Every go-live has a moment, usually somewhere between 6:30 and 8:00 on the first morning, when the command centre finds out which assumptions were wrong. After thirteen years in hospital systems, ten of them in Epic, we have noticed that the list of things that break is shorter and more predictable than most project plans assume.
It is rarely the build itself. Analysts test their own build hard. What breaks is the space between the build and the people using it, and that space looks much the same from one organization to the next.
The five things that break most often
1. Access
Someone logs in and sees the wrong activities, the wrong department or nothing at all. A resident cannot sign. A charge nurse cannot release orders. A pharmacist sees the right screens but cannot verify. Security is built against a role design that was finalized months earlier, and people have changed jobs, started, or picked up second roles since. Access problems are the loudest tickets of day one because they stop work completely.
2. Content that matches the spec but not the practice
Order sets, preference lists, protocols and documentation templates are built exactly as signed off — and clinicians still cannot find what they need, because the sign-off happened in a meeting room, not at the bedside or the clinic desk. A default that made sense in a design session becomes six extra clicks on every patient. The build is correct; the workflow is wrong.
3. Printers, labels and devices
Nothing erodes confidence faster than a specimen label printing two floors away. Device and printer mapping touches every location, every workstation and every department, which is exactly why it is often tested only in a sample. The sample passes, and the third floor west discovers its mapping at 7:15.
4. Hand-offs between areas
Each department tests its own workflow, and each one passes. The failures sit between them: orders that need releasing when a patient moves from pre-op to the OR to recovery, referrals that leave one clinic and never arrive at the next, results routed to a provider who left last spring. Nobody owns the seam, so nobody tests it end to end.
5. Converted data
Future appointments, active medications, treatment plans and problem lists come across from the legacy system. Most of it is fine. The part that is not — a treatment plan landing on the wrong cycle day, an appointment without its provider — tends to surface when a patient is already in front of a clinician.
Three that are cheap to prevent
None of these needs more budget. They need the right people in the room a few weeks earlier than usual.
Test access with real accounts, not test accounts
Two to three weeks before go-live, have a sample of real end users from every role log in with their own credentials and walk one common task. Test accounts prove the role works; real accounts prove the person was given the right role. This catches most day-one access tickets for the cost of a few hours.
Walk every printer and device, location by location
Give a small team a printed list of every location and a script: print a label, a wristband, a requisition and a report from each workstation type. Check off each one. It is tedious and it is the single best hour-for-hour investment we know of before go-live.
Run the seams, not just the departments
Build dress rehearsals around patient journeys that cross at least three areas — admission to procedure to discharge, referral to clinic visit to follow-up — and run them with the clinicians who do the work. Assign a named owner to each hand-off. The point is not to test the system again; it is to find the places where two correct workflows do not fit together.
What this means for planning
When we plan a go-live, we hold back time for these three activities and protect it, because they are the first things cut when a build runs late. They are also what decides whether the first week feels like a rough start or a crisis. If you are approaching a go-live and want a second set of eyes on readiness, our scoping tool is a fast way to start that conversation.