As usual, Tuc is spot on. There have been a number of initiatives in the past to try to bring some realism into the initial budget estimates at the outset, but these have all been scuppered by the disinterest (with some notable exceptions , to be fair) of the guys in the RP branches on short term postings.
Just to quote one example from my experience, my section was once asked to conduct an IA on a project to fit NBC sensors to all current and future platforms across all three services. These sensors would take ambient readings and automatically transmit them via datalink to groundstations and other platforms. The integration costs alone were far in excess of the budget figure, and when we mentioned EMCON, we were told it was not our area of concern. The project was eventually binned, to reappear as a series of much smaller projects.