Prescriptive analytics systems (PAS) represent the most mature form of business analytics (BA), offering advanced decision support. However, current research predominantly focuses on technical facets while neglecting social-technical design aspects. A pluralist research methodology was employed to address this gap, starting with a systematic literature review of over 200 papers. We used these papers to derive a concept matrix of fundamental elements that guided the development of a taxonomy conceptualizing the interplay and collaboration between humans and machines in PAS-based decision-making processes. Based on this taxonomy, we identified four recurring PAS archetypes with salient design characteristics: Informative, executive, adaptive, and autonomous PAS. Our findings have important implications for the BA community, including the need to investigate design options for executive, adaptive, and autonomous PAS; the underrepresentation of the human perspective; the missing links to the broader organizational landscape; and the potential for interpretable machine learning and reinforcement learning in PAS.