With the decreasing cost of data collection and storage and advances in analytics, companies have begun prioritizing deep analysis of data from their buildings and energy systems. Many of these systems already contain large sensor deployments monitoring their health and operation. However, a reliable analysis pipeline can be difficult to set up initially. The data generated is often dirty and it can take significant resources - days to months of manual inspection - to find traces or segments that can be used for analysis and modeling. In this poster, we present some initial attempts to address this problem by identifying and ranking the most useful traces, with respect to application-level relevance.