Migration is the moment of truth for the cleanup. Whatever is wrong in the old system arrives wrong in the new one, only faster and with more users watching.
What is loaded, and who signs
| Object | Records | Owner for sign-off |
|---|---|---|
| Materials (active only) | 18,400 | All field owners |
| Bills of material | 6,400 | Engineering |
| Routings | 6,400 | Industrial engineering |
| Suppliers | 1,240 | Purchasing |
| Customers | 2,860 | Sales |
| Storage locations with stock | 9,600 | Logistics |
| Total | 44,900 |
Open orders follow at cutover: 1,900 purchase orders, 620 production orders and 1,450 sales orders. The 12,800 inactive materials stay in the archive, which is the cheapest cleanup in the whole project.
Three test loads
Each test load is a full load of all objects into a test system, followed by a check: record counts, rejected records by cause, and twenty end-to-end cases per department with real orders.
| Test load | Rejected records | Error rate | Main finding |
|---|---|---|---|
| 1 | 6,286 | 14.0 % | Missing mandatory fields, units without conversion |
| 2 | 1,796 | 4.0 % | Bills with blocked components, invalid work centers |
| 3 | 269 | 0.6 % | Single cases, corrected by hand |
The exit criterion was agreed before load one: below 1 percent rejected records in the third load, and zero in the 2,900 materials of wave 1.
Five rules keep the loads honest. Correct errors in the old system, not in the transfer file, or the next load brings them back. Give every object one owner who signs the result. Map every field with source, target, rule and default, and decide the fields without a target instead of forgetting them. Freeze master data changes one week before cutover with a short written list of exceptions. Count stock before cutover and load quantities per location, never totals per material.
Go-live is a decision
The gear unit plant agreed 40 criteria in five areas, 28 of them must, months before the date.
| Area | Criteria | Of which must | Examples |
|---|---|---|---|
| Data | 8 | 7 | Third test load below 1 percent, stock counted, owners signed off |
| Processes | 8 | 6 | Order to delivery and purchase to payment tested with real orders |
| People | 8 | 5 | Key users per department and shift, all users trained |
| System | 8 | 5 | Interfaces, printers and scanners tested, class 1 defects closed |
| Cutover | 8 | 5 | Cutover plan with owners, fallback point, reduced plan in week one |
The decision rule was written before the first check: go when all 28 must criteria are met and at least 90 percent of all 40. Every open criterion gets an owner, a date and a workaround.
| Check | Criteria met | Must met | Decision |
|---|---|---|---|
| Four weeks before | 34 of 40 | 25 of 28 | Not ready |
| One week before | 39 of 40 | 28 of 28 | Go |
Four weeks before, three must criteria were open: the stock count, the authorizations per role and the fallback plan. Writing the rule early is what made that conversation short. Nobody had to argue about how ready is ready while the date was burning.
The first four weeks
The production plan of week one is cut by 30 percent and the safety stock of critical materials is raised before cutover. A 15-minute daily review with key users and field owners goes through open defects, data errors found and bookings not done. Every data error found in live operation is corrected at its source the same day and counted, because the count shows whether the new process holds.
Counting starts on day one. Inventory record accuracy is the fastest signal that the new bookings work, long before any report is trusted.
Twelve months of data work paid for themselves before the new system went live: on-time delivery 93 percent, the cost of bad data down from EUR 741,600 to EUR 305,560 per year. The system started on clean data, which is the only way a system ever starts well.