Here's my verdict up front: when you integrate ERP and WMS, master data goes first. Always. I've run 30+ go-lives, and every project that tried "let's get the order interfaces live and clean up master data later" blew up on go-live night. The one exception wasn't lucky — it just got delayed long enough for someone to finish the master data work properly.
Think of it as roads and traffic. Integration gets your purchase orders and sales orders moving between ERP and WMS — that's the road. Master data is the signage: item codes, location codes, customer records. If ERP's sign says "SKU-48213" and WMS reads "SKU48213," the traffic crashes. Having the interfaces live doesn't mean you're ready to go live. I've watched that exact script play out three times.
Why master-data-last always detonates
Summer 2024. A beauty brand with its own warehouse in Riverside, California — 50,000 square feet, roughly 4,000 orders a day. The client's IT director decided: "Master data cleanup takes too long. Let's stand up the PO and SO interfaces first and backfill master data after go-live."
Three days before go-live, ERP pushed a purchase order to WMS. Error: item not found. ERP's item number was "SKU-48213"; WMS had imported it as "SKU48213" — one missing hyphen. Neither side would budge, because the interface spec just said "pass the item number through." Garbage in, garbage out.
It got worse. ERP held 6,000+ active SKUs, including 400+ zombie items — discontinued, resurrected, discontinued again, still sitting in the price tables. On go-live morning, the first receiving wave hit "item not maintained" errors at the dock. Trucks stacked up. We didn't start operations until 4 PM, with 12 trailers backed up at the doors. The fix: freeze item creation in ERP that night, declare WMS the item master, and sync backward into ERP. The go-live slipped two weeks and burned three extra consulting weeks.
Contrast that with a home-goods 3PL in Ohio the following year. Day one, we locked three coding standards — items, customers, locations — eight weeks before go-live. Any new SKU entering the building had to clear a master data review first. The order interfaces were integrated last, in the final two weeks. On go-live night we had exactly two issues. One was a field-length mismatch — 20 characters in ERP, 30 in WMS — fixed with a truncation rule in half an hour. It was one of the smoothest go-lives I've ever run.
Side by side, the answer is obvious: master data first isn't the cautious option, it's the only option that survives contact with reality. Master data is the shared language of the two systems. If the language isn't unified, faster conversations just produce bigger misunderstandings.
How I run ERP-WMS integrations: the seven-step sequence
Same playbook on every project now. Follow it and you'll be fine:
-
Lock the coding standards before anything else. Eight to ten weeks before go-live, finalize the item coding standard, location coding standard, and customer/vendor coding standard — signed by all three parties. Whoever signs owns any later change. After sign-off, freeze: special cases go through a change request, not a hallway conversation.
-
Give every master data domain one owner. ERP and WMS both need item data, but only one system is the source of truth. My usual split: items belong to ERP (finance and purchasing live there), locations belong to WMS (the warehouse knows its racks), customers belong to ERP, vendors belong to ERP. The owner creates and maintains; downstream systems read.
-
Do one full master data reconciliation. No shortcuts: walk through active items in ERP line by line against physical stock in WMS. Hunt three things — mismatched codes, mismatched names/specs, and zombie items with no business but still appearing on documents. On one project the cleanup surfaced 7% zombies. Leave those alive and go-live becomes a timed explosive.
-
Sign the field mapping document field by field. Not "roughly working" — every header, line item, item code, unit of measure, lot number, and expiry date gets a source, a destination, and a conversion rule. Units of measure especially: I've done four projects where ERP used cases and WMS used pieces. Get the conversion factor wrong once and an entire document batch is garbage.
-
Stand up the master data interfaces before the transactional ones. The sequence is non-negotiable: item master sync live and reconciled → locations/customers/vendors live → business documents (POs, ASNs, SOs, returns) last. Reversing this order means you're testing against false positives.
-
Test with real dirty data. The test scenarios must include the usual suspects: an item code with an extra digit, a customer name with a trailing space, an overlong lot number. Integration tested only on clean data is integration not tested at all.
-
The night-before-go-live checklist (I use this on every project):
- Are the three coding standards frozen, with zero open change requests?
- Has the full ERP-to-WMS item reconciliation completed with zero discrepancies?
- Are location masters synced to WMS, and have empty locations been verified?
- Are zombie items marked inactive, and can new documents no longer reference them?
- Have the unit conversion factors been reviewed item by item?
- Has the full business-day replay (PO → receiving → putaway → wave → shipping → callback) run green end to end?
- What's the rollback plan if interfaces fail massively on go-live day?
Master data sync frequency is worth pinning down too. Items, customers, and vendors should move near-real-time: an item created in ERP must reach WMS within minutes, not overnight. Location masters live inside WMS and take effect immediately. I've seen a project batch item syncs at midnight; at 10 AM the warehouse needed to ship a new product whose item still hadn't arrived, so pick tasks couldn't be created. Near-real-time is the floor: if latency exceeds 15 minutes, the interface should raise an alert.
When you can go transactional first: the hybrid play
There are exceptions. If you're dealing with: a greenfield warehouse, zero legacy inventory, under 500 items, and only two document types — then run a hybrid: master data rules and transactional interfaces in parallel, but master data interfaces still go live first. That's really just master-data-first with a compressed timeline.
The only genuine exception I've seen: the ERP is a legacy system nobody can touch, and WMS is the new build, with the old ERP having no master data governance at all. My play there is to hand all master data authority to WMS — WMS creates items, locations, customers; ERP only sends documents and receives callbacks. "Master data follows the new system" is still unifying the language before starting the conversation.
Blow-up watch: the traps that detonate on go-live night
Trap one: mismatched field lengths. ERP allows 50 characters for the customer short name; WMS accepts 30. Nobody checked before go-live. Customer names got truncated in transit, and the same customer had different names in the two systems — reconciliation impossible. Fix: write the length into the mapping table for every field, and define the overflow policy (truncate or error out) — pick one.
Trap two: conversion factors buried in code. Some integrators hard-code "1 case = 24 pieces." The client changes the packaging spec; the interface doesn't error; the data quietly goes wrong for three months. Fix: the factor must be a field in item master data, traveling with the item — never hard-coded.
Trap three: return documents referencing dead items. The item got deactivated in ERP, but a return in WMS still references it. The document arrives, the interface throws "item not found," and the return process freezes. Fix: before deactivating an item, check WMS for open documents referencing it — a pre-validation gate.
Trap four: organization-dimension misalignment. ERP's org structure doesn't map one-to-one to WMS warehouse codes; the interface passes an org code and WMS resolves it to the wrong building. Fix: org mapping gets its own table, never mixed into the item mapping.
For your next ERP-WMS integration project, make the first line on the kickoff whiteboard read "master data first" — and get all three parties to sign it. You'll thank me. Every project I've run in that order had a quiet go-live night. Every one that ran the reverse order has called me in a panic at midnight.
By Lin Xiao, WMS Implementation Consultant
