Data Sovereignty for AI Pipelines: Lessons Learned from an Industrial Project at Mondragon Corporation

Abstract

The establishment of collaborative AI pipelines, in which multiple organizations share their data and models, is often complicated by lengthy data governance processes and legal clarifications. Data sovereignty solutions, which ensure data is being used under agreed terms and conditions, are promising to overcome these problems. However, there is limited research on their applicability in AI pipelines. In this study, we extended an existing AI pipeline at Mondragon Corporation, in which sensor data is collected and subsequently forwarded to a data quality service provider with a data sovereignty component. By systematically reflecting and generalizing our experiences during the twelve-month action research project, we formulated ten lessons learned, four benefits, and three barriers to data-sovereign AI pipelines that can inform further research and custom implementations. Our results show that a data sovereignty component can help reduce existing barriers and increase the success of collaborative data science initiatives.