Rich and real-world variants of the well-known vehicle routing problem (VRP) involve customer demands that are uncertain and influenced by contextual factors. Classical metaheuristics rely on deterministic demand estimates, while simheuristics incorporate stochasticity through simulation but generally assume fixed demand distributions. This paper proposes a novel sim-learnheuristic framework for the VRP with context-dependent stochastic demands. The proposed sim-learnheuristics integrates a biased-randomized heuristic, Monte Carlo simulation, and a machine learning model that dynamically selects the probability distribution governing each customer’s demand according to the current operational context. Combining predictive analytics with simulation–optimization allows the proposed methodology to capture both stochastic variability and context-dependent demand behavior during the search process. The framework is evaluated through an extensive computational study using benchmark instances. Its performance is compared with that of a deterministic metaheuristic and a classical simheuristic in terms of expected routing cost, route failures, safety stock requirements, and computational robustness. Results show that context-aware demand modeling improves the representation of uncertainty and produces solutions with lower expected operational costs than traditional approaches.