Nine data model mistakes manufacturers make

Plant, line, cell, asset, work centre, part, order, lot, operator, shift. If two systems disagree on what an asset is, no downstream metric can reconcile — and no amount of dashboard work will fix it.

Write the definitions down before the first integration. One page. Reviewed by production, quality, maintenance and finance together.

When OEE, yield or on-time delivery is calculated inside a BI Dashboard, every new Dashboard re-implements it slightly differently. Define the metric once, in the data layer, with the calculation visible — then let every Dashboard consume it.

Manufacturing data is time-series data with business meaning: local plant time, shift boundaries that cross midnight, daylight saving, downtime intervals that overlap, and events recorded at different resolutions. Storing everything as naive local timestamps destroys reconciliation later.

Store UTC with an explicit plant timezone, keep the shift calendar as data, and never let a Dashboard infer a shift from a timestamp.

A snapshot tells you what the state is now. An event tells you what changed and when. Root-cause work, traceability and any 'why did this happen' question needs events. Plants that only replicate current-state tables can never answer their most valuable questions.

Routings, BOMs, cycle times and standards change. If they are stored without effectivity dates, historical analysis silently uses today's standard against last year's production, and every trend is wrong.

Reprocessing everything nightly looks simple until volumes grow — then windows are missed, and the morning meeting has yesterday's data. Incremental loads with watermarks and late-arriving-record handling are more work once and less work forever.

When a number is challenged in a meeting — and it will be — someone must be able to show its source, its transformation and its refresh time in under a minute. Without lineage, trust never accumulates and everyone reverts to their own spreadsheet.

A Flow that stops is safe. A Flow that keeps running with partial data is dangerous, because the dashboard still looks fine. Every feed needs freshness, volume and schema-change warnings routed to a human who can act.

Shaping data to fit whatever BI product was bought creates a model that cannot answer the next question. Model the manufacturing domain; project it into whatever tool the business prefers this year.