Optimizing B2B customer relationship management and sales forecasting with spectral graph convolutional networks: A quantitative approach

Customer relationship management (CRM) in business-to-business (B2B) environments requires robust strategies and informed decision-making to cultivate strong inter-business relationships, which are pivotal for achieving competitive advantage and maximizing profitability. Traditional CRM analytics, which leverages conventional data mining, machine learning, and deep learning techniques, often fails to address the intricate and interdependent nature of B2B systems. To overcome this limitation, we proposed a spectral graph convolutional neural network (GCN) approach that utilized graph-based modeling to capture the structural complexity of B2B CRM. Companies were represented as nodes, and their interactions as edges within a graph, enriched with Eigenvector centrality and shortest-path graph features, which were particularly suited for spectral GCN operations. Using graph Laplacian-based convolutions, the spectral GCN effectively aggregated global and local relational information, enabling accurate and scalable B2B sales predictions. Experimental evaluations demonstrated that GCN models with spectral attributes significantly outperformed state-of-the-art machine learning and deep learning models, including random forests, convolutional neural networks, feed-forward neural networks, Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CATboost), in terms of accuracy, F1 Score, precision, and specificity. Among the models, the GCN with Eigenvector features achieves the best classification performance, with a high Receiver Operating Characteristic (ROC) value of 0.924, further demonstrating its robustness against variations in feature correlations.