The proposed work investigates the application of deep learning tools, specifically Graph Neural Networks (GNNs), to combinatorial optimization problems within the context of the Stellantis company. The work is divided into three main sections: an in-depth study of combinatorial optimization and GNN architectures, a review of how GNNs can solve such problems, and the implementation of a GNN model on a toy problem inspired by a business scenario of the company. The problem involves organizing orders into loading lists for delivery trucks, optimizing the overall load, minimizing stops, and reducing travel distance, while adhering to specific constraints. This makes the problem a combinatorial optimization problem, which is a class of problems known to be NP-Hard, making exact algorithms that guarantee optimal solutions computationally infeasible at a large scale.