MDM Implementation Styles

Three factors help to broadly categorize which implementation style will be appropriate for a business.

1. Cost — Cost is typically the first and last consideration in choosing an implementation style. Registry models tend to be less complex, or intensive, therefore less expense. From there, the models can be upgraded to other models appropriate for needs. A Registry model can be upgraded to a Consolidation model, for example, with all the accompanying stewardship features added with greater data scrutiny. Costs are important, but the correct style is measured by business goals and needs.
2. Data Sources — The number of data sources is another concern. Businesses with large numbers of data sources, for example, can benefit from the Registry model. Multiple sources present data accuracy challenges, but by fetching data from all sources and depositing them in the MDM hub, then applying cleansing and matching algorithms, these systems can reconcile duplicates, and form a single-source of truth, or “golden record”.
3. Data Scrutiny — Based on the two factors above, Registry implementation style seems to win out, it is both the cheapest, and can easily form a golden record from multiple data sources. However, the Registry’s golden record is not 100% accurate, it’s an approximation made by computer algorithm. If greater data scrutiny is required (see Match and Merge Concepts section), then a human eye can review data and determine accuracy more closely. These systems use additional stewardship features, and therefore cost more.

When data is fetched from multiple sources, to create a golden record it must be matched and then merged together with similar records into a single final record in the MDM’s golden record. The following illustration shows that a match and merge algorithm will determine similarities between data entries based on fields. Some attributes, like DOBs and SSNs, are more significant for identification and therefore weighted heavier as identifiers (names are less of an identifier because of their shared commonality). A match score is used to determine the credible uniqueness of similar entries. If an entry is inadequately matched but seems as a likely fit, it can be relegated to data stewards for further inspection.

Beginning with a Consolidated model, Coexistence MDM implementation styles build loopback features into the MDM system that updates source systems with Master Data records. This results in a Master Data record in both the hub and the upstream data sources. This helps authors retain the most updated information at the sources. This configuration demands that sources have cleansing features to ensure data integrity.

Change management is an important concept in MDM implementation because taking a master data approach to an organization’s data assets not only involves technology and process, but chiefly the people who will be implementing and using master data. Staff need to take a data-first approach to operations, meaning they rely on data, make data-driven decisions, and overall seek to derive value from the data produced by the company about itself. Depending on their familiarity and comfort with using data, high or low, it can foreshadow the change journey ahead.

Change management is a structured process for implementing new operations within organizations. The discipline is so important that it continues to attract many hours of professional study, but central to every school of thought on change management is “inclusion”. Without involving the majority, if not entirety of stakeholders, change management bogs down, invites confusion, and becomes ineffective.

One simple example of why this occurs when implementing new technology is that often new technology becomes the focus because it seemingly delivers value to data processes, and ignores the fact that people will remain the users, stewards, and support staff that will maintain the data processes. In short, the people are the purpose of the data, and the driving force of it, technology helps with this but is a tool.