Customer Story : Colas and Konverso’s IT Virtual Agent

Colas, a subsidiary of Bouygues group, is the world leader in road construction. In the summer of 2021, the company deployed its first IT Chatbot in France with Konverso's Intelligent Virtual Agent.

Its objective is to improve the service provided to users (15,000 users in France, 55,000 employees in 50 countries), while reducing the workload of its internal IT services. This project is a major step in a roadmap launched 3 years ago at Colas, around the ServiceNow ITSM platform and the Microsoft digital workplace.


Over the last three years, Colas had already been working on self-care and knowledge management for its support teams, by deploying the ServiceNow ITSM platform. Since then, the missions of the Colas support team have evolved, and their areas of customer support and intervention have grown with the same number of staff.

Therefore, the Colas Chatbot project has two objectives: to bring innovation to users, through a new way of providing self-services, and to free up time for the service desk teams (about thirty Level 1 support agents) in order to better the customer service and the customer experience.

Colas' Knowledge Manager, Mathieu Sagot, is leading the group's Chatbot project. Integrated into the support teams, he brings to the project a "business" vision, which is important for understanding the challenges of users and IT support agents.

« My role as Knowledge Manager is to ensure that the knowledge bases are well exploited. With Konverso's Virtual Agent, which is based on a wide variety of answer sources, we offer users a self-care tool that brings relevant answers closer to them. For me, the Chatbot is a way to promote knowledge in the company. »

Mathieu Sagot, Knowledge Manager at Colas & Chatbot Project Manager.


Before selecting Konverso, the Colas team looked at several virtual assistant solutions on the market of AI technologies. One of the most important selection criteria was that the intelligent virtual agent be available and functional in Microsoft Teams. This cross-functional communication and collaboration tool has been deployed at Colas for 18 months and adopted by all teams. 

The Chatbot is present where users work to simplify their digital journey as well as their journey with support. It offers a very intuitive interface in Microsoft Teams, which reduces change management and adoption efforts. It is a multi-channel solution with a Chatbot always able to transfer a user to Live Chat/IM, as we will see in more detail in this article.


The implementation of a Chatbot will increase the performance of self-care. The first role of the Chatbot is to provide precise and relevant answers to frequent questions from users. For this purpose, Colas has connected the Konverso Chatbot to the company's knowledge bases in ServiceNow.

Konverso's language processing engine finds the right knowledge articles and offers them to users as answers to their questions. Thanks to artificial intelligence, its answers are better than those of the knowledge base search engine.  In particular, the language processing engine uses advanced similarity learning algorithms and provides an interface for training on historical data. 

The Chatbot will also better handle complex cases related to the context of users (location, equipment used, time of day, etc.). In many companies, listing these numerous use cases can make knowledge articles complex, and they can reach several dozen pages. This "raw" content is complex for users and is not used. In contrast, the Chatbot offers step-by-step guidance to users. By asking clarifying questions, the Chatbot will be able to offer a simple, unique answer adapted to the user's context.


If the Chatbot cannot answer a question, it must be able to transfer a chat conversation to a human support agent. Konverso's intelligent virtual agent must fit into a hybrid set-up where the Chatbot collaborates seamlessly with the support agent teams.

Multiple cases can arise. These transfers must be as smooth and productive as possible, both for the user and for the technician receiving the transfer request. The user will remain in their digital interface (Microsoft Teams or the Self-Service Portal) while the agent will be able to chat from the ServiceNow Agent Workplace.

Konverso's platform is very open and through its multiple integrations it allows Colas to capitalise on its past investments.

In addition, the Chatbot will enable the capture of new knowledge in the company: the expertise of support agents that has not been formalised or structured. This knowledge, generated in instant messaging during exchanges between users and support agents, can be transformed more quickly into precise answers, based on the users' use cases, and learned by the Chatbot.


With Konverso, Colas made the choice to buy a Chatbot rather than build it. Colas compared Konverso to other chatbot development platforms. By choosing Konverso, Colas gets a business solution, based on no-code and easy to use after training.

The solution integrates all the elements in a simple way, in particular a library of conversational business scenarios that can be customised and activated, a ready-to-use NLP module, and an analytics library to monitor the performance and optimisation of the solution. Colas also benefits from the methodological support of the Konverso team for the configuration and evolution of the Chatbot, to maximise its ROI and to ensure a great customer experience.

Defining the scope of a Chatbot is crucial from the start of the project. Mathieu Sagot, Knowledge Manager at Colas and Chatbot project manager, confirms that it can be very tempting to give the Chatbot too large a scope at the start. It is preferable to start with less, but on a well-tested scope, to then make continuous improvements on the virtual assistant job.

The bot will then learn automatically from conversations between users and support agents. This learning process remains under the supervision of the human technical support agents, to validate the quality of the answers provided.

This virtuous circle of continuous learning is based on knowledge bases, coming from tools like ServiceNow but also from the expertise of support agents. Moreover, the conversational qualities of the Konverso Chatbot allow it to understand questions and their context, to provide relevant answers to users and thus to improve customer experience.

« It's a virtuous circle because it's simple for the business to oversee the Chatbot's continuous learning, improve its knowledge and respond better to users in the future. »

Mathieu Sagot, Knowledge Manager at Colas & Chatbot Project Manager.

Mathieu Sagot


By framing its project with the Konverso team, Colas was able to identify and prioritise the most important subjects, with a more immediate ROI, and thus avoid scattering its efforts.

The recurring requests from users are comparable to those of most large companies. These include password requests, office software, printing, and VPN issues, which are very important in the context of Covid-19 with remote access.

Colas also secured the launch of this first Chatbot project, by integrating it naturally into the digital journey of the user and the agent. 55% of requests reach the support agents by telephone and 45% through the ServiceNow virtual agent portal.

Part of this flow now goes through the Chatbot, with the impact of reducing tickets at the Service Desk. The Service Desk team has reorganised itself by dedicating a few agents to chat, to answer complex questions in real time that are not resolved by the Chatbot.


Before the launch of its project, the Colas team took advantage of the summer to communicate extensively with its users and prepare them for the arrival of the Chatbot. The Chatbot was personalised, in the form of an avatar called "Sam", presented as a new employee soon to join the team.

This approach made it possible to explain that "Sam" is available 24/7, but also that he is just starting out. It is therefore normal that he does not know how to answer everything, but he will then transfer the requests to a support agent capable of answering.

« A Chatbot is a new usage that needs to be explained, and you shouldn't oversell it. If a first project goes wrong and users have a bad experience with a chatbot, it is very difficult to make up for it with them. If we say that the Chatbot is a magician, that it will be able to answer all requests, we will only generate disappointment. So, we explained from the start that there would be no downtime, no sticking points. And that a support agent with whom users are used to talking would always be available. »

Mathieu Sagot, Knowledge Manager at Colas & Chatbot Project Manager.



The digitalisation of the user support centre also allows for the creation of useful data for the evolution of the Chatbot. This data can be validated in real time by the experts when they interact with the users. Based on feedback from around fifty projects, Konverso's teams have measured that the Intelligent Virtual Agent could achieve a 30% support call deflection rate from the start. Then the Chatbot will benefit from improvement loops to reach 40 to 45% deflection by learning from user interactions.

These improvement loops are facilitated by Konverso's "no-code" administration interface. Support agents and knowledge managers can quickly make it their own. Thus, at Colas, after a short stage of getting used to the tool and discovering its possibilities, Mathieu Sagot confirms that he is already 95% autonomous on his Chatbot project. For specific technical points, he can always call on the Konverso team.

In 2022, Colas also plans to implement several automation processes that will be managed by the Chatbot. This will include recurring password requests and PUK code requests for phones.


Colas, a subsidiary of the Bouygues Group, has one mission: to imagine, build and maintain sustainable transport infrastructure. Backed by a network of 800 construction business units and 3,000 material production units in more than 50 countries on five continents, the Group’s 55,000 employees act locally to connect communities and foster exchanges for today and tomorrow.

Colas’ ambition is to be the world leader in innovative, sustainable mobility solutions. In 2020, consolidated revenue at Colas totaled €12.3 billion (55% outside of France).



Konverso Intelligent Virtual Agent for IT Service Desk brings to Colas the following unique capabilities:

  • Supervised Learning: The Virtual Agent extracts data from live chat to learn to respond, leveraging its Machine Learning models; human agents can also supervise this learning process. This reduces the effort, complexity and time needed to train the Virtual Agent whilst rapidly increasing the quality of conversations.
  • NLP and conversational Workflow: The Virtual Agent has conversational AI capabilities. It can handle simple to complex conversations and responses through its NLP and conversational workflow engines. To know more about natural language understanding and NLU models, deep learning and how AI understands our languages, you can read this article written by Amédée Potier, CTO and co-founder of Konverso.  
  • Analytics: Colas’ Help desk specialists can monitor the outcomes and performance of the Virtual Agent, through pre-built layered and drillable dashboards found by AI researchers. (35 preconfigured dashboards and data can be exported to analytics tools)
  • Integration: The Virtual Agent is fully integrated with Colas’ ITSM environment, to automatically leverage all knowledge, data and workflows. As a result, Konverso’s clients can automate activities as well as link to external sources of information for wider coverage.