Challenge
The same physical material recorded under different codes and descriptions across plants, regions and subsidiaries.
Master Data Optimization at FMCG Scale
Find duplicate and near-duplicate materials across plants and ERPs.
An AI-assisted similarity engine clusters look-alike materials, quantifies consolidation savings per cluster and gives data stewards a review workflow.
Delivered
A similarity engine using descriptions, attributes and purchasing patterns, plus a steward review workflow.
Decision support
Each cluster comes with a consolidation business case: keep, block or obsolete.
Challenge
Global manufacturers carry hundreds of thousands of materials in their ERPs, with the same physical material recorded under different codes, names, and descriptions across plants, regions, and acquired subsidiaries. The result: missed volume discounts, overstocking, maverick spending, and an inability to consolidate procurement spend. Master data clean-up is universally known to be needed and universally deprioritized - until the cost of inaction adds up.
Approach
We built an AI-assisted similarity engine that detects duplicate and near-duplicate materials across plants and ERPs. The engine uses material descriptions, attributes, and purchasing patterns to cluster look-alike materials, then quantifies the consolidation savings per cluster - based on real purchasing volumes and vendor discount curves.
A steward workflow lets data managers review each cluster and decide: keep, block, or obsolete. Each clean-up action carries a business case behind it.
Inside the similarity engine
Two screens from the standalone demo version of the engagement; the figures are illustrative.
Demo screen · 01 / 02
Cluster materials
One row per material in each similarity cluster, with material code, description, similarity score and the cluster's review status.
- Similarity score per material
- Review status: not reviewed, in progress, cleaned
Demo screen · 02 / 02
Overview
Total materials, duplicate share, inactive materials and estimated annual saving, with an illustrative duplicate-rate trend.
- Candidates for blocking or clean-up
- Estimated saving from consolidation clusters
Outcomes
The figures come from the demo set (see the note below), not from the client site.
228 materials analyzed in the demo set, with 98.2% identified as sitting in duplicate clusters
48 candidates flagged for blocking or clean-up
€1,382,973 estimated annual saving from 84 consolidation clusters (illustrative)
Faster clean-up with a business case behind every change
Lower overstocking, fewer missed discounts, less maverick spend
One trusted material view supporting better sourcing and operational decisions
Numbers shown are illustrative - derived from a standalone demo version of the engagement built on synthetic data. Actual outcomes at the client site are NDA-protected.
Technology
Python
AI/ML similarity engine
ERP integration
Power BI
Solution areas: Data Foundations · Data Products · AI Solutions
