An EHR replacement gets approved as a technology decision and delivered as an operations project. The software installs fine. What takes eighteen months is everything the software forces the organization to decide out loud for the first time.
Most published accounts of these projects skip the middle. They describe the contract signing and then the ribbon-cutting photo from the command center. The part in between is where the money goes, and it looks nothing like a software rollout.
The decision rarely starts with the EHR itself
Very few hospitals wake up wanting a new record system. The trigger is usually something else. A merger leaves two health systems running different platforms and reconciling patient identities by hand. A vendor announces end of support for the version a hospital is three upgrades behind on. The CFO looks at a denials report and realizes the charge capture logic was never built correctly in the first place. Sometimes an ambulatory group threatens to leave because they cannot get their own data out of the system in a usable format.
Selection then runs six to twelve months. Scripted demos, site visits, reference calls, a scoring matrix that ends up weighted toward whatever the loudest specialty cares about. The scoring matrix matters less than people expect. Two enterprise systems will both do 90 percent of what a hospital needs. The differences that actually hurt show up later, during build, when a specific workflow has no clean equivalent, and somebody has to decide who absorbs the extra clicks.
Most of your patient history does not come with you
This surprises clinical leadership more than anything else on the project.
Discrete data migrates: demographics, insurance, problem lists, allergies, medications, immunizations, and usually a limited window of lab and vital sign results. Progress notes, scanned consent forms, faxed referrals, and older results typically do not migrate as structured content. They either convert into flat PDFs attached to the chart or they stay behind entirely.
The old system does not get turned off. It moves into a read-only state and stays there for years, because state retention rules commonly run seven to ten years for adults and longer for pediatric records. That is a licensing, hosting, and access-management cost that many budgets forget to include.
Then comes abstraction, which is the least glamorous work in the entire project. For every patient with an appointment in the first several weeks after go-live, someone opens the legacy chart and manually re-enters what the clinician will need. Nurses and medical assistants do this on evenings and weekends for a rolling window of scheduled visits. Teams spend months arguing about migration scope and then hand-carry the important charts across anyway.
Build is where undocumented workflows get argued about
Build is thousands of small decisions. Order sets, note templates, charge triggers, result routing rules, security classes that control who can see behavioral health notes, printer mapping, downtime paper forms.
Each one exposes a process nobody had written down. The pre-op team has a way of handling same-day add-ons that exists entirely in one charge nurse’s head. The infusion center bills differently than the policy says it does. A hospital going in with 600 order sets from its legacy system will usually find that a third are near-duplicates created by individual physicians over the years, and the request to consolidate them is where governance meetings get tense.
The unpopular position is usually the right one. Standardize where the clinical evidence does not justify variation, and spend the customization budget on the workflows that genuinely differ.
The interface inventory is always bigger than the list you were given
IT hands over a list of interfaces. Call it sixty. Then somebody walks the building and asks each department what they actually use, and the real number lands somewhere between 150 and 300 connections, plus a long tail of small applications, some running on a desktop tucked under a counter in the sleep lab.
A team scoping Epic EHR integration will map the obvious feeds early: ADT, lab, pharmacy, radiology orders and results, transcription, the clearinghouse. Those get attention because they are visible and funded. The ones that break go-live are almost always the small ones. A tumor registry. A dictation tool one surgeon depends on. A state immunization registry submission that fails quietly for three weeks before anyone notices.
Departmental systems complicate this further, because most of them are staying. Diagnostic medical imaging usually keeps its existing PACS and VNA, so the project is not replacing imaging; it is rewiring orders, worklists, and result delivery around a new record system. The same applies to cardiology, anesthesia, oncology treatment planning, and the lab. Add connected devices, and the count rises again: infusion pumps, physiologic monitors, ventilators, analyzers.
Testing is where the schedule actually slips
Build finishes roughly on time more often than people assume. Testing is what reveals the truth.
Unit testing confirms a single piece works. Integrated testing confirms that registering a patient produces an account, that an order reaches the right system, that a result posts to the correct chart, and that the encounter generates a claim that survives the clearinghouse. That last chain is the one most commonly shortchanged, and it is the one with direct cash consequences.
Then come dress rehearsals: a full technical cutover practiced end to end, timed, with the rollback plan tested rather than assumed. Projects that skip the second rehearsal usually pay for it during the real cutover weekend.
Training is a staffing problem before it is an education problem
Classroom hours vary by role, from about four hours for a scheduler to sixteen or more for a physician or an inpatient nurse. Most systems require a proficiency assessment before granting access, which means training is not optional and cannot be quietly skipped by a busy service line.
The hard part is coverage. Pulling every nurse on a unit through eight hours of class means backfilling those shifts, and the backfill budget is frequently underestimated. Super users get pulled from their units weeks in advance. At go-live, most organizations aim for roughly one support person per eight to ten users on the floor during peak hours.
Go-live week
Schedules get cut. A typical plan reduces clinic volume by 40 to 50 percent in week one and steps it back up over three to four weeks.
A command center opens, tickets get triaged, and leadership huddles every few hours. Ticket volume usually peaks on day two or three, not day one, because day one is quiet enough for people to be careful and day three is when the backlog and the fatigue meet.
What goes wrong is small and constant. Badge printers. Label printers, especially. A field dropping off an interface message. A physician looking for an order that exists under a different name now. Registration lines growing because a workflow that took four clicks takes nine. None of it is dramatic on its own. Together, it is exhausting, and the exhaustion is the real risk to patient safety that week.
The revenue dip is predictable, so plan for it
Claims slow down. Coders work in an unfamiliar system, charge capture gaps surface, and denials rise while payer edits get corrected. Days in accounts receivable climb, and cash collections drop before they recover.
Boards should hear this number before go-live, not after. Organizations that come through cleanly hold several months of operating cash or an arranged credit line, and they treat the first ninety days of revenue cycle metrics as a project deliverable with a named owner. Recovery typically takes two to four months. Where it takes longer, the cause is almost always build logic that was wrong from the start rather than staff who were slow to learn.
Months three through twelve are the actual project
Go-live is not the finish line, and treating it as one is the most common mistake in the whole effort.
Legacy reports are gone. Every operational dashboard, quality measure, and board report has to be rebuilt against a new data model, and that work often stretches past the first year. Note templates that looked fine in testing need trimming once physicians use them at volume. Alerts fire too often and need tuning before people stop reading them.
Most hospitals disband the project team the month after go-live, right when the organization finally knows enough to ask good questions. Keeping a funded optimization team for twelve months is what separates a replacement that eventually pays off from one that people simply learn to tolerate.
