City of Montréal supports citizens with a virtual agent built using Microsoft Copilot Studio

The City of Montréal is the second-largest city in Canada and the largest city in the province of Québec. With over 1.7 million citizens, Montréal is also the largest French-speaking city in North America. Montréal is distinguished by its cultural vitality and world-renowned creative forces while also being known as an exceptionally inclusive and dynamic city.

The city provides timely and relevant information to its citizens through multiple channels, including a call center and public website. To enable even faster access to information, Montréal added a conversational agent to its website using Microsoft Copilot Studio. Citizens can now use natural language to inquire about a wide range of topics such as public services, administrative procedures, tax payments, and public space maintenance.

The agent draws from over 40,000 pages of information on the city’s website. Additionally, the agent integrates with two internal systems. One covers specific information on waste collection, and another provides information on city facilities such as public libraries.

In the agent’s first four months of operation, results have already been extremely positive with high customer satisfaction scores. The city anticipates that the agent will also help the call center better manage high call volumes, including seasonal spikes, enabling them to focus on higher priority items such as emergency calls.

A fully customized virtual agent

The city attributes much of the success of their new agent to Copilot Studio. “The idea of developing a virtual assistant for our citizens was already part of our responsible AI integration strategy and Microsoft Copilot Studio definitely helped accelerate that plan,” says Mohamed Arhab, Solution Architect in the IT Department of the City of Montréal. Using Copilot Studio, the team was able to develop the virtual assistant entirely on its own, without support from outside developers or consultants. They were also able to leverage existing APIs to connect to backend systems. “With Copilot Studio, we didn’t have to develop any new custom APIs. This saved significant development time and resources.”

Arhab was especially impressed with how Copilot Studio could be used to customize the virtual assistant. As he explains, “One key reason we chose Copilot Studio was the option to easily combine classic, pre-built responses of a chatbot with the AI-generated responses of an agent. This hybrid option enabled us to achieve a higher level of accuracy than just using generative AI alone.”

Currently, over 85% of conversations are handled with generative answers connected to the public website, the rest being managed by default and custom topics and API calls to back-end systems.

Direct connectivity to backend systems provides an even better experience. Before the agent was deployed, inquiries about waste collection and facilities would only return a link to a web app which required a separate search. Now, the assistant can instantly provide a customized schedule for waste collection according to a user’s location. Similarly, when getting information about a city facility, a user can ask for a specific location and time without having to run through a long schedule on the site.

To further increase accuracy and improve intent recognition, the team added their own custom entities such as Canadian postal code and a list of Montréal's boroughs.

Based on internal testing, this ability to customize the agent has helped produce a 95% accuracy rate. As usage grows and the agent is refined, the team expects to increase this accuracy even further.