Lesson 4 — The MDM Processing Spine
A strong MDM program does not begin by matching every record it can find. Chapter 10 presents a sequence that moves from shared meaning to source intake, preparation, identity, and controlled sharing.
Five key processing steps
1. Data Model Management
2. Data Acquisition
3. Data Validation / Standardization / Enrichment
4. Entity Resolution + Identifier Management
5. Data Sharing + Stewardship
Memorize the order, but more importantly understand why each step exists.
1. Data Model Management — define enterprise meaning
Source systems often speak in local semantics. Account, Customer, Party, or Location may mean different things in different applications. Attribute granularity can differ too: one system stores a single Name field while another stores first/middle/last components.
MDM needs an enterprise logical/canonical understanding of the entity and attributes rather than simply adopting one source schema as the enterprise truth.
The model answers: - What entity are we mastering? - Which attributes belong to it? - What do those attributes mean? - At what granularity? - Which relationships matter?
A common failure is to begin matching before these semantics are aligned.
2. Data Acquisition — bring candidates into the process
Acquisition obtains candidate entity records from relevant sources. The program should know: - which sources are authoritative for which attributes; - source quality; - identifiers; - update frequency; - lineage and transformation; - usage and dependencies.
Acquisition is not the same as modeling or matching. It supplies the data that later steps operate on.
3. Validation / Standardization / Enrichment — make evidence comparable and useful
Validation
Check values against rules and expectations.
Standardization
Normalize representation of what you already have.
Example:
- 9015551212
- (901) 555-1212
- +1 901 555 1212
Standardization can put these in one comparable form before matching.
Enrichment
Add trusted supplementary information.
Example: add validated latitude/longitude or a standardized postal attribute from a trusted service.
Standardization = make what you have comparable.
Enrichment = make the record fuller with trusted additions.
Do not jump to algorithm tuning when the real problem is poor input preparation.
4. Entity Resolution + Identifier Management
Only after adequate modeling and preparation should the process decide whether records represent the same entity. Then it manages Global IDs, source IDs, mappings, and history.
This is where false positive/negative risk and reversible match workflows become critical.
5. Data Sharing + Stewardship
The mastered result must actually be used. Sharing publishes governed data/services. Stewardship handles uncertain cases, exceptions, corrections, feedback, and recurring root causes.
A technically brilliant match engine that no applications consume is not successful MDM.
Why this is a sequence
Consider two records:
- ROBERT JONES, 901-555-1212
- Bob Jones, (901)5551212
If the enterprise has not agreed what constitutes a Party, which identifiers matter, or how phone values should be standardized, the match decision is weaker. The processing spine reduces avoidable uncertainty before the identity decision.
Exam trap
“Matching quality is poor” does not automatically mean “switch to probabilistic matching.” First inspect modeling, source quality, validation, and standardization.
Stop and check
A stem says customer phone formats differ and duplicates are being missed. Best first response: standardize/validate the inputs before changing the matching method.
Source anchor: pp. 340–347.