbots that scrape information from the respective office sites and dynamically determine the best answer to a question on any given day. We don’t have to design any of those flows manually.
My team worked with Google Cloud Premier Partner Quantiphi to implement a generative AI-powered version of our virtual agent (who we call Saige), and their enthusiasm about the project was palpable. The project was completed and connected every department in the county in just ten weeks, thanks to the support from Quantiphi.
Improving interactions with constituents as Saige learns over time
I grew up tinkering with computers in the 1980s, and I’ve been told my entire life that computers will make our lives easier, but training people on new technologies is often cumbersome and complicated. I believe that if a task using technology is more difficult or time-consuming than doing it manually, no one will adopt that new tool. With Vertex AI, I’m seeing technology make good on its promise to improve lives.
We’ve already seen Saige grow and evolve to make our jobs easier. As it gathered information from various department sites, it helped us identify sources that needed to be updated. If it responds to a question incorrectly, I can click “This was not helpful,” and ask the question again. Saige is able to search for and find new information immediately. This is especially valuable because information changes; holidays change operating hours or a file is updated, and our small teams don’t have time to manually update every chatbot workflow.
As Saige’s knowledge repository grows, our offices continue to get more and more time back to manage more complex issues that require the face-to-face, human touch. Our offices get more efficient, people get answers faster, and every interaction we have with a constituent helps inform the next.
Setting new goals for an AI-powered Sullivan County
Hearing about new experiences from our teams and the people of Sullivan County is one thing, and it’s another to be able to truly measure impact. We’ve implemented a Looker dashboard that is helping us track exactly how much impact our chat features have on our offices. I can easily view total user sessions, success rates of chat interactions, and peak hours, so we can refine our support to best meet our community’s needs.
I can also see what questions people are asking, which helps me to understand what subjects are trending. Gathering experiential data alongside quantifiable data helps us offer better services and information about popular topics, in more prominent places, on our websites.
When I think of the future of Sullivan County’s online resources, I think about what I would want as a constituent coming to our sites for information. As our chat function continues to improve over time, I’d love to offer additional services directly through chat, such as payments or form submissions. All of these tools can come together to further augment our amazing staff and provide Sullivan County constituents with the best possible service.
Interested in seeing how the state of New York is transforming ? Learn more about how gen AI is transforming public sector services in New York.