Smart buildings are a multidisciplinary topic that has attracted interest in recent years. Research often focuses on smart homes, with little attention paid to other building categories. However, these contribute almost as much to the building sector’s emissions as residential buildings. An essential step towards more efficient electricity supply and better grid stability is demand-responsive electricity generation. In the smart grid, however, this requires a forecast of electricity demand and thus a plan of future electricity consumption. Automated planning is a field of computer science that deals with the generation of plans in modelled environments. Uncertainty about the future demands for the consideration of associated risk and the quality of generated plans. In this context, the risk attitude of decision-makers is another decisive factor defining the steps chosen to operate a building. The solution of a planning problem by means of search algorithms as well as the approaches of optimisation and Satisficing are discussed. We present a hierarchical planning domain for the purpose of estimating the electricity demand of a smart non-residential building. Considering all parties involved makes it a multi-objective planning problem. This is why there are no unique optimal but multiple non-dominated solutions. A suitable trade-off between comfort and energy consumption can be found by choosing a suggested plan. The perceived satisfaction that is expected from a plan also depends on the willingness to take risk. Considering all these factors offers the possibility to quantify the expected impact on occupant satisfaction which is associated with a certain energy demand. An implementation of this domain and its application in illustrative scenarios is presented in the following. Proving a good abstraction level of the domain, as well as an effective generation of non-dominated plans. For application in the real world, however, the implementation faces limits of space complexity. The search for non-dominant plans could be made more efficient through analysis of the domain and the search algorithm. Different search algorithms and heuristics are discussed, as well as the dominance of plans in the case of multiple objectives. A comparison with sequential decision processes further reveals a gap in the seamless integration of automated planning and acting. The need for hand-crafting domains yields another topic for future research, namely automatic domain generation.