The construction company announced in 2024 a partnership with Google Cloud around AI. On the occasion of Google Cloud Next, its CIO Jean-Philippe Faure has drawn up a first step.
For a construction giant like Eiffage, the adoption of new artificial intelligence tools is not done anything. It is not only a question of making tools available to employees, they must also be taught to use it correctly: “At Eiffage, we use four different types of LLM”, explains the DSI of the Jean-Philippe Faure group. “We propose Imagen and Gemini from Google, the LLM of Mistral AI, but also Claude. And users can choose their solutions according to the use case that interests them, but they are also offered a robot to guide their choices according to their need,” summarizes the director.
Prompt, but effectively
Eiffage puts these artificial intelligence tools available to just over 2,500 employees within its teams, subject to having first validated a dedicated training of around 80 minutes. For the moment, 2,600 employees (out of just over 84,000 employees throughout the group) have already passed this training. But Eiffage does not intend to stop there: “We do not want people to make stupid prompts. These training courses are essential: we would like to have very high level prompters.” If for the moment, the training remains limited, Eiffage would like to be able to set up a workshop, “can be on two or three days”, in order to develop even further the skills of its employees on artificial intelligence tools.
Improving prompt is also one of the main challenges of a partnership with Google Cloud announced in 2024: Eiffage has set up a no code platform called Workflow Builder to help employees achieve more effective prompts to obtain the desired results. “What is needed, on a prompt is never to take the first result. Because the machine seeks to consume as little as possible, and the first result is generally a fuel result. You have to ask it to enrich, enter a tennis game with the machine, until you get a satisfactory result,” says Jean-Philippe Faure.
Understand uses and control consumption
And this population of Early Adopters is observed closely in order to understand what are the main use cases of artificial intelligence tools in the company. Unsurprisingly, Eiffage employees use AI as you and me: “What we observe is that 40% of prompts aim to improve texts, 40% aim to make decision-making, and we also see in 5% of cases relating to computer code”, confides Jean-Philippe Faure. The improvement of text, and especially its standardization, has an interest in a company of this size, which does not really seek a “literary style” or “a complex thought”, but a transmission of clear and understandable information, without frills. A style for which language models are excellent. Unsurprisingly, language models are used to generate account meetings reports. Among the other uses put forward by Eiffage, visual recognition tools are for example used to analyze the images taken on construction sites in order to quickly identify differences in safety rules.
Obviously, the other aspect observed closely is the consumption and the use made of these different tools, which generally charge at the request. To avoid the well -known drifts of the cloud, Eiffage therefore monitors its users closely: “We observe consumption, we have a reporting tool that allows us to know it to the near Euro, by consumer, and as soon as we see someone who consumes too much, we call him to understand what happened,” explains the DSI. On average, users of AI at Eiffage are entitled to a credit of 5 euros per month, 10 euros for the biggest users or even 30 for certain specific profiles, “but it is very rare to see this kind of scenario”, assures Jean-Philippe Faure.
