Intangible cultural heritage (ICH) products possess distinctive cultural value yet confront four critical bottlenecks in live-streaming e-commerce: prolonged production cycles, elevated logistics and packaging costs, limited market penetration, and the progressive aging of master artisans. Existing literature predominantly addresses isolated design or marketing dimensions and concentrates on standardized goods, leaving a notable research gap in holistic, data-driven frameworks that simultaneously optimize the entire “production–sales–logistics” chain under live-streaming contexts. Moreover, prior work has not adequately resolved the challenges of data sparsity and pronounced seasonal demand volatility inherent to non-standard ICH offerings. To bridge these gaps, this study proposes a threefold contribution: (1) a data-driven collaborative optimization framework that integrates online live-streaming platforms with offline experiential touchpoints; (2) a hybrid demand-forecasting pipeline combining CNN-LSTM for short-term demand fluctuations and Temporal Fusion Transformers (TFT) for long-term trend prediction; and (3) a hybrid recommendation algorithm that fuses collaborative filtering with content-based techniques to mitigate data sparsity while preserving cultural relevance. Empirical validation demonstrates that the proposed framework significantly outperforms traditional baselines in forecasting accuracy (RMSE and MAE) and elevates click-through and purchase conversion rates, thereby enhancing both commercial viability and cultural preservation.