Trace clustering is a fundamental task in process mining for identifying distinct behavioral patterns from event logs. However, many existing approaches apply clustering directly to the complete log, implicitly assuming a single coherent process and overlooking structural heterogeneity introduced by multiple business types or variants. To address this limitation, we propose a simple yet effective context-based business segmentation step, in which event logs are pre-segmented using domain-specific attributes before clustering. This semantic segmentation reduces contextual heterogeneity and enables clustering algorithms to focus on structurally comparable traces, thereby improving both clustering quality and interpretability. We evaluate the proposed approach on a synthetic event log and the real-world BPI 2017 dataset. Experimental results show consistent improvements across standard clustering quality metrics, including the Silhouette Score, Davies–Bouldin Index, and Calinski–Harabasz Index. Moreover, process models discovered from post-segmentation clusters exhibit clearer behavioral separation and more interpretable structures. These findings indicate that semantic business-type segmentation constitutes a valuable preprocessing step for coherent process variant discovery and supports downstream tasks such as prediction and concept drift detection.