Master Data Management (MDM)

The foundational process of survivorship—ingesting data from multiple source systems, cleansing it, and applying configurable match-and-merge rules to construct a single, best version of the truth for each critical business entity (e.g., Customer, Product, Supplier).

• Match Rules: Deterministic (exact field match) or probabilistic (fuzzy logic, phonetic algorithms like Soundex).
• Survivorship Strategy: Defines which source system 'wins' when attribute values conflict (e.g., CRM always wins for phone number, ERP always wins for billing address).
• Merge & Unmerge: The golden record must maintain a persistent, non-recyclable UUID to track lineage back to original source records.

MDM systems must model complex, often overlapping, real-world relationships between entities. This goes beyond flat records to define parent-child roll-ups and peer-to-peer mappings.

• Legal Hierarchies: Tracks the actual corporate structure (e.g., Subsidiary A reports to Ultimate Parent B) for financial consolidation.
• Sales Hierarchies: Maps the commercial territory structure, which may differ entirely from the legal structure.
• Product Taxonomies: Manages deep category trees (UNSPSC, eClass) to ensure a single product is correctly classified for procurement, sales, and logistics simultaneously.

The human-in-the-loop workflow engine that manages exceptions. When automated match algorithms fall below a defined confidence threshold, records are routed to a Data Steward for manual review via a task-driven inbox.

• Role-Based Security: Stewards see only the entities and attributes relevant to their domain (e.g., Customer Steward vs. Material Steward).
• Audit Trail: Every manual merge, attribute override, or status change is logged immutably to satisfy regulatory compliance (GDPR, CCPA).
• Policy Enforcement: Validates data against business rules in real-time (e.g., 'A customer cannot be marked as Active if the address is null').

Modern MDM platforms are multi-domain, meaning they manage the lifecycle of multiple entity types within a single, unified data model rather than siloed applications.

• Party Domain: Customers, Suppliers, Employees, Patients.
• Product Domain: Finished goods, Raw materials, Services, Assets.
• Location Domain: Stores, Warehouses, Regions, Geo-coordinates.
• Reference Data: Standardized code sets (ISO country codes, currency codes) mapped across the enterprise to ensure semantic consistency in analytics.

The golden record is useless if it stays locked in the MDM hub. Bi-directional synchronization pushes the trusted master data back out to subscribing operational systems (CRM, ERP, eCommerce) in near real-time.

• Message Queues: Uses JMS or Kafka to publish change data capture (CDC) events.
• Global Data Synchronization Network (GDSN): For product data, MDM syndicates standardized attributes to trading partners and retailers via certified data pools.
• Loose Coupling: Source systems continue to operate independently; the MDM acts as the central broker, mapping local keys to the global Enterprise Identifier.

The algorithmic core that answers the question: 'Are these two records the same thing?' It uses probabilistic record linkage (Fellegi-Sunter model) to weigh the agreement and disagreement of multiple attributes.

• Blocking: Reduces the computational search space by grouping records into blocks (e.g., only compare records with the same ZIP code) to avoid an O(n²) comparison explosion.
• Fuzzy Matching: Tolerates typos, transpositions, and abbreviations using algorithms like Levenshtein distance or Jaro-Winkler.
• Identity Resolution: Specifically tuned for person data, resolving identity across systems with sparse or conflicting PII.