With the explosive growth of Over-the-Top (OTT) platforms, reducing subscriber churn has emerged as a central problem for service providers. This paper introduces a single framework for churn prediction and customized subscription plan recommendation, capitalizing on the advantages of Graph Neural Networks (GNN) and Reinforcement Learning (RL). Churn prediction was carried out on four models: Random Forest, XGBoost, Long Short-Term Memory (LSTM), and GNN, with the GNN model showing better ability to detect potential churners by being able to model rich relationships in user data. For personalized recommendations, both Cosine Similarity-based and Deep Q-Learning (DQN)-based approaches were utilized, with DQN providing dynamic, user-specific plan recommendations. The model was trained and tested on a balanced dataset with 147,269 training samples and 25,000 test samples, and performance was gauged using accuracy, recall, F1-score, and AUC-ROC. Analysis shows that the use of GNN for churn prediction and DQN for recommendation enhances the efficacy of retention mechanisms and increases user satisfaction. The technique empowers OTT providers with an effective means of proactive user interaction and churn prevention.