Master data management

Master data management is the set of practices for maintaining consistent descriptions of core entities such as organizations, products and locations. It addresses identifiers, ownership, quality rules and distribution of approved records across systems. Its scope is organizational as well as technical: different departments may use the same term at different levels of granularity, so a shared record requires explicit definitions and responsibilities.

In practice, an organization identifies source systems, defines canonical entities and reconciles duplicates and conflicting attributes. Policies govern who can change records, how evidence is reviewed and how updates reach consumers. A golden record can be one output, but some use cases need multiple context-specific views rather than one universal value for every field.

Defensibility requires knowing which system asserted an attribute, why the shared view uses it and which consumers saw which version. A single source of truth slogan is insufficient when source authority differs by property. Entity resolution establishes identity; survivorship rules select attributes; provenance and audit trails make the resulting decisions inspectable.

The failures of master data programmes are mostly failures of decision rights. Someone must own each entity type and each attribute, with authority to resolve disputes between systems. Rules for survivorship should be written, versioned and approved, not embedded in a configuration nobody reads. A change to a rule should be tested against a sample and its effect on existing records reported before it is applied. A data steward needs a queue of cases that is short enough to clear and a record of each decision. A correction in a source needs a defined path to the master and from the master to consumers.

Measure the programme by outcomes. Useful measures include the proportion of records matched without review, the error rate found in sampled reviews, the time from source change to consumer visibility, and the number of conflicts open beyond a stated age. They tell a reviewer more than counts of records loaded. The article on auditing the multiple-sources-of-truth assumption offers a way to test whether the organisation really has one authoritative account for a given fact.

Master data management relies on entity resolution, produces golden records under survivorship rules, and stands next to the idea of a system of record. Its audit needs an audit trail. It is a governance practice with technical supports and not a product, and the best technical design cannot compensate for the absence of agreed ownership.