MDM Implementation Styles
There are four MDM implementation styles: Consolidation, Registry, Centralized, and Co-existence. The style that best suits your needs depends on the domains you want to involve in your MDM system.
Consolidation
The consolidation approach to MDM involves copying data from source applications into a central hub, where it is matched and merged. The hub creates golden records, which it makes available for distribution to downstream applications or direct consumption by business users and data stewards.
This approach provides added capability in the form of stewardship, which gives data stewards the ability to make data entries, deal with duplicates, and manage rejects. This is a great way to take centrally stored master data and use it for analysis and reporting.
Registry
The registry style of MDM is focused on identifying duplicates within data sets pulled from multiple source systems. The system cleans and standardizes the data, attaching a unique identifier to duplicate records. The MDM hub stores an index of this source data, keeping track of cross-references between matching source data.
Registry-style MDM is cost effective and doesn’t require much intrusion into source systems. However, it has higher latency and more limited capabilities than other MDM styles.
Centralized
The centralized, or transactional, MDM style guarantees the highest-quality data, but it’s also the most intrusive and time-intensive. In this type of system, all systems must constantly connect to the data hub for any updates to their master data, which can disrupt existing business and technical processes.
Coexistence
The coexistence style of MDM is the gold standard for large-scale data distribution models. It creates a consolidated data hub that then feeds updated records back to sources. This provides real-time updates in both the master data and the source systems.
The most significant benefit of this style is that it allows data creation on multiple systems. However, it is complicated to deploy and keep secure, and it requires constant data cleaning.