Planning-stage energy models are almost always wrong. Not slightly wrong — routinely 40-60% off actual energy consumption once a building is occupied. The industry knows this, accepts it, and keeps producing them anyway because regulation requires a number, not an accurate number.
The gap exists because early-stage models use assumed occupancy patterns, generic equipment loads, and idealised control strategies that bear no resemblance to how buildings actually operate. A model assuming 8-hour occupancy in an office that runs 18 hours will always underpredict consumption dramatically. The firms producing more accurate predictions are doing two things differently: calibrating models against metered data from comparable completed buildings rather than CIBSE defaults, and running probabilistic sensitivity analysis across occupancy and equipment load ranges instead of single-point estimates.