Every wrong value in the system was once entered by somebody who had no rule and no responsibility for it. Cleaning the data without changing that produces the same mess two years later.
Fields belong to departments, not to IT
Data belongs to the department that knows the truth and lives with the result. The gear unit plant listed 64 planning-relevant fields and assigned each one to exactly one owner role.
| Owner role | Fields | Examples |
|---|---|---|
| Engineering | 14 | Description, base unit, classification, bill of material |
| Production planning | 13 | Planning method, lot size, safety stock, planner |
| Purchasing | 11 | Lead time, minimum quantity, supplier, purchasing unit |
| Logistics | 9 | Storage location, packing unit, count class |
| Industrial engineering | 8 | Routing, work center, run time, setup time |
| Controlling | 6 | Standard price, valuation class, profit center |
| Quality | 3 | Inspection requirement, shelf life, certificate |
At the start, 23 of the 64 fields had a named owner: 36 percent. The other 41 were filled at creation by whoever raised the request and never looked at again.
Owner does not mean typist. The owner decides the value and answers for the hit rate of that field. Who enters it is a second question, and most plants use one or two data stewards who enter what the owners decide.
The matrix and the board
The matrix has one row per field with the owner, the roles consulted, the rule that says when the value is correct, and the review interval. It is signed by the heads of all departments and hangs where creation requests are processed.
Around it sit three roles: the field owners, the data stewards who process requests and run the audits, and a data board of the plant manager and the seven owners that meets monthly for 45 minutes to look at the key figures, the cleanup tracker and any rule that crosses departments.
Expect the fight to be about the fields between departments, such as the in-house lead time of manufactured parts, which planning, industrial engineering and production all feel they own. One owns, the others are consulted. Decide it in the room and write it down.
Five steps, five days
About 45 materials are created per week in the plant. At the start, a creation took 11 working days and 38 percent of new materials were released with empty planning fields, which means they were planned with defaults from day one.
| Step | Who | Target |
|---|---|---|
| Request with a duplicate check | Requester | Day 0 |
| Basic data: description, unit, classification | Engineering | Day 1 |
| Sourcing data: lead time, lot size, supplier or routing | Purchasing or production planning | Day 3 |
| Logistics and cost data | Logistics, controlling | Day 4 |
| Completeness check and release for planning | Data steward | Day 5 |
The rule that carries the whole process: a material cannot be planned or ordered before the data steward releases it. After twelve months the plant was at 4 days and 3 percent incomplete releases.
Changes and duplicates
About 140 master data changes arrive per week. Each names the field, the old value, the new value, the reason and the effective date, and is released by the owner of that field. Mass changes are tested on ten records first and approved by the data board.
Before a new number is created, the requester searches by classification and key attributes, not only by description, and descriptions follow one pattern per material group: noun first, then size, then material or standard. A second number for the same part splits demand and stock and doubles the maintenance forever. The plant estimated 640 duplicates among its 18,400 active numbers, 3.5 percent, and merged them during the cleanup.
Start the ownership matrix and the creation process before the cleanup ends, so that the cleaned data meets a process that keeps it clean.