Definition of master data management
Master data management, abbreviated to MDM (gestion des données de référence in French), refers to the set of rules, processes and tools by which an organisation maintains a single, shared version of the entities that several of its applications use: customers, products, suppliers, sites, employees.
This definition describes common usage: its wording varies from one book or vendor to another. The DAMA-DMBOK, a reference framework published by DAMA International, presents itself as a common language for data management professionals. Its publisher specifies that it is neither a prescriptive standard nor a regulatory text.
One entity, several records: the problem MDM addresses
In an organisation, the same customer may be created in the sales tool, in invoicing and in the after-sales service application. Each application assigns it its own identifier, spelling and address. As long as these records are not linked, simple questions remain without a reliable answer: how many customers, what revenue for this one, which contracts are under way with this supplier.
Master data management addresses this problem of sharing between systems. It does not concern all data. It targets data that describes common entities and changes little. The facts attached to them (orders, payments, service calls) remain in the applications. Nor does a defect arising from uncontrolled entry in a single application fall under it: it is dealt with at source, as explained in our article on the levers to improve data quality.
From matching records to the golden record
An MDM system chains five mechanisms.
- The entity model sets the attributes retained, their definition and the common identifier.
- Matching identifies the records that designate the same entity, through equality of a reliable identifier or approximate comparison of names and addresses.
- Survivorship rules state, attribute by attribute, which source prevails in the event of a conflict. Applying them produces the golden record.
- Stewardship entrusts a designated person with the cases the rules do not settle.
- Distribution returns the golden record and its identifier to the applications and analytics tools.
The system is organised by domain (customer, product, supplier) or covers several domains at once.
MDM, reference data, PIM, catalogue: not to be confused
| Notion | What it manages | What separates it from MDM |
|---|---|---|
| Reference data | Code lists: countries, currencies, units | English distinguishes it from master data; French usage may group both under “données de référence” |
| PIM (product information management) | The descriptive content of products for sales channels | MDM holds the product’s identity and the attributes shared between applications |
| Data catalogue | The description of datasets: content, owner, classification | The catalogue holds metadata, MDM holds the data itself |
| Data warehouse | The history of facts, consolidated for analysis | Matching done in the warehouse does not flow back to the applications |
A concrete example: the business customer with three records
Take an equipment manufacturer where the same business customer appears in the sales tool, in invoicing and in the after-sales service application, under three slightly different company names. Matching relies on the company’s SIREN number (the French company identifier) when it is filled in, then on comparing name and address. The survivorship rules retain the company name from invoicing and the contact from the sales tool.
Two records carry the same SIREN and different addresses. A duplicate, or two establishments of the same company? The rule does not settle it, and the case goes to the data steward. The golden record then receives a common identifier, sent back to the three applications. Revenue and service call history can then be read on the same entity.
Implementation styles, from a simple index to a central repository, and the running of such an initiative are presented on our page on data quality and MDM. Designating owners and stewards falls under data governance.