Buildings are amongst the largest consumers of electrical energy in developed countries. Building efficiency can be improved by adapting building systems to a change in the environment or user context. Appropriate action, however, can only be taken if the building control system has access to reliable real-time data. Sensors providing this data need to be ubiquitous, accurate, have low maintenance cost, and should not violate privacy of building occupants. We conducted a 3 year study in a mid-size office space with 15 offices and 25 people. Specifically, we focused on sensing modalities that can help improve energy efficiency of buildings. We have deployed 25 indoor climate sensor nodes and 41 wireless power meters, submetered 12 electric loads in circuit breaker boxes, logged data from our building control system and tracked activity on 40 desktop computers. We summarize our experiences with the cost, data yields, and user privacy concerns of the different sensing modalities and evaluate their accuracy using ground-truth experiments.