I think there was a YC funded iPad EMR startup that tried to be cool / hip / provider first until they got smacked in the face by reality.
This.
Once I worked making healthcare software that would basically save costs by avoiding complications. We were bidding into a large hospital network. After the long sales cycle, we checked the most boxes, clinicians liked the feel of ours the best, they said ours was the most intuitive, helped them do their job the fastest, etc.
The clinicians weren't the ones writing the check though.
Our competitor approached the main insurance provider in the area and convinced them they could save x% in additional claims if the hospital network would adopt their software. Competitor's deal was to split the bill between the insurance company and the hospital network. To the admins writing the checks, they had the choice between two vendors that more or less did the same thing but one came in at half the cost. Clinicians' preferences didn't matter, it was a no brainer for them.
No amount of software engineering would have saved that deal.
In part I think it happens more commonly because there is so much "other peoples money" in healthcare. Everyone is spending from someone elses checkbook and so transactions seem very distant.
Also, compliance doesn't fully explain why competitors are able to convince clinicians to switch systems. Word-of-mouth means that people will know your EMR is a flaming piece of garbage, but (as you noted) companies would much rather cut up-front prices so they can milk an extended contract than improving product quality. All that said, I think there's a bit of a chicken-and-egg thing going on here. Good people don't join the space because the culture sucks and the pay is bad, but those are because those with the talent and drive all self-selected out of it. I get it, health IT is a huge drag and not at all sexy. But just look at how often folks on HN ask about "doing social good" and how many complaints there are about healthcare delivery. Trying to run a "disruptive" VC-backed startup is IMO pretty crazy, but contributing to an OSS project is far less risky and more achievable.
The problem is that the Venn diagram between the people who actually have to use or administer the EMR and the people who decide which EMR to implement is basically two entirely-disconnected circles.
That said, I don't really recall the doctors themselves using an EMR much, either; they'd usually punt that to assistants and/or nurses. Probably different in smaller practices/clinics.