Master data management gives the business one trusted version of its core entities, such as customers, products, suppliers, and locations. It matches records that describe the same real-world thing, decides which value wins when sources disagree, and shares the result with every system that needs it.
What counts as master data
Master data describes the main things a business deals with and changes slowly compared with transactions. A customer or product is master data; an individual sale is a transaction. Gartner defines master data management as a technology-enabled discipline in which business and IT work together to ensure the uniformity, accuracy, stewardship, semantic consistency, and accountability of the enterprise official shared master data assets. Reference data, such as country codes, is related but managed separately.
Four common styles
| Style | How it works | Best when |
|---|---|---|
| Registry | Keeps an index that links records across systems; data stays in place | You need a fast, low-disruption view and sources cannot change |
| Consolidation | Copies and merges data into a hub for reporting; sources are not updated | The main need is trusted analytics |
| Coexistence | Hub holds the golden record and synchronizes changes both ways | Several systems update the same entity |
| Centralized | All creation and change happen in the hub, then flow out | You can enforce a single point of entry |
From several records to one
The core mechanics
- Standardize. Clean formats first: casing, abbreviations such as St and Street, and phone and address formats.
- Match. Use deterministic rules (same national ID or email) and probabilistic scoring (similar name plus same address) with thresholds for automatic match, manual review, and no match.
- Merge with survivorship. For each field, define which source wins: most trusted source, most recent update, most complete value, or a steward decision.
- Publish. Send the golden record to consuming systems, and keep a cross-reference of the original source identifiers so changes can be traced.
- Steward. Review uncertain matches and exceptions in a work queue, and feed corrections back into the rules.
A worked example
The three records in the figure are scored: names are very similar, the address matches after standardizing St to Street, and the third record shares the first name and surname. Together they pass the automatic-match threshold. Survivorship uses the most trusted source for the address (Source B, which has a verified street format), the most complete value for the name, and the only available value for the email. One customer is now counted once, and a marketing campaign no longer mails the same person three times.
Starting small
Choose one domain with a clear cost of duplication, usually customer or supplier. Define the data elements, agree the matching and survivorship rules with business stewards, run a pilot on a subset of records, and measure the duplicate rate before and after. Expand only when the first domain is stable.
Common pitfalls
- Buying a tool before deciding which style and domain you need.
- Skipping standardization, which lowers match accuracy.
- Ignoring ownership: someone must be accountable for each master domain.
- Merging too aggressively. Prefer manual review for uncertain matches to avoid wrongly combining two different people.
Implementation checklist
- First domain chosen with an accountable owner
- Sources and key attributes inventoried
- Standardization rules defined and tested
- Match thresholds and survivorship rules agreed
- Steward work queue and review process set up
- Cross-reference to source identifiers maintained
- Duplicate rate and match accuracy tracked
The hardest part of master data management is agreement, not technology. Spend time getting business stewards to agree which source is trusted for each attribute, and the rules become straightforward to build.
Interactive: from three records to a golden record
Step through standardizing, matching, and merging. Names and addresses are fictional.
Source A
Ahmed Ali
12 Palm St
-
Source B
A. Ali
12 Palm Street
-
Source C
Ahmed Ali
-
ahmed@example.com
Step 1 of 4
Source records as received. They look different but may describe one person.
References and further reading
- Master Data Management (MDM) glossary definition (Gartner)Accessed November 7, 2026
- DAMA-DMBOK: Data Management Body of Knowledge (DAMA International)Accessed November 7, 2026