Understanding customer behaviour is a challenging problem. While the customer produces a large amount of data with each touch point, most of the proposed models focus on one data source in their predictive analysis approaches. This research proposes a customer profile model based on 360 customer view. To this end, we first model a simplified data model and the basic entities based on the existing models. Then, we perform extensive feature engineering techniques, including extracting new features and transforming features to enhance their behaviour in the predictive model. Through the experimentations, we show that the models based on graphs achieve good performance. To this end, we propose a graph-based neural network capable of multitasking without sacrificing the task's performance. We examine three tasks to predict customer intentions. The final results reveal that the set of features with customer information from different data sources positively influences the predictive algorithms' performance.