MDM
Also: MDM, Master Data Management, Reference data management, Golden record management, Gestion des donnees de reference (GDR)
Master Data Management: ensuring a single, authoritative version of key reference data (customers, products, suppliers) across all systems.
What it is
Master Data Management (MDM) is the discipline, set of processes, and technology used to create and maintain a single, trusted, and consistent version of an organization's core reference data. This master data describes the key entities a business depends on: customers, products, suppliers, employees, accounts, and locations. Unlike transactional data (an order, a payment, a click), master data changes slowly and is reused across many systems.
MDM combines governance (who owns and approves data), matching and merging logic (deduplication), and a distribution mechanism so that every downstream system references the same authoritative record, often called the golden record.
Why it matters
Without MDM, the same customer can exist as "Acme Corp", "ACME Corporation", and "Acme Inc." across CRM, billing, and support tools. This fragmentation causes:
- Bad decisions from inconsistent or duplicated reporting figures
- Wasted spend (duplicate mailings, redundant supplier onboarding)
- Compliance risk (inability to honor a data deletion request across systems)
- Failed AI initiatives, since models trained on conflicting entities produce unreliable output
How it is used in practice
Typical MDM implementations follow a few common styles:
- Registry style: leaves data in source systems and only maps identifiers to a central index.
- Consolidation style: pulls copies into a hub for cleansing and reporting.
- Centralized (or transactional) style: the hub becomes the authoritative source that systems write back to.
Core capabilities include data modeling, matching and survivorship rules (deciding which value wins when records conflict), stewardship workflows for manual review, and syndication back to consuming applications.
Worked example
A retailer sells through an online store, physical shops, and a call center. A customer, Maria Lopez, appears three times:
- Web: `maria.lopez@email.com`, no phone
- Store loyalty: `M. Lopez`, phone `+33 6 12 34 56 78`
- Call center: `Maria L.`, same phone
The MDM hub runs matching rules on email, phone, and name similarity, links the three records, and applies survivorship rules (most recent verified email wins, most complete address wins). It produces one golden record with a stable ID. Marketing now sees one customer instead of three, finance reconciles lifetime value correctly, and any AI recommendation engine trains on clean, unified history.
Bottom line: MDM is the foundation that makes analytics, personalization, and automation trustworthy.
See also
Frequently asked questions
What does MDM stand for, and what does it actually manage?
MDM stands for Master Data Management. It covers the processes, governance rules, and technology used to maintain one trusted version of an organization's core reference data: customers, products, suppliers, employees, accounts, and locations. It does not manage transactional data such as an order, a payment, or a click, which changes constantly and is not reused as a shared reference.
What is the difference between master data and a golden record?
Master data is the category of slowly changing reference data (customers, products, suppliers); the golden record is the single authoritative version of one specific entity within that category. The golden record is produced by matching duplicate records across systems and applying survivorship rules that decide which value wins when two sources disagree, for example the most recently verified email or the most complete address.
Why do AI projects fail without master data management?
Because a model trained on conflicting versions of the same entity produces unreliable output. If "Acme Corp", "ACME Corporation", and "Acme Inc." are treated as three separate customers, a recommendation engine or a lifetime value model learns from fragmented history. MDM unifies those records first, which is why it is treated as a prerequisite rather than a parallel project.
Which MDM implementation style should I start with?
Three styles exist and they differ by how much authority the hub takes. Registry style leaves data in the source systems and only maps identifiers into a central index, so it is the lightest to deploy. Consolidation style copies records into a hub for cleansing and reporting. Centralized (or transactional) style makes the hub the authoritative source that applications write back to, which delivers the most control but requires the deepest integration work.
What are survivorship rules and how do they work on a real duplicate?
Survivorship rules decide which value is kept when linked records disagree on the same field. Take a retailer where Maria Lopez appears as "maria.lopez@email.com" with no phone on the web, "M. Lopez" with a phone in the store loyalty program, and "Maria L." with the same phone in the call center. Matching rules on email, phone, and name similarity link the three; survivorship rules then keep the most recent verified email and the most complete address, producing one golden record with a stable ID.