Conclusion first: before a new WMS goes live, historical inventory must be reconciled in the old system, migrated into the new one, and then locked by a physical count. Skip any of those steps and pickers will be sent to empty locations on day one, while customer service and finance come looking for you at the same time. I have seen this failure three times. The excuses were different each time. The outcome was identical.
Why does historical inventory fail so often? Because it is not one static table. It is three things bolted together: book stock, in-transit stock, and exception stock. Book stock is the number in the system. In-transit stock is shipped but not received, received but not put away, or sitting on a transfer in between warehouses. Exception stock is damaged, blocked, quarantined, or consigned goods. Migrate only the book number and you have thrown the other two away. On day one the system says 100 units, the location holds 80, and the missing 20 are floating somewhere in transit with nobody able to explain them.
Step 1: Freeze the moment, reconcile three ledgers of historical inventory
In June 2023 I took on a go-live for an auto-parts client in East China. The old system showed 42,000 units, the finance ledger showed 41,000, and a spot count showed 39,000. Three ledgers, a 3,000-unit gap. The client wanted to cut over that weekend. I stopped it. Four days of reconciliation found 1,800 units on transfer orders that had shipped without being deducted, 900 units damaged but never written off, and 300 units posted to the wrong SKU.
The procedure is exact, and it is meant to be followed literally:
- Set a freeze time, for example Friday 18:00. After that point the old system records new receipts and shipments only — no edits to historical documents.
- Pull three reports: the old system's stock table, the finance inventory value table, and the most recent count sheet. Reconcile quantity and value SKU by SKU.
- Sort every difference into one of three buckets: missing documents get completed, physical differences go through write-off or write-up approval, and wrong-SKU postings get an inventory adjustment document. Every difference needs a document number. A verbal explanation does not count.
- List in-transit documents separately: purchase orders in transit, sales orders in transit, transfers in transit, and unclosed production issues and returns, each with an expected arrival or closing date.
My bar for moving on is strict: zero quantity difference, and value difference within 0.1%. Clients always push back on that bar at first. After reconciliation they understand: the arguing it prevents after go-live is worth far more than those four days.
Step 2: Migrate historical inventory in three batches, never all at once
The worst way to migrate is everything in one load. My batching method:
- Batch one, master data: SKUs, locations, suppliers, customers. Run a validation script first and clear three error types — duplicate codes, empty required fields, and nonexistent locations — before going further.
- Batch two, normal stock: items in normal status, with a clear location, sold or moved in the last 90 days. This is usually around 85% of total volume.
- Batch three, exception stock: blocked, damaged, quarantined, consigned, and near-expiry goods, each with a reason and an owner. Exception stock must never be mixed into the normal batch. Mixing it in plants a bomb for go-live week.
In March 2024, a cosmetics client had more than 600 near-expiry units in batch three. The client wanted them loaded as normal stock to "just get live." I insisted on listing them separately. In week two after go-live, 120 of those units would otherwise have been picked and shipped as normal goods — and in the event, a mixed load in a parallel test run did exactly that, triggering complaints, returns, and compensation that cost far more than the half day separate listing took.
After loading, run three checks: total stock in the new system equals the old system's frozen total; spot-check 10% of high-value SKUs location by location and batch by batch; and run 50 simulated pick orders to confirm the recommended locations actually hold stock. If any check fails, the whole load goes back. "Go live now, fix it slowly later" is not an option.
Step 3: The night-before checklist, then lock the numbers
The night-before checklist is the final gate on every project I run. It goes in the project channel, item by item, ticked off in public:
- All documents created after the freeze time have been loaded, and the difference report is at zero.
- The exception stock list is signed off, and blocked goods are blocked in the new system too.
- Picking, counting, and transfer flows have each run five real test orders.
- The rollback plan is written down: if picking accuracy falls below 98% within four hours of go-live, who decides to fall back to the old system and manual orders, and whose phone number gets called.
Here is the counterintuitive part: more history is not better. Teams want to migrate five years of transaction detail, as if completeness were safety. It is not. It slows the system and buries the queries that matter. My rule: migrate all opening balances, migrate the last 12 months of movements as summaries, and leave anything older as an archive in the old system. A go-live is an engine swap, not a house move where you carry every box out of the attic.
One more failure point deserves its own warning: quantities reconcile but costs do not. Opening quantities can match while the value is far off, and finance only discovers it at month-end close. Quantities and costs must be validated in the same batch, and a cost difference beyond 0.1% blocks go-live just like a quantity difference does.
Get opening stock wrong and day one falls apart. That is not a scare line; it is what three rescue jobs taught me. Freeze and reconcile, migrate in batches, and run the night-before checklist — and the historical inventory gate holds.
By Lin Xiao, WMS Implementation Consultant







