Firms with privately known types may have reputational incentives for effort, if customers share their experiences on a social network. The structure of the social network matters for incentives; more connected customers may be more influential but harder to influence. We model a firm with privately known skill who serves a sequence of networked customers, and can exert non-contractible effort to increase the probability customers receive good service. Customers observe the purchase decisions and experiences of their neighbours, and purchase if they are sufficiently optimistic about the firm’s skill. We show that for many networks, equilibria featuring high effort for early customers also feature herding; later customers copy the purchase decision of critical earlier customers, and so the firm’s effort choice for later customers is irrelevant. If customers are located on a directed rooted tree, herding occurs immediately and the firm exerts high effort for at most one period. For general networks, we propose a method to identify the critical customers whose purchase decisions are influenced by the firm’s effort choice, and to identify the customers who will herd. We give conditions for there to exist a unique equilibrium strategy for the firm.