I've watched health IT implementations fail for over fifteen years. The technology gets better every cycle. The failure rate doesn't. That should tell you something about where the problem actually lives.
The Pattern
- Most health IT failures aren't technology failures. They're failures of change management, workflow understanding, and organisational readiness.
- The three most common causes: underestimating clinical workflow complexity, treating training as an event rather than a process, and missing the gap between vendor capability and operational reality.
- AI is now entering the same cycle. Early pilots look promising. The organisational challenges that kill implementations haven't gone anywhere.
~70%
of health IT implementations fail to deliver expected benefits
Source: Standish Group, CHAOS Report, 2020
Here are the three ways your implementation will fail. I've seen each one multiple times.
You'll skip the workflow analysis. Not deliberately. You'll do a requirements workshop. Stakeholders will describe their ideal workflow. The vendor will map it to their platform. Everyone will agree it looks great. But nobody will have spent a week in the practice watching how things actually work. The gap between the documented workflow and the real one is where implementations die.
You'll underinvest in training. Two half-day sessions before go-live. A user manual nobody reads. A help desk that takes 48 hours to respond. Within a month, staff will have developed workarounds that bypass half the system's features. Within six months, leadership will wonder why the promised efficiency gains haven't materialised.
You'll launch everything at once. The vendor has a project timeline. The board has a deadline. So you'll go live with the full system across all sites simultaneously. And when something breaks, which it will, you won't have the support capacity to fix it everywhere at once. Phased rollouts are slower. They're also the only approach I've consistently seen work.
3x
higher adoption rates in phased health IT rollouts vs. big-bang implementations
Source: Journal of the American Medical Informatics Association, 2019
The organisations that succeed do something unfashionable. They go slower. They spend more time in practices before they configure anything. They budget 40% of the project for change management, not 10%. And they accept that the real go-live isn't the day the system turns on. It's six months later, when staff have stopped using their workarounds.
AI won't change this dynamic. If anything, it will make it worse. AI tools promise faster results with less effort. That makes them even more tempting to deploy without the organisational groundwork. The pilots will look great. The production implementations will hit the same walls they always have.
The technology isn't the problem. It never was.