Data-intensive technologies draw high investments. Yet, data science projects are reported to suffer from poor collaboration, unrealistic expectations, and difficulties in realizing practical solutions between business and data science units. Moving beyond the currently prevalent approach to study data science practices, our study emphasizes the use of boundary objects between data science and collaborating fields. We interviewed collaborators from diverse fields in six organizational data science initiatives. Our inductive analysis of this rich data source uncovered six distinct mechanisms and six archetypes of boundary objects in data science projects. While archetypes that we label Alignment, Temporary, Collaboration, and Outcome are procedural and appear in selective stages of the data value creation process, the archetypes Infrastructure and Upskilling support projects along the value creation process. The archetypes and their mechanisms inform the management of data science initiatives, help to advance boundary object theory, and provide instruments to study data science initiatives.