SmartVisual: AI-Enhanced Personalization of Digital Marketing Campaigns via Deep Image Understanding

The growth of visual content in digital marketing has led to the need to have hyper-personalized campaigns that emulate individual user tastes. This paper introduces a multimodal deep learning-based customer behavioral embedding framework enhanced by AI, Smart Visual, which can provide a personalized recommendation of ads. To extract hierarchical image features, Smart Visual uses Convolutional Neural Networks (EfficientNet and ResNet) and Transformer-based attention layers to extract semantic and contextual information between marketing images. Joint multi-modal embeddings comprising of customer embeddings based on clickstream data and demographic data are combined with image representation. Moreover, a Generative Adversarial Network (GAN) is used to create visually attractive versions of ads that apply to individual customer groups. As it has been revealed through experiments, Smart Visual is a much stronger tool compared to traditional static-only campaigns and single-mode AI models in terms of engagement (10.93 %), clicking (16.2), conversion (6.9), and recommendations (87.95). These findings suggest that bringing together visual, semantic, and behavioral information, with dynamic image enhancement, is the key to successful hyperpersonalization. The suggested framework offers an end-to-end, scalable AI-based digital marketing framework that provides a connection between content and user preference prediction and creates benchmarks to continue the research in the area of personalized visual marketing.