Realising the promises of artificial intelligence in manufacturing by enhancing CRISP-DM

Artikel i vetenskaplig tidskrift, 2024

To support manufacturing firms in realising the value of Artificial Intelligence (AI), we embarked on a six-year process of research and practice to enhance the popular and widely used CRISP-DM methodology. We extend CRISP-DM into a continuous, active, and iterative life-cycle of AI solutions by adding the phase of ‘Operation and Maintenance’ as well as embedding a task-based framework for linking tasks to skills. Our key findings relate to the difficult trade-offs and hidden costs of operating and maintaining AI solutions and managing AI drift, as well as ensuring the presence of domain, data science, and data engineering competence throughout the CRISP-DM phases. Further, we show how data engineering is an essential but often neglected part of the AI workflow, provide novel insights into the trajectory of involvement of the three competences, and illustrate how the enhanced CRISP-DM methodology can be used as a management tool in AI projects.