Behavioral AI for enterprise customer lifecycle management : from transactional systems to explainable predictive decision support
View/
Keywords
Customer Churn ; Behavioral AI ; Customer Segmentation ; Predictive Analytics ; Explainable AI ; Machine Learning ; Customer Lifecycle Management ; Behavioral Modeling ; Enterprise AI ; Churn PredictionAbstract
In enterprise environments, customer churn extends beyond a purely predictive problem, representing a behavioral and decision-making challenge that directly affects customer lifetime value, operational efficiency, and long-term growth. This complexity becomes particularly pronounced in digital healthcare and online pharmacy ecosystems, where customer interaction patterns differ across Prescription (RX), Over-the-Counter (OTC), and FREE product interactions.
This thesis develops a behavioral AI framework for customer churn interpretation, prediction, and decision support within a real-world online pharmacy environment. Rather than relying on static inactivity thresholds or purely model-driven approaches, the study begins with reconstructing customer behavior from transactional systems and progressively builds toward segmentation, personalized churn interpretation, predictive modeling, and operational decision logic.
The analysis is based on a large-scale transactional dataset covering 122.8 million order lines, 45.2 million completed orders, and 10.9 million customers over a 24-month period. Customer behavior is modeled at the ordering-customer level using leakage-safe behavioral signals derived exclusively from observable transaction patterns. Particular emphasis is placed on return dynamics, purchasing consistency, recency-frequency behavior, and structural differences across customer segments.
The findings demonstrate that inactivity cannot be interpreted uniformly across the population. Distinct behavioral rhythms exist across segments and individual interaction patterns making static churn definitions unreliable. To address this, the framework introduces segment-specific and personalized churn definitions based on empirical return behavior and customer-level expected interaction patterns.
Building on this foundation, the framework is extended into predictive Machine Learning (ML) models that forecast future customer states (ACTIVE, AT RISK, CHURNED) using behaviorally grounded features. Beyond predictive performance, the work emphasizes explainability, scalability, and operational usability.
The thesis concludes with the design of a decision-support framework that connects behavioral analytics, predictive intelligence, and operational prioritization. The proposed approach demonstrates how transactional systems, behavioral modeling, and ML can be integrated into a coherent, explainable, and scalable architecture for enterprise customer lifecycle management.
Postgraduate Studies Programme
Πληροφοριακά Συστήματα και ΥπηρεσίεςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
143Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα