Business problem

Operations discovered late deliveries only after customers complained, with no shared view of supplier performance.

Dataset

40M fact rows of purchase orders, shipments and receipts across 320 suppliers.

Methodology

• Composite model blending imported dimensions with DirectQuery facts.
• Defined OTIF and lead-time variance measures with operations.
• Built exception queues with automated alerting.

Key findings

• Twelve suppliers drove 60% of late deliveries.
• Lead-time variance, not average lead time, predicted stockouts.

Recommendations

• Put the twelve worst suppliers on a performance plan.
• Set safety stock from variance rather than average lead time.