The need of the hour is the optimization of digital marketing campaigns, especially in businesses that want to refine customer targeting and conversion rates. This paper will study deep learning-based predictive analytics for the purpose of predicting user purchase behavior based on online platform data, email analytics, and ad platforms, mainly based on the demographic, behavioral, and transactional data about user browsing history, clickstream patterns, social media activities, and email campaign response. Four deep learning models—Bidirectional Encoder Representations from Transformers (BERT), XGBoost, long short-term memory (LSTM), and ResNet50—were employed to predict the buying intentions of users. The data were preprocessed and split into 70% for training and 30% for testing. BERT was the best model, with an accuracy of 97.88%, precision of 96.5%, recall of 98.2%, and an F1-score of 97.3%, leading all the other models in every performance metric. XGBoost outscored all others at 95.6%, while LSTMs and ResNet50 scored 92.3 and 88.56%, respectively. Further AUC-ROC analysis established that BERT was superior with an AUC of 0.986, which verifies that it can distinguish between purchase and non-purchase behaviors superbly. Such results indicate that BERT is useful for more accurate and data-driven decisions in digital marketing. Companies can now optimize marketing campaigns and target the customers better, maximizing the conversion rates through deep learning models like BERT to make the advertisement campaign cost-efficient and successful.