The challenge
The operations team spent every Monday morning rebuilding the same Excel file. Three plant managers exported CSVs from two ERPs and an MES system, reconciled them by hand, and emailed a static PDF to the executive committee on Wednesday. By the time the board reviewed Friday, the numbers were already five days old.
Two consequences:
- Production decisions reacted to last week, not last shift. Defect spikes were caught after the batch had shipped.
- The same three people spent ~14 hours a week on a task that produced no insight beyond what the raw systems already contained.
They had tried a BI tool eighteen months earlier. The implementation stalled because no one owned the data quality problem upstream.
Our approach
We refused to start with the dashboard. The dashboard was the easy part.
Week 1–2 — Data audit. We mapped every field that landed in the weekly Excel back to its source system. 30% of the columns came from manual edits the team had stopped questioning. We documented each one and decided, with the operations director, which were genuinely needed and which were historical artefacts.
Week 3–5 — Pipeline. A single ingestion layer pulling from the two ERPs and the MES, with named transformations replacing the Excel logic. Every transformation has a test against the prior week’s known-good Excel output, so we knew exactly when our numbers diverged from theirs and why.
Week 6–7 — Dashboards. Three views: shop-floor (refreshed every 15 min), plant manager (hourly), executive (daily). Each view shows fewer numbers than the old PDF, not more. We pushed back hard on every “can we also add” request — the win was clarity, not coverage.
Week 8 — Handover. Two ops analysts trained to maintain and extend the pipeline. We stayed for an additional 30 days as on-call backup.
The outcome
The Monday Excel ritual is gone. Reporting effort dropped from ~14 to ~5 hours a week, redirected to actually investigating anomalies the dashboard now surfaces.
Two specific wins in the first quarter post go-live:
- A defect spike on Line 2 was caught within the same shift instead of after the batch shipped. Estimated avoided rework: €18k on that single incident.
- Inventory carrying cost dropped because the planner could see WIP in real time and stopped pre-buffering against the reporting lag.
The CEO’s quote, which we put on our home page with permission: “We went from 3-day reporting cycles to real-time dashboards in eight weeks.”
Most importantly, the team owns it. Six months in, they have added two new views without us, and the data quality has improved because the upstream owners now see their numbers daily.