The Socially-Aware Robot Assistant (SARA) is an embodied conversational agent that works toward using detection of visual, vocal and verbal cues as an input to estimate the strength of its relationship (namely the level of rapport) with a user. SARA then answers to the user through similar visual, vocal and verbal behaviors with the goal of building and maintaining rapport with that user as we hypothesize that this will improve task performance and user satisfaction over time. In this paper, we report results of a field trial with a semi-automatic SARA system that took place in a large high-profile conference. Participants interacted with SARA during the whole conference, receiving recommendations about sessions to attend and/or people to meet. We analyzed these interactions to shed light on the dynamics of the rapport level between SARA and the conference attendees, and investigate how SARA's task performance would influence the evolution of rapport over time. Although we did not find evidence supporting our claim that the recommendations' outcomes would influence rapport dynamics, our findings emphasize the importance of interactional features plays in both rapport and SARA's task performance. We thus propose design guidelines for the next generation of socially aware personal assistants.