Data quality is poor. Build the dashboard first, or clean first

“Wait until the data is clean, then visualize” sounds rigorous. In practice it never starts. Source systems still take wrong codes, master data still lives in one sheet per person, cleanup scripts have no business owner, and a week later it is dirty again. A dashboard is not a prize for clean data. It is a mirror that makes pollution visible. Ship a thin version first so quality work has something to accept against.

• 01
  One customer, many codesSales splits into strangers. HQ thinks the market is more fragmented than it is.
• 02
  Materials named but uncoded, or one item with many namesStock and cost do not match. Margin jumps between stores.
• 03
  Returns ignored or thrown into sales at randomThe first weekly-meeting item is a definition fight. A pretty chart does not help.
• 04
  Each team cuts its own time windowOperations looks at order time, finance at invoice time. Two “todays” never bite.
• 05
  Manual edits with no trailIt looks aligned until someone asks for a source. Next time nobody trusts it.

The right order is not “wash the whole warehouse white.” It is “lock must-watch KPIs and queue pollution by decision harm.” If margin is shredded by material codes, fix materials and documents first—not ten years of history. When master data is missing, visualization lies; see Without master data, visualization lies. When the system is live and numbers are still wrong, see The system is live and numbers are still wrong. How to trace it. Write the definition first so “clean” has a meaning; see After one definition, weekly meetings argue less. A cleanup with no audience has no deadline, and nobody will sign “how clean is done.”

Some teams put a wall screen up first to force better entry. That works only if the chart drills and a dirty point opens a responsible person. A decorative screen you cannot click makes the floor feel smeared by “the system,” and two ledgers continue. Why drill-down is a precondition, see If a number cannot open a document, what is the dashboard for. In a chain, one forked code forks the roll-up; see How chain roll-up avoids forked definitions. Five years of dirty orders can freeze for display and stay out of bonuses. Incremental data from a midnight cutoff on the new definition is more deliverable than a fantasy full wash.

Cleanup needs a deadline and a responsible role

Write each stain as: who changes the source system, who blocks at the API, who owns the number in the weekly meeting. Cleanup with no role is an IT night shift. Daylight returns and it is dirty again. History can freeze. Incremental starts on a date. DaXi puts “KPIs we watch first” and “fields we clean first” on the same list so the project does not stall on a boundless governance slogan. To start from a pollution list, go to the Data service page and name the number you least dare to reconcile. Sequence, see From brief to launch: how a data project actually runs.