This paper introduces a novel unsupervised method for detecting lead-lag clusters in multivariate time series, applied to US equity data. The method constructs a directed network from pairwise lead-lag metrics, then applies Hermitian clustering to identify clusters with high flow imbalance. The authors evaluate several lead-lag metrics (e.g., cross-correlation functionals and signatures) and clustering algorithms on synthetic data, finding that Hermitian clustering with Pearson ccf-auc performs best. On US equity data, the method uncovers statistically significant clusters that align with industry sectors and market capitalization, with large-cap stocks leading small-cap stocks. A trading signal based on cluster leadingness generates significant out-of-sample returns, demonstrating practical applicability. The work is the first to provide data-driven clustering of lead-lag networks in finance and shows promise for forecasting in noisy, high-dimensional settings.