In digital markets with peer-to-peer reviews, new products encounter the so-called "cold-start" problem. Little-known products are bought too rarely and remain little known. In this paper, we examine the inefficiency associated with social learning on Airbnb, including its implications for hosts’ price, entry and exit decisions. We estimate a dynamic structural model of demand and supply for Airbnbs in Manhattan, New York. We find that addressing the cold-start problem by lowering the price of new listings relative to incumbent ones leads to a welfare increase exceeding 8% of total host revenue.