This paper presents an innovative, data-driven framework for determining the Economic Smart Specialization (ESS) of local communities through the use of discrete-time Markov chains. By synthesizing three critical layers of information—intrinsic industrial specialization derived from NACE-based revenue data, inter-industry supply chain flows–the proposed model captures the complex interplay between economic structure and geographic proximity. Specialization is modeled as a stochastic process, enabling simulation of how sectoral strengths evolve and diffuse across regions. The framework yields a normalized ESS matrix that provides a probabilistic index of specialization at the locality level, informed by both upstream and downstream industrial relationships. Through Principal Component Analysis (PCA) and relative standard deviation metrics, the model quantifies the diversity and complexity of economic specialization under multiple transition scenarios. Applied to more than 3000 localities in Romania, the methodology demonstrates high scalability and adaptability, offering policy makers a powerful tool to design targeted, resilient, and context-sensitive smart specialization strategies. This research bridges regional economic modeling and machine learning, contributing a reproducible, interpretable, and policy-relevant approach to enhancing territorial innovation potential in line with the principles of the Smart Specialization Strategy (S3).