Some thoughts:
* Most companies are not software companies, and most c-level managers are not tech leaders. They treat IT as an annoying expense, like taxes. The idea of hiring software engineers to gather requirements and run a lean process to build efficient software from the ground up on a modern web stack is as foreign to them as setting up a chemical processing pipeline and logistics management network is to me.
* Rolling your own solution takes time, and more importantly, energy and involvement of the whole company. I would argue that it might take less time than a big SAP implementation, but with the latter you can just hire a big consulting firm (there's that cost center again...) and they'll deal with everything for you so your people can just work.
* This is a bit of a chicken and egg thing, but: A lot of SAP's customers aren't startups. If you're hired in the be the CEO of an oil company running major pipelines, you've already got SAP running everywhere. And you do your first big move and acquire a small competitor, and THEY have SAP running too. So why are you going to embark on a huge software development project when a) you're not a software company, b) you're not a startup leader, etc. etc. etc.
Although to be honest, the success (?) the US Government has had doing basically what you propose via the USDC / 18F does give me some hope. But alas, much of the world is too far gone already (sigh).
"We held firm to the NO CUSTOMIZATION rule, we re-engineered our processes to fit SAP. Other than a few hiccups when integrating all aspects of a company that has 100 sub companies, and is under federal, state and local regulation the project basically went off without a hitch." [0]
At these kinds of numbers, I've got to ask a genuine question here: what's the benefit of using SAP as opposed to rolling your own solution?
From reading some of the other comments here, it sounds like getting your business to integrate with SAP involves trying to fit a lot of square pegs in round holes and doesn't always succeed in the end anyways.
At these levels of stratospheric costs, it seems like it would often be cheaper and less risky to simply use your standard software stack of choice and build a custom web app over a database that mirrors your existing business processes perfectly, and then transform/integrate those processes over time as needed for increased efficiency.