Political clientelism is a dyadic relation in which a politician (the patron) gives material goods and services to a citizen (the client), in exchange for political support. If, at different stages of this relationship, both the patron and the client have incentives to defect and not honor informal agreements, what makes clientelism self-enforcing? The following paper presents a gametheoretical model of political clientelism in which a candidate disciplines a majority of voters through the promise of a future flow of benefits. A mixed strategy involving a randomized allocation of resources among constituencies makes clientelism feasible when the politician’s action is contingent on the result of the election. Higher campaign budgets and lower voter aversion towards clientelistic parties, as well as higher patience and higher heterogeneity across groups of voters, make clientelism more likely. Swing voters tend to be gifted more frequently than core supporters with this frequency increasing as group heterogeneity increases, presenting a positive association.