Uncovering the Contingencies of MLOps Adoption: A Literature-Based Analysis

Kurzfassung

The increasing adoption of Machine Learning (ML) applications presents significant challenges for organizations with regard to their development, deployment, and maintenance. The paradigm of MLOps (ML Operations) seeks to address these issues by promoting principles such as automation, reproducibility, and collaboration throughout the ML lifecycle. However, the implementation of MLOps is often perceived as difficult and complex. Therefore, this study draws on contingency theory, a well-established organizational theory, to examine factors that influence the adoption of MLOps. A structured literature review identified 38 relevant scientific articles. Through the application of inductive coding, five MLOps contingencies, namely data, governance, lifecycle management, organizational readiness, and technical infrastructure, were revealed and presented in a conceptual model. The discussion of the findings and limitations highlights substantial potential for future research, including the development of MLOps guidelines and empirical studies investigating the specific impact of the identified contingencies on successful MLOps implementation.