Internet-based recommender systems have traditionally employed collaborative filtering techniques to deliver relevant "digital" results to users. In the mobile Internet however, recommendations typically involve "physical" entities (e.g., restaurants), requiring additional user effort for fulfillment. Thus, in addition to the inherent requirements of high scalability and low latency, we must also take into account a "convenience" metric in making recommendations. In this paper, we propose an enhanced collaborative filtering solution that uses location as a key criterion for generating recommendations. We frame the discussion in the context of our "restaurant recommender" system, and describe preliminary results that indicate the utility of such an approach. We conclude with a look at open issues in this space, and motivate a future discussion on the business impact and implications of mining the data in such systems.