Le cas Michelin : 114 401 réponses, 2 963 563 mots

In the aftermath of the pandemic, the Information Systems Department of Michelin Group is questioning the actual practice of digital work among white-collar employees. How can we characterize digital work practice in the field? What are the expectations and needs related to digital work expressed by office workers? We have translated these industrial inquiries into the following research question: "How to observe digital work practice among white-collar employees from a big quali corpus?". We proposed to address these questions using an existing textual database of over 100,000 responses to two open-ended questions from an annual job satisfaction survey. Consequently, we explored the literature on digital work and textual data analysis. We then proceeded in two distinct methodological phases, chronologically separated by the emergence of mainstream generative AI. After performing top-down classifications, our first "classic" approach proposes a manual qualification of classes (by coding a part of the sentences). In contrast, the second approach leverages the analytical capabilities of ChatGPT 4.0 to perform the same qualification. The following research question thus arises: "What is the difference between the 'classic' class qualification through manual coding and class qualification assisted by AI?". Our work presents several contributions. On one hand, our research proposes a methodological approach for researchers and practitioners, allowing the longitudinal observation of a complex phenomenon from a Big Quali corpus. On the other hand, our work demonstrates the interest of class qualification assisted by artificial intelligence. Finally, the rapidness and neutrality induced by AI allowed us to perform consecutive top-down classifications enabling thematic deepening.