This paper studies informational herding models. Our contributions are essentially threefold.
First, far from being a truly new class of models capturing a new phenomenon, we show by a reverse mapping that social learning is formally but a special case of myopic single-person experimentation. As such, the incorrect herding outcome is just the familiar failure of complete learning in an optimal experimentation problem.

Second, and more naturally, we show that established notions of (limit) cascades and herds are constrained efficient. When the herding externality is correctly internalized by individuals who care about successors, or are so induced by a social planner, incorrect herds still obtain.

Thirdly and most substantively, we describe forward-looking behaviour in this model --- before we reach the limit.Individuals select actions in part to signal their private information. We show that optimality demands that any action be chosen over a single interval of beliefs. We then exhibit a set of indices that capture the privately estimated social value of every action. The optimal decision rule is simply: Choose the action with the highest index. While they have the flavour of Gittins indices, they also incorporate the potential to signal to successors. We then apply these indices to establish a key comparative static, that the set of stationary `cascade' beliefs strictly shrinks as the planner grows more patient. We also show how these indices yield a set of history-dependent transfer payments that decentralize the constrained social optimum.