Before you extract anything
- Decide which master data objects move, and in which order: dependencies first, for example warehouses before item/warehouse records.
- Decide which records move at all. Items not sold for years and blocked customers may not need to come along.
- Agree who owns each object, and who signs off that it is ready.
Extract complete tables
Work from complete extracts, with the key fields always included, so every row stays identifiable. Splitting exports to get around record limits is a common source of missing or duplicated rows.
Check against rules, before the data moves
| Dimension | Example check |
|---|---|
| Completeness | Every stocked item has a receiving location |
| Validity | The warehouse on every item/warehouse record exists |
| Consistency | Every warehouse an item is sold from has an item/warehouse record |
| Uniqueness | No two customers share the same VAT number |
| Plausibility | Lead times fall within a sensible range |
Fix what you find at the source, in the on-premise environment, so the corrections are part of what moves.
Load with a safety net
- 1Validate every row before writing, and compare it with what is already in the target: new, changed or unchanged.
- 2Never write rows the target would refuse; report them with the reason instead.
- 3Save the records an import will change before it changes them, so there is always a way back.
- 4Load in the order of the dependencies you agreed at the start.
Checking the data with the same rules in the old and the new environment shows directly whether something was lost or changed on the way.
After the load
- Run the quality rules on the cloud environment and compare the results with the on-premise baseline.
- Put the fields that steer M3 under governance from day one, so the clean state stays clean.
- Check key transactions on real records before users do.