Back in 2000, I was hired at Juniper Networks to take them from FrameMaker to XML and single-sourcing. They'd intended to do it from the start but and they'd just reached the point, from a corporate capability perspective, to be able to make the change. No longer a rapidly growing startup, they could now devote resources to true single-sourced documentation.
Resources that turned out to be me. Two years later, I was presenting on what we'd done at ACM SIGDOC 2002. And, as it turns out, my presentation happened to be right before Michael Priestly's.
By 2002, he and IBM had started really promoting DITA to the world. I remember him asking me why we didn't choose DITA at Juniper. Truth was, we had considered it and had used much of their thinking in what we designed. The reason we hadn't selected DITA was that it hadn't been ready when we'd started our project two years earlier. I'll always remember that decision and the way things played out. If we'd been later, our decision would have been much different. At that time, though, we didn't have the choices we all have today.
We did get one thing right: We selected Arbortext, a toolset that was robust and complete enough that 1 person could support a 45-member techpubs team for 4+ years. And while I would still choose Arbortext today, I absolutely would not create a custom schema (and resist DITA specialization as well).
It's a sentiment that's shared by nearly everyone who started their XML publishing projects before ~2004. In fact, in 2010, 5 years after DITA was released and 10 years after I started that first XML publishing project at Juniper, I did exactly that. I surveyed long-term, early adopters of XML for publishing to find out what they had to say. Had it been worth it? Were they seeing the returns they expected? Did everyone succeed? Who (or what) failed and why?
In the end, I talked to dozens of people. What did they say? "Do it. Just don't do it the way we did. It's worth it." Every single one of them.