Government chatbots are an increasingly essential tool for communication with residents. While some argue that they are too error-prone to be effective, training an AI model specifically on clean government data can improve output.
In Amarillo’s case, the city of more than 200,000 was able to delay a $1.8 million investment in hiring new call center agents thanks to Emma, according to Rich Gagnon, its CIO and assistant city manager. Call center agents in the city take about 32,000 questions a month. Deployed in December 2024, Emma is now answering 16,000 of those, and officials plan to increase that number.
In the video clip below from the city of Amarillo, Gagnon talks about Emma, who demonstrates her ability to respond to a question. Users can pose questions either by text or by asking out loud.
The vision for the tool came from community engagement efforts around digital inclusion. Only about 2 percent of residents do not have access to an Internet connection, compared to about 40 percent prior to the COVID-19 pandemic, Gagnon said. During the city’s work to expand access, community engagement highlighted the many languages and dialects residents speak — as many as 62, the CIO said. This led leaders to explore communication solutions that enabled multilingual access, ultimately leading to the AI tool’s implementation.
Emma’s current capabilities include English and Spanish, the most commonly spoken languages in Amarillo, but five more languages are coming online soon, Gagnon said, noting the city worked with nonprofit organizations that serve refugees in the community to identify the next languages to prioritize.
AI-generated information is not always accurate, but Emma is trained on government website data, ingested every 24 hours with the capability for officials to refresh at any time. Officials also improved their website content to leverage plain language. This allows emergency messaging such as flood information to be distributed in a timely manner.
UneeQ helped the city design the digital human image that was representative of the diverse community of more than 200,000 people it would be serving. Call center staff and community groups did too. Officials hoped that by using a digital human image rather than a text-based chatbot, Gagnon said, trust could be built.
“Conversational AI starts with listening,” Tyler Merritt, UneeQ’s chief technology officer, said, explaining that the digital human provides the experience of being listened to more than a traditional chatbot. The trust component is often missing in government AI implementations, he said: “Once AI proves that it could be useful and helpful, then trust is established.”
To further build trust, the city’s Digital Dignity, Rights and Privacy Ordinance guided the tool’s implementation, Gagnon said. The city collaborated with the organization Digital Rights House, members of the public, and an advisory board to ensure the tool was something the community would accept and feel safe using.
City staff can train people who lack digital literacy skills to effectively use Emma, but Gagnon said senior testing revealed a high level of interest in engaging with Emma — and not just about city services, but about their grandkids, recipes and more. Now, officials are working with a nonprofit organization and higher education institutions to pilot a senior companion.
By interacting with Emma, users provide city government data — albeit, in an anonymous way — that informs decision-making by providing insight into the kinds of questions that were asked and whether the answers satisfied them, Gagnon said.
Analyzing large amounts of textual data can be time-consuming, so the city is working with analytics companies to better leverage the insights made possible through Emma’s interactions with constituents, and to better understand trends.
Emma was programmed to not remember contextual information from previous conversations with users, but some individuals expressed interest in having her remember those interactions. City officials are exploring adding a sign-in capability as an optional feature for users who prefer that.
“We actually have to manage her like a vendor does product,” Gagnon said, indicating the tool itself drives its path forward based on the insights it enables.
Her capabilities are expanding into other areas of service, too, like emergency services or being able to tell users whether a certain book is available at the public library. Officials are also inputting additional city data, like the charter, ordinances and regulations, to help Emma answer questions about building and home improvement.
Currently, users ask questions and Emma provides answers — but officials are starting to explore using Emma to ask questions to members of the community to get rich, qualitative data about sentiments and needs, resulting in what Gagnon referred to as a “24-7 insight engine” for city leadership.
“I don’t think she’s ever going to be done,” Gagnon said.
