Master Data Management (MDM), or a data repository, stores, manages, and distributes reference data (such as third parties, products, organisations, nomenclatures, employees, etc.) within an organization.
An MDM centralizes key data for a given scope in a single master location - aka "Single Source of Truth"- to facilitate sharing, ensure quality, access security, and data governance, among other benefits.
The increasing needs for information system urbanisation have accelerated their expansion—you daily benefit from their utility without realizing it. To better grasp this concept, I'll draw on examples and lessons from real-world projects. Then, I will cover recurring challenges that cause MDM implementations to fail and how to avoid them.
Examples of Data Managed via MDM
At the international level, the 249 country codes and their subdivisions are (…)
We can distinguish two main categories: Master Data (business data) and Reference Data, often called static tables due to their rare updates (e.g., ISO country codes, currency codes).
On an international scale : REX 1: In an industrial group operating in 40 countries, a multi-domain repository linked projects, third parties (clients/suppliers/others), vessels, legal entities, etc. The third-party repository integrated Dun & Bradstreet's external data (a company business data provider), adding strategic financial insights.
At national level: At company level:
REX 2: A doctors' NGO manages a catalog backed by a MDM: 37k field-used items (medical and non-medical) across 12 categories (nutrition, kits, meds, transport...). Each item has structured fields such as description, usage tips, storage..., easing searches, updates, translations. The online catalog, used by other NGOs, gets 3M+ annual views.
REX 4: International pharma company ditched scattered Excel files across departments (regulatory, supply chain, finance, procurement...) for a multi-domain repository. Example of impact analysis (in 2 seconds): List Japan-based suppliers and their roles (manufacturing, packaging, QA...) in logistics chains with related products
REX 5: City hall of Paris: A 2008 audit criticized poor real estate asset management. The Architecture Dept. launched four parallel projects (asset inventory/MDM app/target processes/change mgmt. for 1200 civil employee). Sample of query: In 2 seconds, list by district, all the buildings with at least 2000m² usable area under emphyteutic lease or public service delegation contract. Result: Palais Brongniart run by GL Events.
In public services, audits from accounting firms or oversight bodies highlight mismanagement of assets or key data.
The first MDM domain installed in the companies In European and American firms, (…)
Is a CRM (Salesforce, SugarCRM...) a MDM?
A CRM becomes a MDM only if it holds all necessary customer reference data—not just sales-related—and fits robust data governance with clear quality and access rules. CRMs often lack built-in handling for unification, normalization, deduplication, requiring dedicated MDM tools.
PIM vs. MDM differences
| Attribute | Managed in PIM | Managed in MDM |
|---|---|---|
| Product marketing descriptions | Yes Update of product sheets |
No or limited |
| Multilingual content |
Yes Translation management |
Context specific |
| Images, videos, marketing materials | Yes Often integrated with DAM (Digital Asset Management) | Rarely or via external integration |
| Adaptation to sales channels |
Yes Personnalization according to the channel (web, paper, marketplace) |
No |
| Product Background Information |
Yes Promotional data, seasonality, specific use |
No |
| Product variant mngt |
Yes Variants, detailed options |
Often limited |
| Marketing business collaboration |
Yes Collaborative enrichment workflows |
Context specific |
| Variable data (contextual) |
Yes Very detailed data according to target |
Generally "cleaned" and homogeneous data |
| Export media for specific channels |
Yes Specific formats: catalogues, e-commerce |
No |
In-house solution vs. market solution
REX 2: A water treatment/distribution firm assessed in-house vs. MDM market solution costs. IT roadmap targeted 4 domains with validation workflows and external source automation. Projections showed that cumulative intenal development and maintenance costs exceeding the cost of implementing a market solution from the second domain onward.
Deploying and maintaining a high-quality MDM requires co-building six key pillars (scope, processes, CRUD matrix, governance, etc.) aligned to organizational strategy. When solid, benefits emerge fast, positioning MDM as urbanization enablers; in contrast, weak pillars risk failure.
MDM rollout mirrors multi-domain ERP: both are cross-functional, disrupt processes, demand business/manager/executive buy-in, and force workstyle changes. MDM implementation often precede the ERP ones.
Critical phases: Co-designing target processes for data enrichment without burdening business (complexity, delays, tasks)—method matters as much as tech (auto-complete, post-entry validation, intuitive UX). Data retrieval/preparation for MDM injection can be a separate and critical project.
REX: On a framing concerning natural gas network operator's equipment MDM, we arranged a meeting with the Paris City's team that successful installed a real estate MDM. Expert-novice exchanges clarified unexpected key phase considerations in similar public contexts.
Example 1: Sports Vacation NGO
Context: Stages for teens/young adults; send travel docs, emails, invoices pre-departure. France: for 18-25 year old people, the address change rate is ~28-30% yearly. Third parties evolve and can cumulate attributes : prospect, member, parent, volunteer, staff. Track active periods (start/end dates) are required to enable targeted campaigns and stats:
Data quality uses at least six criteria: completeness, accuracy, consistency, uniqueness, timeliness, validity (depict the reality). Measure in repositories or other sources (SharePoint, Excel, CRM/ERP, externals).
The level of data quality can be measured in your MDM or in other sources used by your organisation (SharePoint, Excel, CRM, ERP, external databases, etc.).
Specif tools are available: Talend, Soda, IBM InfoSphere, Datagalaxy for dashboards/alerts. MDM software also includes quality, usage and use level reports.
Data quality reports has be shared as part of the data policy in order to highlight MDM value. The utility level increases with the data quality and interaction ease: search, filter, export, update, customize, interoperate. In multinationals, one MDM feeds dozens of apps as urbanization base —nicknames evoke this: golden records, Single Source of Truth (SSOT), etc.
A poor data quality drives shadow IT; referential data often duplicates across 10+ scattered bases.
The major failure of the National Payroll Operator (ONP) project in 2014 clearly illustrates the difficulties involved in implementing a unified MDM. Launched with great ambition, this project aimed to centralise the payroll of 2.7 million French civil servants via a single information system synchronised with eight different referential systems. Among the main causes of this failure, the financial auditors highlights the ignorance or serious underestimation of technical risks, in particular the automated management of common referential systems, which had not been anticipated before the contract was signed. In addition, the fragmented governance between ministries and the ONP complicated the coordination necessary for the project's success, leading to the programme's abandonment after a significant financial loss: €346 million!
The scoping phase, which did not have any AI tools at its disposal, did not sufficiently measure and anticipate:
Example 2: A good MDM = A comprehensive inventory is sufficient!
One of the common pitfalls is to put all your efforts into the inventory phase without worrying about maintaining its quality. Remember the inventories of supermarkets that closed their doors annually to count all their products on the shelves and in their stocks. However, if there is no formalised management of incoming and outgoing flows, an inventory remains accurate for only one day and ceases to be accurate the moment a customer, supplier or employee interacts with the products.
For example, a national operator undertook an exhaustive inventory of technical assets located throughout the country in order to create a single national reference database, spending months of costly referencing work. Unfortunately, strong convictions compensated for weak change management. As a result, the target update processes had not been tested before this meticulous and costly inventory was carried out. After the inventory, knowledgeable stakeholders rejected the application of the target processes and retained their habits and their own regional databases. Inevitably, the reliability and usefulness of the repository deteriorated day by day...
Find sources and additional information related to this article