Streamlining the Operation of AI Systems: Examining MLOps Maturity at an Automotive Firm

Developing and operating AI systems based on machine learning (ML) has unique challenges that render traditional practices inappropriate (e.g., managing data drift). To that end, MLOps emerged as a novel paradigm for managers and teams to develop and operate such ML systems successfully. Organizations currently employ different maturity levels for MLOps, whereas higher maturity typically corresponds to more automated, streamlined, and reliable workflows. However, we have limited insight into factors influencing MLOps maturity in ML projects. Therefore, we conducted a case study on MLOps maturity in three ML projects at an automotive firm. We identified several contextual factors that facilitate or inhibit MLOps maturity, such as the ML model’s complexity, the quality of new data, and the appropriateness of available MLOps tools. Our study contributes to research on managing and organizing AI by providing factors that explain the different adoption of MLOps in practice.     «

Developing and operating AI systems based on machine learning (ML) has unique challenges that render traditional practices inappropriate (e.g., managing data drift). To that end, MLOps emerged as a novel paradigm for managers and teams to develop and operate such ML systems successfully. Organizations currently employ different maturity levels for MLOps, whereas higher maturity typically corresponds to more automated, streamlined, and reliable workflows. However, we have limited insight into fac...     »