Master Data Management: Golden Records and Entity Resolution

What Is Master Data?

Master data is the core business entities shared across an organization: customers, products, suppliers, employees, locations. It is the data that every system references but no single system owns.

When master data is inconsistent, problems cascade:

The same customer appears three times with different addresses
Product hierarchies conflict between ERP and e-commerce
Financial reporting cannot reconcile across business units

MDM Core Concepts

Concept	Definition
Golden Record	The single, authoritative version of a master entity
Entity Resolution	Matching and merging records that represent the same real-world entity
Data Deduplication	Removing redundant records while preserving data completeness
Survivorship Rules	Logic that determines which field value "wins" when merging records
Hierarchy Management	Maintaining parent-child relationships (product categories, org charts)

MDM Implementation Styles

Style	Description	Best For	Complexity
Registry	MDM stores links to source records but doesn't hold master data	Low disruption, quick wins	Low
Consolidation	MDM creates golden records from sources, read-only hub	Analytics and reporting	Medium
Coexistence	MDM creates golden records and syncs back to source systems	Operational consistency	High
Transaction	MDM is the system of record; all changes go through it	Maximum control	Very high

Most organizations start with consolidation and evolve toward coexistence as maturity grows.

Entity Resolution: The Hard Problem

Entity resolution determines whether two records represent the same entity. This is harder than it sounds:

"John Smith" at "123 Main St" and "J. Smith" at "123 Main Street" -- same person?
"Acme Corp" and "ACME Corporation" and "Acme Co., Ltd." -- same company?

Approaches:

Deterministic matching: exact match on defined keys (email, tax ID)
Probabilistic matching: scoring based on multiple fuzzy attributes
ML-based matching: trained models that learn matching patterns

Key decisions:

Match threshold: too low = false positives (over-merging), too high = false negatives (under-merging)
Human review workflow for uncertain matches
Ongoing matching (new records must be matched continuously)

When MDM Pays Off

M&A integration: merging customer bases from acquired companies
Regulatory compliance: single customer view for KYC/AML
Customer experience: consistent profile across channels
Financial consolidation: single product and customer hierarchy for reporting
Supply chain optimization: unified supplier and product master

When MDM Struggles

Organizations without executive sponsorship (MDM is political, not just technical)
Companies that underestimate data stewardship effort
Projects that try to boil the ocean (starting with all entities at once)
Implementations that ignore source system data entry quality

Common Failure Patterns

Pattern	Why It Fails
Big bang approach	Trying to master all entities at once overwhelms the team
Technology-first	Buying a tool without defining governance processes
No stewardship	Golden records degrade without ongoing human review
Ignoring sources	Fixing master data without fixing data entry at source is futile
Over-engineering match rules	Diminishing returns on match precision past a threshold

Implementation Roadmap

Pick one entity domain: start with customer or product, not everything
Profile the data: understand duplication rates, completeness, consistency
Define survivorship rules: which source wins for which attribute
Implement matching: start deterministic, add probabilistic as needed
Establish stewardship: assign data stewards for exception handling
Measure and iterate: track duplicate rate, golden record coverage, steward throughput