Bulk terminals are complex environments due to a number of variables that affect terminal performance. Although the analysis of big datasets is destined to become an important component of terminal management, previous research has not addressed this issue yet. This paper aims to shed new light on the operation of dry bulk terminals through a two-stage method based on unsupervised machine learning techniques. The first step gives an overview of the terminal's performance, revealing the strongest associations between the variables, while the second calculates an anomaly score for each vessel through an optimised implementation of the isolation forest. As a result, we detect anomalous services which could be directly attributable to the terminal operator. This method can be used to increase transparency in service and assist the terminal operator and ship agents in future contracts.