Amazon uses deep learning algorithms to generate demand forecasts for over 400 million products daily, predicting how much of each item to stock in which fulfillment centers — reducing excess inventory, preventing stockouts, and enabling faster delivery.

Amazon's forecasting team has evolved through multiple generations of machine learning models, from random forests to the Multi-Horizon Quantile Recurrent Forecaster (MQ-RNN) and transformer-based architectures drawn from NLP research. The current system analyzes sales trends, customer behavior, social media signals, weather data, and regional events. Amazon Science has published detailed accounts of this forecasting evolution. The system enables Amazon to pre-position inventory closer to likely buyers, underpinning its ability to offer same-day and next-day delivery at scale. In 2024, Amazon delivered over 9 billion items same-day or next-day globally.