Let me be blunt: a WMS migration cutover without downtime is two problems, not one. One is technical — dual runs, data, interfaces. The other is human — who can call a halt at 2 a.m. I have watched too many teams spend everything on selection and arrive at cutover night with nobody willing to decide. My verdict up front: no downtime is not the goal. It is the floor. Miss the floor and the finest features mean nothing.
WMS migration cutover: my verdict up front — run both systems or don't sleep
A dual run means the old and new systems process the same orders and account for the same stock twice, then you reconcile line by line. Do not dual-run everything at once; that burns people out for nothing. Start with one order type — small-item picking, say — and expand to bulky goods and returns once it runs clean. Reconcile three things only: on-hand inventory, wave counts, and carrier labels. When all three match, the dual run counts. Run it for two to three weeks, until variance falls below the line agreed in advance, and only then set a cutover date.
Here is the trade-off most teams get wrong: a longer dual run feels safer, but past a month it turns against you. The team works days and reconciles nights, variances pile up, and the cutover gets harder, not easier. Two to three weeks is my rule — long enough to prove the system, short enough to stay sharp. The price is double labor: entry, checking, and exceptions all need two crews. Put that cost in the budget; do not discover the headcount gap on cutover night. Three cutover patterns exist, and the choice starts with your own building:
| Pattern | What happens | Risk | Fits whom | | Big bang | Everything switches in one night | Highest — one failure stops the building | Small sites, low volume | | Dual run | Old and new run side by side for two to three weeks | Medium — labor doubles | Most warehouses | | Phased by process | Inbound first, outbound later, in stages | Lower — but the timeline stretches | High volume, cannot stop |
The freeze window is where cutovers die
In June 2023, an apparel warehouse in California cut over two weeks before peak season. The project team set a 4-hour freeze window, proud of how short it was. At 1 a.m. on cutover night, count variances surfaced in more than 300 locations, the window blew past its limit, and at 7 a.m. the first outbound wave sat stuck in the new system while carrier trucks waited at the dock for 3 hours. I still remember that lesson: a freeze window is not a stunt. It is a safety net, and it must be sized for the worst case, not the smoothest one. I have watched too many a WMS migration cutover die in the freeze window, always the same way — a window set too short, one variance, and the whole plan collapses.
Before the freeze, reach a quiet point: waves picked and confirmed, put-away cleared, replenishment either finished or deleted. What remains is stock sitting in bins and nothing in motion. A half-finished task cannot be carried across a freeze; forcing it plants a bomb for cutover night. The quiet point is a floor condition, not a system status — someone walks the aisles with a list and signs it off. Taper outbound volume deliberately, weeks ahead, with sales and customer service in the room — not merely told after the fact.
WMS migration cutover night, run by the minute
Cutover night leaves no room for improvisation. The runbook is timed to the minute, with an owner and a rollback trigger on every line. The table below is an illustrative example, timed for a mid-sized warehouse:
| Time | Action | Owner | | 22:00 | Stop new waves, close out in-flight tasks | Warehouse manager | | 23:00 | Freeze data, load and reconcile | System lead | | 00:00 | Clear variances, confirm locations one by one | Count lead | | 02:00 | Test orders: pick, pack, and label end to end | Test lead | | 04:00 | First outbound wave, command center watching volume | Duty manager | | 06:00 | Review, decide whether the day proceeds | Project commander |
In October 2024, an auto-parts warehouse in Texas cut over on exactly this schedule. At 2 a.m., test orders showed the label-printer driver did not recognize the new system. Within 15 minutes the command center decided: outbound labeling went back to the old system, the new system kept picking, and trucks left on time at 6 a.m. Not one order was lost; the driver problem was fixed in daylight. Skill on cutover night is not preventing problems. It is deciding within 15 minutes when one appears. Write those decisions into the runbook in advance: which failures send you back to the old system, which ones you push through — if it is not written down, nobody dares decide at 2 a.m.
Rollback: who can call it matters more than how
The most valuable part of a rollback plan is not the technical steps. It is three lines written down in plain language: who has the authority to call a halt, what triggers it, and how many minutes it takes to return to the old system. Triggers must be numbers: three failed test orders in a row, or first-wave outbound volume below half the normal rate for an hour — then roll back, no meeting required. Numbers matter because at 2 a.m. nobody's judgment is sharp; numbers decide for you.
A WMS migration cutover stays reversible only while the way back stays open: keep the old system writable for at least one full business cycle, and do not stop dual-writing in a hurry — once dual-write stops, rollback is an empty promise. Keep the old logins, printers, and handhelds in place; recovering them is a separate scheduled step, two weeks after a successful cutover, never sooner. Rehearse the full runbook at least once, and deliberately break one step in rehearsal to see how fast the team gets back to the old system. A rollback plan that has never been rehearsed will not save you on the night it counts.
At 2 a.m., the third test order just failed. Do you dare call the halt? If you cannot answer that, do not set a cutover date.
By Lao Mi, Supply Chain Veteran (20 yrs)







