Master Data Management (MDM)

The central mechanism of MDM is the consolidation of fragmented, duplicate records into a single, best-curated version known as the golden record. This involves applying survivorship rules to resolve conflicting attributes from multiple source systems. The golden record is not necessarily a physical merge; it can exist as a virtual consolidated view that links back to source data while providing a unified, trusted identifier for the entity.

MDM relies on sophisticated matching algorithms to identify that disparate records refer to the same real-world entity. This process, known as entity resolution or identity resolution, moves beyond simple deterministic matching to include:

• Probabilistic linkage using statistical likelihood ratios
• Fuzzy matching with edit distance thresholds and phonetic encodings
• Transitive closure to group multiple pairwise matches into a single entity cluster

MDM distinguishes between two critical architectural roles:

• System of Record (SOR): The authoritative source where master data is created and maintained. The MDM hub itself often becomes the SOR for core entity identifiers.
• System of Reference (SOR): A read-only, consolidated view that harmonizes data from multiple systems without owning the data. This distinction allows MDM to coexist with legacy applications while providing a unified access layer.

MDM is inseparable from data governance. It operationalizes governance policies by enforcing:

• Data stewardship workflows for manual curation and dispute resolution
• Business rules for validation, standardization, and survivorship
• Role-based access controls to manage who can create, update, or merge master data Without governance, an MDM hub degrades into another unmanaged data silo.

While early MDM implementations focused on a single domain like customer or product, modern platforms support multiple interconnected domains. A multi-domain MDM manages the relationships between entities—linking a customer to their contracts, locations, and products—creating a rich, navigable semantic network that reflects the real-world connections within the enterprise.

MDM architectures are categorized by how the master data is stored and accessed:

• Registry Style: Maintains a thin index of cross-referenced keys; source systems remain the owners of attribute data.
• Consolidation Style: Periodically aggregates and cleanses data from sources into a central hub for reporting and analytics.
• Coexistence Style: Allows updates in both the MDM hub and source systems with bidirectional synchronization.
• Centralized Style: The MDM hub is the sole system of record for all master data attributes.

The single, best-curated version of a master data entity created by resolving and merging all conflicting attributes from duplicate records. Survivorship rules determine which source system's value prevails when contradictions arise—for example, preferring the most recently updated address or the billing system's tax ID. The golden record serves as the canonical reference that downstream analytics, CRM, and ERP systems consume, ensuring every department operates from identical, trusted data.

The comprehensive process of identifying, linking, and merging disparate records that refer to the same real-world entity within or across datasets. Also known as identity resolution or data matching, it employs both deterministic rules (exact field matching) and probabilistic algorithms (Felligi-Sunter statistical weighting) to overcome typographical errors, missing values, and inconsistent formatting. Entity resolution is the computational engine that transforms fragmented data silos into unified, actionable records.

Two fundamental matching paradigms. Deterministic linkage classifies a pair as a match only if a predefined set of identifiers agree exactly—offering high precision but low recall when data contains typos. Probabilistic linkage uses statistical likelihood ratios (Felligi-Sunter model) to compute match weights based on field agreement and disagreement patterns, tolerating errors. Modern MDM systems often combine both: deterministic rules for high-confidence matches, probabilistic scoring for ambiguous pairs requiring clerical review.

The critical preprocessing step of mapping semantically equivalent attributes from different source schemas to a common canonical format before record linkage can begin. For example, mapping cust_name in one system and fullName in another to a unified customer_full_name field. Without rigorous schema alignment, even identical entities appear as unrelated records. This process often involves data standardization—parsing, cleaning, and normalizing values (dates, addresses, phone numbers) to eliminate superficial differences.

The strategic process of choosing specific record attributes to partition a dataset into mutually exclusive blocks, dramatically reducing the quadratic computational complexity of record linkage from O(n²) to near-linear. Effective blocking keys—such as zip code, birth year, or phonetic name encoding—group records that are likely to match. Poor key selection causes missed matches (records falling into different blocks). Advanced techniques include TF-IDF blocking for text-heavy data and Locality-Sensitive Hashing (LSH) for privacy-preserving scenarios.

The process of transforming raw data into a consistent, normalized format by parsing, cleaning, and reformatting values before matching. Common operations include: - Address normalization (123 Main St. → 123 MAIN STREET) - Date formatting (01/02/2025 → 2025-01-02) - Name parsing (Doe, John A. → JOHN DOE) - Phone number canonicalization Standardization directly improves linkage accuracy by eliminating superficial discrepancies that would otherwise cause false non-matches.