Consensus decision-making, a widely utilized group decision-making process, requires the consent of all participants. We consider consensus stopping games, a class of stochastic games arising in the context of consensus decision-making that require the consent of all players to terminate the game. We show that a consensus stopping game may have many pure stationary equilibria, which in turn raises the question of equilibrium selection. Given an objective criterion, we study the NP-hard problem of finding a best pure stationary equilibrium. We characterize the pure stationary equilibria, show that they form an independence system, and develop several families of valid inequalities. We then solve the equilibrium selection problem as a mixed-integer linear program by a branch-and-cut approach. Our computational results demonstrate the effectiveness of our approach over a commercial solver.