Fully assessing the robustness of a software application in-house is infeasible, especially considering the huge variety of hardly predictable stimuli, environments, and configurations that applications must handle in the field. For this reason, modern testing and analysis techniques can often process data extracted from the field, such as crash reports and profile data, or can even be executed directly in the field, for instance to diagnose and correct problems. In all these cases, collection, processing, and distribution of field data must be done seamlessly and unobstrusively while users interact with their applications. To limit the intrusiveness of in-the-field monitoring a common approach is to reduce the amount of collected data (e.g., to rare events and to crash dumps), which, however, may severely affect the effectiveness of the techniques that exploit field data. The objective of this Ph.D. thesis is to define solutions for collecting field data in a cost effective way without affecting the quality of the user experience. This result can enable a new range of testing and analysis solutions that extensively exploit field data.