While there is ample evidence that social and communication networks play a key role during the spread of new ideas, products, or services, network effects are expected to have diminished influence in the stationary state, when all users are aware of the innovation, and its usage pattern is determined mainly by its utility to the user. Here we study four mobile phone-based services available to over six million subscribers, allowing us to simultaneously monitor the communication network between individuals and the time-resolved service usage patterns. We find that usage highly correlates with the structure of the communication network, and demonstrate the coexistence on the same social network of two distinct usage classes, network effects being responsible for the quantifiable differences between them. To test the predictive power of our theory, we demonstrate that traditional marketing techniques are ineffective in permanently boosting service adoption, and propose a hub-based incentive mechanism that has the potential to enhance usage for one of the two service classes.