The article examines the evolution of machine learning methods in marketing, with a focus on the development of personalization technologies and consumer demand forecasting. The historical stages of methodological transformation are analyzed – from statistical tools for segmentation and clustering to modern deep learning and reinforcement learning models, which enable the consideration of individual behavioral patterns and exogenous factors. The importance of transitioning from static models to dynamic systems that integrate behavioral, content, and contextual data is emphasized, as it creates new opportunities to enhance the effectiveness of marketing strategies. The practical part of the study is based on experimental modeling using an open retail sales dataset. A comparison was conducted between traditional statistical methods (linear regression, ARIMA) and modern machine learning algorithms (gradient boosting, RNN, LSTM, GRU) using MAE, RMSE, and R2 metrics. The results confirm that modern algorithms provide higher forecasting accuracy and greater robustness to market fluctuations, enabling companies to manage resources more efficiently, personalize offers, and increase customer loyalty.