Fortune 500 Supply Chain Intelligence

Challenge

A Fortune 500 supply chain organization needed reliable procurement and delivery intelligence across a multi-phase dataset with no consistent baseline methodology. Raw data contained measurement errors from supplier change orders, inconsistent fiscal-year logic, and unvalidated DEI spend classifications. The team needed a defensible, auditable analytics foundation before any dashboard could be trusted for executive decision-making.

Methodology

Phase 1: Supplier Management and Data Architecture (P2)

Profiled source schemas across supplier registration extracts. Built a master normalization pipeline in Python: P2_profile_source to map field types and identify gaps, P2_clean for schema standardization, and P2_validate for approval-status crosscheck. Centralized all input and output paths in path_config.py to ensure reproducibility across phases.

Phase 2: Delivery Performance and DEI Pipeline (P3 and P4)

P3. Ingested order records across multiple fiscal years using regex-based fiscal-year filters and cancellation handling. Built SCO exclusion logic to separate non-operational change order events from true delivery performance. Cross-validated P3 outputs against P4 on shared benchmarks to confirm consistency across phases.

P4. Built a multi-script DEI spend classification and validation system covering gap analysis, filter verification, and reclassification checks. Patched Tableau workbook XML directly to reconcile dashboard logic against source-of-truth data.

Phase 3: POS Pipeline and Backlog (P8, P9, P10)

Built three delivery pipelines from a unified source extract.

• P8. MTM firm backlog generator with SCO exact-match filter and site-and-product aggregation; delivered as Excel.
• P9. Open PO demand-excess aggregator at the site, product, and month level; delivered as Tableau.
• P10. POS pipeline outlook extractor with positive-value filters and site-level monthly pipeline output; delivered as Tableau.

Each phase ran through a validation gate before loading into the workbook layer.

Key Findings

• SCO exclusion logic validated across phases. Manual flagging without systematic validation had created measurement error across the full dataset. The automated exclusion pipeline now produces an auditable record for every removed record and prevents recurrence.
• DEI spend classified and validated across supplier categories. Spend data was previously unstructured with no consistent classification methodology. A single source of truth now enables compliance reporting and supplier-diversity strategy decisions at the program level.
• POS pipeline outlook operational across site and product dimensions. Pipeline visibility had been fragmented across source extracts with no consolidated view. The monthly pipeline dashboard now enables inventory and demand planning at the site level.
• Corrected delivery trend isolated from SCO noise. SCO spikes clustered in mid-period had been artificially depressing headline delivery numbers. Stakeholders now have a baseline that reflects true operational performance, not data artifacts.

Tools and Approach

• Python (Pandas, openpyxl): ETL pipeline across six phases; profile, clean, enrich, validate pattern applied consistently
• Tableau (.twbx): dashboards across P2 through P10; XML-patch scripts used to reconcile workbook logic against source data
• Excel: MTM backlog deliverable (P8); multi-year source extracts
• Data governance: centralized path configuration, audit-trail documentation, cross-phase validation gates

Results and Impact

• Delivered Tableau dashboards spanning supplier management, delivery performance, DEI spend, and POS pipeline workstreams
• Built a reproducible six-phase ETL architecture with a consistent profile, clean, enrich, validate pattern across all phases
• Established auditable SCO exclusion logic separating operational performance from change-order noise
• Validated DEI procurement spend classification across supplier categories for compliance reporting
• Deployed cross-phase validation gates to confirm output consistency before dashboard loading

What I Would Do Differently

A unified semantic layer connecting all six phases would eliminate the cross-phase validation overhead that currently requires manual reconciliation. The Tableau-patch approach works, but a live-connection architecture would remove the need for workbook XML patching entirely. An automated regression test suite against known baseline outputs would further reduce the manual validation burden on each sprint cycle.

---

MBA practicum context. Raw source data, client field names, vendor identifiers, and performance metrics are not published. Methodology, pipeline architecture, and dashboard structure are public; proprietary inputs are not.