The decisions that decide a cutover are operational, and most of them are made before the weekend starts. By Friday night the load either runs or it does not, but the load is not the test. The test is whether the 6am shift can receive a truck, put stock away, pick a wave and get it out the door without a person from the project standing at every bay. Everything in this note is aimed at that one Monday morning.
Cutover is an operational event.
Treat cutover as an IT milestone and you will measure the wrong thing. The go-live status call fills up with green ticks: extracts completed, tables loaded, reconciliation reports generated, no errors in the log. All true, and none of it tells you whether the floor can work. A warehouse that has loaded cleanly and cannot despatch is still a warehouse that cannot despatch.
So set the acceptance test in floor terms and rehearse the people who will apply it. Can a storeperson scan an inbound handling unit and see the right put-away instruction? Can a picker walk a wave without a queue forming at the exceptions desk? Can a truck leave on time with the right paperwork? Those are the questions the weekend exists to answer. If the answer is not yes by Sunday evening, a clean load has bought you nothing.
Stock accuracy at freeze.
You cannot go live on inventory you do not trust. The count strategy at freeze is the single decision most likely to sink Monday, because everything downstream (put-away, replenishment, pick confirmation) assumes the opening balances are real. Get this wrong and the floor discovers it one empty bin at a time, in front of a customer order, on day one.
Pick the count strategy honestly
There are three routes and each carries a cost. A full physical count is the most trustworthy and the most expensive in floor hours and downtime. Cycle-count sampling is faster but only as good as the sample design and the accuracy of the bins you did not touch. A pure system-to-system reconciliation is the cheapest and trusts the legacy data completely, which is exactly the data you are leaving because you stopped trusting it. Choose with your eyes open, and match the route to how much the legacy stock record has actually earned.
Decide the variance rule before you find one
The moment that breaks teams is a variance found at 2am on Sunday. A bin says one hundred, the count says sixty-two, and nobody agreed in advance who adjudicates or what the tolerance is. Now you are debating accounting policy at the worst possible hour. Decide it beforehand: what variance you accept, what you recount, who signs a write-off, and what threshold escalates to the go / no-go call. Write it on the runbook so the 2am decision is a lookup, not an argument.
Open transactions at freeze.
Real warehouses are never at rest, so freeze always lands in the middle of something. Inbound deliveries part-received. Orders part-picked with handling units already staged. Stock in transit between manufacturing and the warehouse, on a truck, in nobody's system cleanly. Every one of these in-flight items needs an explicit decision, because the load will not make it for you.
There are only three answers per item and you choose one for each category: close it before freeze so it never crosses, migrate it mid-flight with its current state intact, or park it and re-enter it after go-live from a controlled list. Closing is cleanest and costs you a rushed Friday. Migrating mid-flight is the hardest to get right and the easiest to leave half-done. Re-entry is safe but only if someone owns the list and works it on Monday. What you cannot do is discover a part-picked order on the floor with no record of the decision.
The runbook is judged on Monday morning, not on Sunday night. A clean load at midnight means nothing if the first shift is improvising by nine.
The go / no-go call.
Define the criteria and the decision-maker in advance, in writing, before anyone is tired. Not “we will see how it looks” on the night. A short list of hard gates: stock accuracy above the agreed threshold, open transactions all resolved or listed, critical interfaces confirmed live, first-shift staffing in place. One named person makes the call against that list, and everyone agrees the list beforehand.
A no-go recommended on the evidence at 9pm Saturday is a professional act, not a failure. The wrong move is to press on because the comms already went out, the town hall already promised a date, and nobody wants to be the one who blinked. Sunk communications are not a reason to open a warehouse that is not ready. Give the decision-maker explicit cover to call no-go and it becomes a decision you can actually make.
The rollback trigger.
Separate from go / no-go, and just as easy to skip. Rollback is the condition that sends you back to the legacy process after you have already gone live. Define it as a specific, pre-agreed condition, not a feeling: for example, despatch throughput below a set rate for a set number of hours with no clear fix in sight. Name who can pull it and how long they have to decide, because a rollback window closes fast once stock has moved under the new system.
If you have not defined the trigger, you will not use it. Instead you will limp: the floor half-works, the exceptions desk grows a queue, and everyone tells themselves it will settle by Wednesday. Sometimes it does. When it does not, the team that wrote the trigger down had a clean exit and the team that did not had a slow bleed. Decide which team you want to be while the room is calm.
First-shift staffing and floor-walkers.
Put senior EWM presence at the bays for the first shifts, not in a control room reading dashboards. The questions that stall a floor on day one are small and local: why did this handling unit not propose a bin, why is this wave not releasing, why does the scanner say the pick is short. They get answered fastest by someone standing next to the storeperson, at the rack, in the aisle.
Rehearse the escalation path so a floor-walker who is stuck knows exactly who to raise it to and how, without leaving the bay. A named EWM lead, a named integration contact, a single channel, and a rule about what gets fixed on the floor versus what gets logged for later. The goal is that a problem at bay six never becomes a queue of eleven people waiting on one project manager's phone.
Rehearse the runbook.
A dry run finds the missing step while it is still cheap. Walk the full sequence on a non-live weekend or in a staging environment: the freeze, the count, the load, the reconciliation, the go / no-go gate, the first-shift kick-off. Time each step. The rehearsal is where you learn that step fourteen depends on an interface nobody scheduled, or that the count reconciliation takes four hours and you budgeted one.
Every gap you find in rehearsal is a gap you did not find at 3am with a truck at the dock. That is the whole trade. The runbook is not a document you write to satisfy governance, it is the thing the weekend actually runs on, and the only way to know it holds is to run it before it counts.