Mémoire-projet

Abstract

This thesis is set against the backdrop of the accelerating digital transformation of industrial organizations and the growing diffusion of generative artificial intelligence tools in B2B commercial environments. Given the limitations of universalist frameworks drawn from the existing literature which presuppose abundant, homogeneous and structured data, this work examines the real conditions for integrating AI in a high contractual-intensity environment, where projects are singular, data is largely tacit, and confidentiality concerns are structural.

This thesis focuses on the case of John Cockerill Metals, an industrial engineering entity within the John Cockerill Group, specializing in the design of complex steel equipment, operating in Wallonia, Brussels and internationally. The research question is: 'How and to what extent should artificial intelligence be integrated into the industrial Sales & Proposal process in order to maximize operational benefits while managing organizational and confidentiality risks?'

To address this, a critical literature review was conducted, drawing on theoretical frameworks relating to AI as a commercial performance driver in B2B contexts, the organizational conditions for successful integration, including AI capability, systemic digital transformation and data quality, as well as governance, risk and confidentiality. This theoretical framework was compared with the results of an exploratory qualitative study conducted through semi-structured interviews with eight John Cockerill Metals employees from the Sales, Proposal, Engineering, IT and Knowledge functions, complemented by direct observation of the process and documentary analysis.

The results reveal that documentary repetitiveness constitutes the main friction point in the process, and that the most immediately exploitable AI use cases are concentrated in mechanical and analytical tasks with low contractual sensitivity. Usage maturity varies significantly across actor profiles, and data confidentiality emerges as a structural barrier to the adoption of external tools. Furthermore, training and change management prove to be decisive variables, consistently underestimated in the existing literature.

Based on these findings, concrete recommendations have been formulated for John Cockerill Metals, structured around three axes: structuring data to make AI operationally deployable, creating the roles and responsibilities necessary for effective AI governance, and deploying mandatory, differentiated training grounded in the day-to-day practices of the teams concerned.