Design Principles for Machine Learning Marketplaces in Enterprise Systems

Abstract

While standardized enterprise systems (ES) have become widely accepted, this is not the case for machine learning (ML) implementations, which are mostly developed individually in company-specific projects. Necessary historical data and rare ML capabilities result in a low cross-market ML utilization. To overcome the high usage barriers of ML, it should be incorporated into ES in a standardized manner. Therefore, we propose to implement an ML marketplace. While marketplaces in ES already exist, this paper proposes a marketplace dedicated to the exchange of ML models in a federated learning approach. Accordingly, this work formulates four meta-requirements based on interviews, which are structured by marketplace governance dimensions. With these meta-requirements, an ML marketplace was implemented in a design science research project, from which eight design principles are derived. The design principles address governance dimensions for making ML accessible to many companies and allow them to integrate ML into existing ES.