Data quality behaves like product quality. It is specified by rules, produced by a process, measured on samples and improved at the cause. One large cleanup is inspection without a process behind it, and it does not hold.
Seven figures, measured the same way
| Key figure | Formula | Start | Month 12 |
|---|---|---|---|
| Field hit rate | Correct field checks / all checks | 77 % | 97 % |
| Bill of material accuracy | Correct bills / bills checked | 73 % | 98 % |
| Routing accuracy | Correct routings / routings checked | 65 % | 95 % |
| Inventory record accuracy | Locations within tolerance / counted | 78 % | 97 % |
| Fields with an owner | Fields with one owner / planning fields | 36 % | 100 % |
| Creation lead time | Working days from request to release | 11 days | 4 days |
| Incomplete releases | Released with empty planning fields / created | 38 % | 3 % |
The first four come from samples, repeated every four months: 200 materials, 60 bills, 60 routings, plus the daily counts. The last three come out of the creation process itself and can be read from the system every month.
Accuracy cannot be read from a system, because the system does not know what is true. Completeness can: empty fields, invalid planners, materials without movement. Report both, and never let a complete field be reported as a correct one.
The figures the plant feels
Data figures convince data people. These convinced the plant manager: on-time delivery rose from 81 to 93 percent, assembly orders that started with missing parts fell from 23 to 6 percent, and the share of planner time spent on manual corrections fell from 30 to 12 percent.
Start where the money moves
Nobody cleans 18,400 materials with the same effort. Sort them by consumption value and the work sorts itself.
| Work package | Records | Minutes each | Hours |
|---|---|---|---|
| Wave 1: materials with 90 percent of consumption value | 2,900 | 12 | 580 |
| Wave 2: materials with the next 9 percent | 6,300 | 5 | 525 |
| Wave 3: remaining materials, mass update by rule | 9,200 | By rule | 120 |
| Bills of material with orders in the last twelve months | 2,350 | 30 | 1,175 |
| Routings with orders in the last twelve months | 2,350 | 20 | 783 |
| Total | 3,183 |
Three thousand one hundred eighty-three hours are two people for a year at 1,600 hours each. The gear unit plant named two data stewards for twelve months and gave each field owner two hours per week. The stewards prepared the values, the owners confirmed them, and wave 1 took six weeks instead of the 29 weeks that owners working alone would have needed.
Three more packages belong in the plan: 12,800 materials without movement for three years, flagged and blocked for planning; 640 estimated duplicates, merged starting with standard parts; and the 4,050 bills and routings without orders in the last year, which get the status “check before next order” instead of being cleaned now.
The tracker and the rule for done
One row per work package with records in scope, records done, hours, owner and due date. The data stewards review it weekly, the data board monthly.
A package is done when the next audit sample confirms the target, not when the last record has been touched. That rule is unpopular in week 20 and indispensable in month 12, because it is the only thing that separates a cleanup from a feeling.
Three quick rules for the first month
While the waves are being planned, three rules already move the figures: replace the system lead time of purchased parts by the median of the last ten receipts wherever the deviation is above 20 percent; flag every material without movement for three years and take it out of the planning run; and put the line material, seals, screws and labels, into the bills of the twenty products with the highest volume.