Dynamic Graph Model of the Customer Journey as a Tool for Event-Based Analysis and Decision Support

This paper addresses the problem of formally representing the customer journey in commercial processes, where customer touchpoints are distributed across advertising channels, websites, CRM systems, communication services, sales, payment, and post-purchase support. Traditional sales funnels and static customer journey maps provide a high level of managerial interpretability but fail to adequately capture repeated visits, skipped stages, varying progression rates, and changes in customer behavior over time. To address these limitations, the study aims to develop and experimentally validate a dynamic graph model of the customer journey extracted from a unified event log. An algorithmic framework is proposed that includes event normalization, construction of a directed weighted graph, filtering of stable transitions using an adapted Heuristic Miner approach, computation of structural and temporal distances between trajectories, DBSCAN clustering, graph drift analysis, and generation of Next Best Action recommendations. The experimental evaluation was conducted using a synthetic event log comprising 1,200 customer trajectories and 7,236 raw events. The resulting graph contains 37 states, 198 observed transitions, and 146 stable edges. The experimental results demonstrate that the proposed model enables the identification of problematic states, estimation of the probabilities of direct customer loss, detection of both typical and anomalous customer journeys, and association of diagnostic findings with the corresponding managerial control domains. The scientific contribution of the study lies in the transition from a descriptive customer journey map to a computational graph-based representation suitable for continuous monitoring, explainable analytics, and decision support.