SMOAD is a novel scalable model for automatic dialogue summarization. It employs a hierarchical attention mechanism to effectively capture long-range dependencies between utterances in a dialogue, and utilizes a copy mechanism to incorporate important information from the original dialogue into the summary. SMOAD is trained on a large dataset of dialogues and achieves state-of-the-art performance on several benchmark datasets. It is also efficient to compute, making it suitable for real-world applications.