Early in Tellus I made a call that looked wrong on paper. We took on design work at a loss, sometimes handing it out for free, because I had no way to predict when a project would actually get approved. Permitting was the black box that wrecked every schedule I tried to build, and I decided the only way to fix that was to own the front end of it myself.
Once we controlled the design, we controlled the pace of approval, and once we controlled that, we started selling plan expediting as its own service line. The thing I gave away built the thing I later charged for.
I think about this every time somebody asks whether it makes sense to eat margin on one part of a job. My answer is yes, if that part is the bottleneck nobody else can control. Owning the choke point is worth more than the margin you lose absorbing it, because the choke point is what you get to monetize once you understand it better than anyone else in the market.