Duplicate MRO items to SAP–Maximo MDM with AI similarity; requisitioners find parts

Overview

A heavy equipment manufacturer was struggling with duplicate and inconsistent maintenance, repair, and operations (MRO) item records across plants. Requisitioners could not confidently find the right part, planners triggered accidental reorders, and storerooms carried excess stock of look?alike items. Intelligex implemented a master data management initiative that added AI?based similarity search and governed workflows for new item creation, integrated into the client’s existing enterprise resource planning and maintenance systems. With a standard data model, classification, and human?in?the?loop stewardship, requisitioners located the correct part quickly, planners consolidated demand on the right stock codes, and inventory signals became more reliable.

Client Profile

• Industry: Heavy equipment manufacturing
• Company size: Multi?plant operations with centralized procurement and regional maintenance teams
• Stage: Established operations with legacy data and plant?specific processes
• Department owner: Procurement, Supply Chain & Logistics
• Other stakeholders: Maintenance and Reliability, Engineering, Production Planning, Plant Storerooms, Finance and Inventory Control, IT and Data Governance

The Challenge

The MRO item master had grown organically through plant expansions, system migrations, and vendor transitions. The same bearing, seal, or hose could appear under several stock codes with slightly different descriptions, manufacturer names, or units of measure. Some records carried free?text notes from past buyers, while others were sparse and hard to interpret. Requisitioners often could not find items without knowing the exact phrasing used at a specific plant, so they opened new item requests rather than risk a wrong substitution. Planners saw demand dispersed across duplicates, obscuring true usage and safety stock needs. The result was overstock in some locations and emergency buys in others.

Legacy processes for item creation were decentralized. Plant storerooms or maintenance teams requested new items with minimal validation. Buyers and data admins did their best with what they had, but there was no effective way to detect near?duplicates before records were approved. Attempts to clean the data relied on one?time spreadsheet projects, which improved a subset of records but did not change how new items were introduced. The organization needed a sustainable fix that would work inside current systems and daily workflows.

Budget and time did not allow for a wholesale platform change. The client relied on SAP ERP for procurement and inventory, and IBM Maximo for maintenance and work management. Any solution had to integrate with both, align with existing vendor and item masters, and respect each plant’s operating tempo. The aim was not to slow down maintenance, but to remove friction from finding and stocking the right parts.

Why It Was Happening

Root causes traced back to fragmentation and weak governance. Item creation followed different patterns by plant, and the data model allowed free?form text for attributes that should have been structured. Manufacturer names were typed in various forms, part numbers were stored inconsistently, and classification codes were applied unevenly or left blank. Without a consistent taxonomy or attribute templates, similar items were hard to compare and even harder to detect as duplicates.

There was also no system?supported way to suggest existing items during request or approval. Stakeholders relied on keyword search, which breaks down when descriptions are inconsistent. A digital shelf of look?alike items grew over time, hiding the true picture of demand and availability. Downstream, that ambiguity showed up as line changes during purchasing, mismatched receipts, and excess inventory. The absence of clear ownership for template content and approval gates allowed the drift to continue.

The Solution

Intelligex delivered a master data management layer that standardizes MRO item data, detects duplicates through AI?based similarity, and enforces a governed new item workflow. We did not replace core systems. Instead, we integrated with SAP ERP and IBM Maximo for item synchronization and request handling, added a similarity search service that understands manufacturer part numbers and technical attributes, and embedded human review at decision points. A harmonized taxonomy and attribute templates ensured that items could be compared on like?for?like terms.

• Integration with SAP ERP for item master, vendors, and purchasing views; bi?directional sync with IBM Maximo for maintenance catalogs and storeroom references.
• Similarity search using embeddings over descriptions, manufacturer names, and part numbers, backed by Elasticsearch vector search documentation patterns for approximate matching.
• Standard classification and attribute templates aligned to UNSPSC and category?specific attributes for bearings, seals, electrical, hydraulics, and fasteners.
• Manufacturer and manufacturer part number normalization with alias tables and validation against approved source lists.
• Governed new item workflow with pre?submission suggestions of potential matches, required attachments for technical specs, and routed approvals to data stewards and engineering where needed.
• Duplicate management tools to merge records under a single golden item, with redirect rules for historical references and purchase history consolidation.
• Dashboards showing data quality trends, duplicate volumes by category, open requests by plant, and cycle times for approvals.
• Role?based permissions, reason codes for overrides, and an immutable audit log of changes, merges, and approvals.

Implementation

• Discovery: Profiled the item master across plants to quantify duplicate patterns and common description variants; reviewed current item creation steps; cataloged critical attributes by category with Maintenance, Engineering, and Procurement; identified integration touchpoints in ERP and EAM.
• Design: Defined the canonical item model, taxonomy, and attribute templates; established naming and description standards; designed the new item workflow with review gates and exception paths; specified similarity thresholds and match tiers for suggestions.
• Build: Implemented data services for classification, manufacturer normalization, and attribute validation; stood up the similarity search service and indexing pipelines; built connectors for ERP and Maximo; developed merge tooling and redirect logic for superseded items.
• Testing/QA: Ran similarity search in shadow mode against live requests, showing match suggestions without blocking the legacy process; tuned embeddings and thresholds based on steward feedback; validated merge outcomes in a non?production clone with representative purchasing and maintenance scenarios.
• Rollout: Phased deployment by category and plant; started with high?volume MRO classes and storerooms ready to adopt governance; enabled suggestion?only first, then activated required review gates; maintained a clear rollback plan at each cutover.
• Training/hand?off: Delivered role?based training for requisitioners, planners, and data stewards; published quick guides on naming standards and attachment requirements; established a governance council for template changes and periodic audits; incorporated human?in?the?loop review for complex or safety?critical items.

Results

Requisitioners searching for parts began to see relevant, existing items presented first, with clear attributes and approved manufacturers. The urge to create new codes diminished because the right items were easier to find and trust. Planners observed consolidated demand on the correct stock codes, enabling more accurate reorder points and fewer accidental reorders of near?identical parts. Storerooms gained confidence that stock on the shelf aligned with what maintenance actually requested.

Data stewards spent less time firefighting duplicates and more time improving templates and standards. Merged items retained procurement and usage history through redirect rules, so reporting did not break when duplicates were retired. Purchasing saw fewer last?minute line changes and clearer supplier alignment to the approved item and manufacturer lists. Overall, cycle time for finding and approving the right part improved, rework decreased, and auditability increased without disrupting maintenance schedules.

What Changed for the Team

• Before: Requisitioners tried multiple keyword searches and often gave up, filing new item requests. After: Similarity search suggested existing items with matching attributes and approved manufacturers.
• Before: Data admins approved items with limited context, and duplicates slipped through. After: Governed workflows required attribute completeness and showed likely duplicates before approval.
• Before: Planners saw fragmented demand across look?alike items. After: Consolidation onto golden records clarified true usage and stocking needs.
• Before: Merges were avoided for fear of breaking downstream links. After: Merge tooling preserved history and redirected old references safely.
• Before: Standards lived in slide decks and were inconsistently applied. After: Naming, classification, and attribute templates were enforced in the system.

Key Takeaways

• Reliable MRO data starts with a governed model, not a cleanup project; prevention at item creation is the leverage point.
• AI?based similarity search reduces duplicate entry by surfacing existing items at the moment of request and approval.
• Category?specific attribute templates make items comparable and help requisitioners trust what they find.
• Integrating with ERP and EAM preserves existing workflows while improving data quality in place.
• Human?in?the?loop stewardship is essential for safety?critical or complex items and builds trust in automation.

FAQ

What tools did this integrate with?
We integrated the MDM layer with SAP ERP for item and purchasing data and with IBM Maximo for maintenance catalogs and storeroom references. The similarity service followed patterns from Elasticsearch vector search documentation. The approach also aligns with governance capabilities available in SAP Master Data Governance.

How did you handle quality control and governance?
We established a governed workflow with mandatory attributes, attachment requirements for technical specs, and role?based approvals. Data stewards owned classification templates and naming standards. Overrides required reason codes, and all actions were captured in an immutable audit log. Periodic sampling and data quality dashboards supported continuous improvement.

How did you roll this out without disruption?
We ran similarity suggestions in shadow mode first, so requisitioners and stewards could see recommendations without changing behavior. Deployment was phased by category and plant, and we enabled review gates only after thresholds were tuned. Clear rollback paths and targeted training ensured maintenance work orders and purchasing continued without delay.

How was the similarity search built and tuned?
Descriptions, manufacturer names, and part numbers were embedded into vectors and indexed for approximate matching, following established practices from Elasticsearch. We combined semantic similarity with exact and fuzzy token matches and normalized fields like manufacturer and unit of measure. Thresholds were tuned with steward feedback to balance helpful suggestions with noise control.

How did you manage legacy duplicates and inventory implications?
We used merge tooling to consolidate duplicates into a golden item while preserving historical references. Purchasing, receiving, and work order links were redirected, and stock on hand was aligned to the surviving item. Communication to plants flagged the change in catalogs, and planners reviewed reorder settings after consolidation to ensure availability and avoid excess.