It was to partition the app into at least the primary segments that did different functions and had different behaviors, system resource load profiles, etc., and then mandate that those all run on separate machines (this was before everything was cloud-based), and that the relationship to entities was not 1:1. Defined clean interfaces up front. The primary goal was to be able to scale very rapidly for any big customer and/or upon discovering a scalability issue by temporarily throwing hardware at the problem, and secondarily to allow teams to rev different portions of the system on different schedules.
Both worked great, and it wasn't long before we were taking major customers away from competitors because they had scalability issues and we didn't. I'm in R&D and manufacturing now, but see no reason that this approach of fundamental modularity should not be relevant. Heck, I find high modularity to be a good approach relevant to designing industrial processes, or just setting up a shop... just keep everybody's fingers in their own pies, make the organization implement the best architecture, not make your architecture implement your org chart.
(see point about about the not understanding what is being built and constantly changing the scope)
That would have allowed modules/components to operate independently ,reved on different schedules, and more, but evidently too much thought at the outset ,so they just didn't even bring it up ,or passed on the opportunity if it was brought up.
They think they can't spend the resources to get it right, but they'll magically have more resources to fix the problems.
Now everyone pays.