Master Data Management
AI-Powered MDM Glossary & Technical Reference
A glossary of key terms and acronyms for AI-powered and agentic master data management, covering golden records, entity resolution, matching, guardrails, and governance.
This glossary defines the core terms and acronyms used across AI-powered and agentic Master Data Management, from entity resolution and matching to agent orchestration, guardrails, and stewardship.
A
Agent Orchestrator
Coordination layer that sequences the work of multiple AI agents, passes context and results between them, and applies policy when deciding what happens next. In MDM, the orchestrator runs the Observe → Analyze → Plan → Execute → Verify loop and escalates uncertain decisions to a steward.
Agentic AI
AI that pursues a goal across multiple steps by choosing and sequencing actions rather than returning a single prediction. In MDM, agentic AI coordinates specialized agents for discovery, matching, resolution, governance, and publishing while keeping humans in control of thresholds and ambiguous merges.
AI Agent
Goal-directed software worker with a defined role, a bounded set of tools or data access, and explicit permissions. A Matching Agent, for example, can score duplicate candidates but should not merge records without the resolution and governance stages approving the action.
Anomaly Detection
AI technique that identifies unusual patterns or outliers in master data that deviate from established baselines, indicating potential quality issues or duplicate candidates.
Attribute Matching
Process of comparing specific fields (attributes) between records to determine similarity. For example, comparing customer names phonetically or addresses via distance algorithms.
Authority Source
The designated system of record for a particular attribute within the golden record. For example, the ERP might be the authority for a customer's financial rating, while the CRM is the authority for contact information.
C
Cleansing
The process of correcting or removing inaccurate, duplicate, or incomplete data to improve quality. Often performed before or during MDM implementation.
Conformance
Degree to which data adheres to defined formats, standards, and business rules. A customer record conforms if all required fields are populated and values fall within valid ranges.
Consolidation
Process of combining data from multiple sources into a unified master record, typically by identifying duplicates and merging their attributes.
D
Data Governance
Framework of policies, procedures, roles, and controls that ensure master data is created, maintained, and used correctly throughout its lifecycle.
Data Lineage
Complete tracking of master data from source systems through transformations to downstream consumption, showing which systems depend on which master records.
Data Steward
Individual or role responsible for maintaining quality and governance of a specific master data domain (e.g., customer data steward, product data steward).
Duplicate Detection
Specific process of identifying multiple records representing the same entity within or across systems. Machine learning accelerates this by scoring match probability.
E
Entity
A real-world business object that appears as data in multiple systems: a customer, product, supplier, location, or employee.
Entity Resolution
Process of determining whether records from different systems (or different records in the same system) refer to the same real-world entity and merging them into a single master record.
G
Golden Record
Single, authoritative master version of an entity combining all information about that entity from all source systems, with documented lineage for each attribute.
Graph Database
Database optimized for storing and querying relationships between entities. Often used in MDM implementations to store the complex linkages between master records and their source system lineage.
Guardrail
Constraint or check that bounds what an AI agent may do: allowed tools, data scope, confidence thresholds, approval requirements, or protected attributes. Guardrails make agent autonomy auditable and keep high-impact merges inside governance policy.
H
Human-in-the-Loop (HITL)
Governance pattern in which a person reviews or approves decisions the system is not confident about or that exceed a defined risk level. In agentic MDM, HITL usually applies to ambiguous matches, conflicting survivorship values, and merges affecting high-value accounts.
M
Master Data
Core business entities critical to operations: customers, products, suppliers, locations, employees, accounts. The foundational data that multiple systems depend on.
Master Data Management (MDM)
Discipline and set of technologies that create, maintain, and govern authoritative master records for core business entities across the enterprise.
Match Confidence Score
Probability rating (typically 0-100) assigned by machine learning algorithms indicating how confident the system is that two records refer to the same entity. Used to prioritize human review and determine which matches can be applied automatically.
Match Threshold
Score boundary that maps a match confidence to an action: auto-merge above the upper threshold, steward review inside the ambiguous band, and no merge below the lower threshold. Thresholds are business policy and should be documented, versioned, and reviewed as data and models change.
Merge Rule (Survivorship Rule)
Logic determining which source system's value is retained when merging duplicate records into a golden record. For example, 'use the CRM address if it's more recent than the ERP address.'
Model Drift
Gradual loss of matching accuracy as source systems, data patterns, or business definitions change. Drift is why agentic MDM needs continuous evaluation of precision, recall, steward overrides, and false merges rather than a one-time model launch.
P
Profiling
Process of analyzing source data to understand its characteristics, patterns, quality issues, and statistical distribution. Profiling data establishes baselines for both quality rules and matching algorithms.
Probabilistic Matching
Advanced matching technique using weighted algorithms and machine learning to calculate probability that two records represent the same entity based on multiple attributes.
S
Schema
Structure defining what fields a record contains and what data types those fields accept.
Single Source of Truth
Authoritative master data repository ensuring all systems reference the same consistent version of each entity.
Source System
System of record from which master data is originally captured (e.g., CRM, ERP, e-commerce platform).
Survivorship
See Merge Rule.
Key Acronyms
| Acronym | Meaning |
| AI | Artificial Intelligence |
| CRM | Customer Relationship Management system |
| ERP | Enterprise Resource Planning system |
| HITL | Human-in-the-Loop |
| LLM | Large Language Model |
| MDM | Master Data Management |
| ML | Machine Learning |
| ETL | Extract, Transform, Load |
| SKU | Stock Keeping Unit (unique product identifier) |
Related Reading
See Master Data Management in 4DAlert
Explore how 4DAlert implements the concepts in this guide as a working platform.