Agents sequentially choose between options A/B using private signal + observing predecessors' choices. Bayesian updating can produce herding: agents ignore private info and follow the crowd — even to the wrong answer.
Bikhchandani, Hirshleifer & Welch (1992): herds form with probability 1. Once 2 more agents chose one option than another, all subsequent agents herd — ignoring private signals. Cascade can lock in on wrong answer. Information aggregation fails.