The contact hypothesis posits that interacting with outgroup members can reduce prejudice and improve intergroup relations. While the overall effects of contact have been found to be positive, some studies have found null or even negative effects. We aim to contribute to the understanding of the determinants of contact interventions, by unbundling the bundle represented by contact interventions and testing for the effects of a common component of all existing contact interventions, namely bilateral discussions. We take advantage of the controlled environment of lab-in-the-field experiments and implement a ``fast-friends'' intervention, which aims to create interpersonal closeness quickly, and can be implemented in a wide variety of situations. We run our experiment with a diverse sample from northern Senegal. Our intervention is found to be effective in increasing interethnic trust toward the specific individuals the individual meets. However, we find a backlash effect at the collective level, with contact increasing interethnic prejudice. We analyze the heterogeneity of treatment effects with respect to a predetermined dimension - previous contacts - and using machine learning techniques for dimension selection to avoid overfitting. We find no evidence of heterogeneity in either individual-level trust or outgroup prejudice.