It doesn't matter what you build, when it is at scale the sunk costs can be enourmus to realise half way thru it isn't going to meet your needs, or take 9 years longer than expected.
As Fred said about software, but it applies to everything, nothing is harder than deciding precisely what to build.
(First identify the problem, then work out your requirements to solve that problem, then go from there - most people skip at least the first step if not most of the second and then, there's your problems, esp with software).