Bottom line first: set the data freeze point for the moment the last outbound wave of cutover night is fully shipped, right before the first inbound receipt of the next morning. At that instant the legacy system goes read-only and the new WMS opens with exactly one opening stock table. That point is not a guess. It comes out of a drain-down checklist. I have run 30-plus go-lives, and every freeze failure I have seen was the same story: the work still moving on the floor was never drained before the freeze.

Fall 2023, an auto-parts warehouse outside Suzhou: 12,000 SKUs, 800-plus outbound orders a day. The project team froze at 8 p.m. on a Friday. At 8 p.m. sharp, three picking waves were still half-picked, two trucks were loaded but not dispatched, and more than 40 pallets sat in receiving, never put away. The freeze order landed, the legacy system went read-only, and those three waves had no status in the new system. Drivers waited at the gate for four hours while the team hand-wrote dispatch notes. Day one was chaos, the count was off by more than 600 lines, and it took a second week to reconcile. I have seen this exact pit three times. Three times, same cause: the team treated the freeze as a timestamp instead of a checklist.

Where to Put the Data Freeze: Four Stock Buckets, Not One Total

Most teams set the data freeze point from the single inventory total in the legacy system. That is the mistake. At the freeze instant, stock sits in four completely different states, and mixing them into one number is how you build a mess:

  1. Stock on a bin, available to pick. Clean, usable inventory.
  2. Stock already allocated and picked, sitting in a staging lane. The legacy system still shows it as "available," but the physical goods are no longer in the bin.
  3. Stock in receiving, not yet put away. It may not be in the system at all, or it may be received with no bin.
  4. Stock on quality hold, damaged, or awaiting return. Unsellable, but still sitting in the total.

The cutover rule is simple: split the four buckets before the freeze, one table each. Available stock becomes opening balances in the new WMS. Picked-but-not-shipped orders either ship before the freeze or migrate whole, line status intact, nothing dropped. Receiving stock either gets put away before the freeze or stays sealed on the spot, tagged "receive after freeze," and comes in under the new process after go-live. Hold stock goes into a dedicated freeze zone with quantities locked until someone deals with it after launch.

I have seen this pit three times. The third was spring 2024, a consumer-goods warehouse in East China. The night before the freeze I had the team count the staging lanes, and they found 11 pallets that the legacy system showed as "in bin" while the goods had already been picked and stacked in the dispatch area. If we had loaded the legacy total straight into the new system, those 11 pallets would have been ghost stock on day one: the system says it is there, the picker walks over and finds nothing. That is what the pre-freeze count is for.

48 Hours Before the Data Freeze: Drain Everything, Item by Item

The standard industry practice is to start the drain-down 48 hours before the freeze: operations and implementation walk one drain-down list together and clear every unfinished job. My list looks like this:

  • Stop releasing new multi-shift picking waves. Any wave that cannot be fully picked, packed, and staged before the freeze timestamp does not go out.
  • Unfinished waves either get completed the same day or get cancelled in full, orders returned to pending. A half-picked wave never crosses the freeze.
  • Clear receiving completely: goods either get put away the same day or the whole truck stays sealed in place until after go-live. On freeze night, receiving must be empty.
  • Count variances close the same day. Whatever the pre-freeze full or sample count finds, resolve it that night. Anything unresolved goes on an exception list with a named owner and a signature.
  • Freeze master data two weeks early: SKU list, pack hierarchies, bin structure. Nothing changes during the freeze window. Whatever you want to change waits until after go-live.

That last one deserves its own paragraph. Many projects lump the master-data freeze together with the transaction freeze, figuring one freeze covers both. It does not. Master data must freeze much earlier than transactions, generally two weeks out: if SKUs, unit-of-measure conversions, or bin topology change mid-migration, the slotting logic and pack conversions in the new system break with them. In 2024 I worked a project where the customer added more than 200 new SKUs three days before the freeze. They existed in neither the legacy export nor the new system, so goods arrived at the dock and could not be received. Two people spent a whole night keying those SKUs in by hand.

Cutover Night: The Go-Live Night Checklist, No Exceptions

On cutover night my checklist fits on one page, and every item on it is a veto:

  1. The legacy system is read-only, the freeze timestamp is recorded, and nobody can change a number anymore.
  2. The opening stock table is exported: legacy closing balance, new opening balance, and the finance sub-ledger agree across SKU, batch, bin, and status. One line off and we do not go.
  3. The open-order list is reconciled: every unshipped order carries full line status, no half-picked and no half-packed lines.
  4. The exception list is signed: every variance has an owner and a plan, and nothing is parked as "we will fix it after go-live."
  5. The rollback plan is written down: who can call a stop, what happens to orders already entered in the new system, and how the legacy system gets write access back.

Item 5 is the one teams skip. A rollback plan is not written for the steering committee. It is written for the person on duty at 3 a.m. When something really breaks, nobody has time for a meeting. Everyone follows the few lines on that page.

Run the math on what a wrong freeze point costs. Take a warehouse shipping 1,000 orders a day. If the drain-down is sloppy and day-one counts come out 2 percent off, that is 160 SKUs wrong out of 8,000 (hypothetical example, illustrative scenario). At 20 minutes to investigate each variance, that is 53 labor hours, roughly seven people for a full day. That is before mis-ship compensation, stockouts, and customer complaints. A proper drain-down costs a few people one night of overtime. Put the two numbers side by side and the answer is obvious.

Here is the counterintuitive part: an earlier data freeze is not a safer data freeze. Plenty of teams think freezing a week early is the careful move, because there is more time. It is the opposite. The earlier you freeze, the longer the frozen period, and the more the business suffers inside it: new products cannot be shelved, urgent orders cannot ship, receiving piles up at the door. Every extra frozen day is another day of manual workarounds, and manual workarounds are where errors breed. The safe move is a clean drain-down that lets the freeze point sit as late as possible, with the shortest window you can manage. A 48-hour drain-down buys you a clean cutover night. Freezing early is not rigor. It is a lack of confidence in the drain-down.

Here is a timeline you can lift straight into your plan: freeze master data two weeks before cutover, start the drain-down 48 hours out, run the full reconciliation count the night before, freeze at 10 p.m. on cutover night once the last outbound wave is shipped, finish the opening-balance reconciliation and the go/no-go call by 4 a.m., and run the first inbound receipt under the new process at 6 a.m. Then keep both systems reconciled in parallel for 30 days after go-live, and only switch the legacy system off once the numbers agree. Thirty days of dual reconciliation looks like overhead. It is the cheapest insurance I know.

If you are building a cutover plan right now, do not set the freeze point first. Pull the four stock-bucket tables, and look at how much work is really sitting in staging and receiving. Once those tables are on the table, the freeze point sets itself.

By Lin Xiao, WMS Implementation Consultant