We look at the effect of transparency on the incidence of costly back-scratching in a laboratory setting by implementing player identification via photographs. In our experimental design players have an incentive to form bilateral alliances in which they favour their partner at the expense of others. We find no improvement in overall group payoffs from transparency. A plausible story that fits our results is that there may be two countervailing forces at play. First, more rapid alliance formation due to social cues from the photographs being used as a coordination device to facilitate faster alliance formation between some players. Second, shorter alliances due to prosocial forces at the group level. We draw out lessons for policy makers about the limits of transparency in curtailing "grey" types of corruption.