When a dashboard lies, the chart library is rarely the culprit. More often: Nanshan One, Nanshan, SZ-NS, and Nanshan District Store are the same store and appear once on each of four charts. You think you covered four locations. You only covered chaotic codes.
What the chart shows
- 28 stores, Nanshan contributes the most
- 1,200 customers, the head looks scattered
- Many slow SKUs—actually the same item split
After codes are reconciled
- 22 physical stores; Nanshan was counted four times
- One customer, three tax IDs plus a nickname
- Only a few materials are truly slow; the rest are aliases
Master data is not “another master-data management wall.” It is a rule: who may create a store, customer, material, or org; how merges work; how old documents point after a code is retired. Without that rule, any roll-up can treat duplicates as growth. Excel weekly reports often break first on a VLOOKUP to the wrong code; see Excel weekly reports break once you scale. Multi-system conflict is half algorithm, half code; see When every system calculates its own numbers, what breaks.
Visualization amplifies the lie. More map dots look better. A longer ranking looks like effort. Drill-down lies too: open “Nanshan” and you see half the documents; the other half sits under “SZ-NS.” If you cannot open, or open a fragment, the dashboard loses the ability to be challenged; see If a number cannot open a document, what is the dashboard for. Chain roll-up is especially dangerous. HQ watches the wrong store; see After stores roll up, how HQ watches outlier stores.
Mid-size companies do not need a master-data platform suite first. Pick three codes operations must use: store (or plant), material (or SKU), customer (or dealer). Build a mapping table and a retire rule. New systems must map on connect. How to trace numbers that are still wrong, see The system is live and numbers are still wrong. How to trace it. When quality is poor, clean critical codes before you draw; see Data quality is poor. Build the dashboard first, or clean first.
Before DaXi draws a dashboard, we ask for the state of codes: how many sets, who issues them, whether retire exists. If that cannot be said, phase one is a mapping table, not a gauge. To turn a lying chart into numbers you can reconcile, go to the Data service page and write how stores and materials are coded today.
One customer with several tax IDs does not have to merge on day one, but the chart must expand “likely the same entity.” Hiding duplicates to inflate customer count is worse than no visualization. APIs use the unique code. The display layer can show an alias; see Once APIs are standardized, the next chart is faster.
Aliases can be searched. They cannot be the roll-up key
Keep aliases for search. Do not let them enter the roll-up primary key. Separate display name from code so a ranking does not split one entity.
Do not force-change codes on historical documents. Map them. Force-changes break finance reconcile. New documents use new codes. Old ones use the map. That is safer than a one-cut wash.
Close who may issue codes. Store codes from an operations master-data role, materials from planning or purchasing, customers from finance and sales sharing the tax ID. If anyone can invent a name in Excel, visualization will keep lying. Why a dashboard must sit on codes you can challenge, see Why an operations dashboard is not optional. Crew names and machine IDs are master data too. Map them before they hit a wall screen; see How a shop-floor screen keeps crews out of the office. Mid-size companies can start with a mapping table. They do not have to buy a master-data suite; see Do mid-size companies have to buy a data platform.