An Analysis of the Barriers Preventing the Implementation of MLOps

The rapid improvements in machine learning (ML) and the increasing im-portance of ML models in numerous industries have resulted in the emer-gence of MLOps (Machine Learning Operations), a discipline focusing on efficiently managing and operationalising ML workflows. This exploratory study investigates the difficulties encountered when implementing MLOps within organisations and compares MLOps to DevOps. The study begins by conducting an SLR to identify the challenges mentioned in the literature. We then explain the results of conducting semi-structured interviews with 12 ML practitioners working across many industries, perform qualitative content analysis using grounded theory, and discuss findings. Findings are organised along four distinct dimensions: Organisational, Technical, Opera-tional and Business challenges, which are explained in eleven different themes. Our findings show that MLOps has some challenges that overlap with DevOps as well as some specific only to MLOps, like the complexity of data and model. In our discussion, we summarize these challenges and suggest future recommendations.