Business Process Management (BPM) plays a key role in operational excellence and organizational agility; however, traditional process diagnosis methods, relying on manual, static, and retrospective analyses, are increasingly inadequate for addressing the complexity of data-driven processes in the digital transformation era. This study aims to propose, explain, and evaluate an integrated, intelligent, and multi-layered model for applying artificial intelligence algorithms to Business Process Management in order to establish a dynamic and prescriptive decision support system. Grounded in the Design Science Research (DSR) approach, the proposed model is architected as a five-phase processing pipeline. First, a heterogeneous data preprocessing and ingestion layer integrates structured event logs and extracts implicit process knowledge from unstructured textual data using Natural Language Processing (NLP). Second, the Apriori algorithm is employed to discover dependency rules and frequent patterns. Third, DBSCAN is applied for density-based behavioral clustering. Fourth, graph-based process modeling and Social Network Analysis (SNA) are conducted to identify structural patterns and process bottlenecks. Fifth, a prescriptive reasoning engine and dynamic decision support system are developed based on an innovative Process–Role Relationship Matrix aligned with the APQC reference framework. The proposed model was developed in Python and comprehensively evaluated on the large-scale BPIC 2017 loan application process dataset, comprising 1,202,267 events and 31,509 unique cases. Experimental results demonstrated that DBSCAN identified 18 standard behavioral clusters and intelligently isolated 165 highly critical outlier cases, characterized by an average of 4.89 rework cycles and 1,309 hours of processing time. Graph-based analysis accurately identified the A_Validating activity as a central structural bottleneck and revealed a detrimental ping-pong cycle between A_Incomplete and A_Validating, occurring 24,986 times across 11,668 cases (37.03% of all cases). Finally, the prescriptive engine generated targeted operational recommendations for process redesign and intelligent redistribution of organizational workload based on risk scoring.