Using a large administrate dataset covering the universe of phone calls and airtime transfers in a country over a four year period, we examine the pattern of adoption of airtime transfers over time. We start by documenting strong network effects: increased usage of the new airtime transfer service by social neighbors predicts a higher adoption probability. We then seek to narrow down the possible sources of these network effects by distinguishing between network externalities and social learning. Within social learning, we also seek to differentiate between learning about existence of the new product from learning about its quality or usefulness. We find robust evidence suggestive of social learning both for the existence and the quality of the product. In contrast, we find that network effects turn negative after first adoption, suggesting that airtime transfers are strategic substitutes among network neighbors.