This paper presents a multiagent architecture and methods for intelligent decision support in logistics processes. It extends current advanced prediction systems by providing the ability to combine history and situated reasoning. The contribution of the paper is threefold: first, a multi-agent architecture and learning algorithms are developed that enables us to combine background models learned from history data with context-related knowledge about the current situation; second, using a large real data set we show that adding situated knowledge actually improves the performance of a supply chain decision support system; and third, for our settings we evaluate the degree to which agent-assisted decision support is actually usable/sufficient to improve human decision-making and to support automated decision-making in dynamic supply network management scenarios.