A novel model was developed to test the use of short data sets for testing various alarm thresholds as part of an energy management program. Several years of 15-minute interval data were utilized from five buildings in Jacksonville, Florida. The model aggregated the data by day type and occupancy so that there were four period types used. For all of the buildings’ meters, their daily usage by period type was tested against the threshold to determine if an alarm would be triggered, which would then be assigned a reward and cost based upon the type and duration of response. The risk management value was converted to dollars, in order to normalize the energy and time. It was determined that the 5-month short data set was the most appropriate choice for short data sets. In addition, it was concluded that the thresholds should be set between 0.8 and 1.0 standard deviation above the average of the short window. Several recommendations for further study are also enclosed.