Data Quality, Governance & Master Data Management

Quality data: fit for their intended use

Quality data are "fit for their intended uses in operations, decision making, and planning." The exam frames quality along eight characteristics:

1. Uniqueness — each instance exists no more than once (a key identifies each instance).
2. Accuracy — the datum correctly represents the real-world object (data can be *valid* yet *inaccurate*).
3. Consistency — values agree with related data across rows, datasets, and time.
4. Completeness — required values are present (NOT NULL plus business rules about what's required).
5. Timeliness — data are available *when expected*.
6. Currency — data are *recent enough* to be useful.
7. Conformance — data are stored/presented in the format their metadata specifies.
8. Referential integrity — references satisfy relationship rules (foreign keys match existing primary keys).

The key insight is that quality is *multidimensional*: data can be accurate but not timely (a correct address from five years ago), or consistent but inaccurate (the same wrong value copied everywhere).

The six-step data-quality improvement program (this is Part B Q8b)

1. Get business buy-in       frame quality as a business imperative; quantify the cost
                             of bad data and compute ROI
2. Conduct a data-quality    statistically profile files, check against business rules,
   audit                     review data-entry/maintenance controls
3. Establish data            assign accountability for each subject area to named data
   stewardship               stewards with clear roles
4. Improve data capture      automate entry, use drop-downs/preset options, trigger
   processes                 immediate validation feedback
5. Apply modern data-        use proven data modeling + ETL scrubbing tools
   management technology      (pattern matching, fuzzy logic) + sound DB design
6. Apply TQM principles      treat quality as an ongoing process — prevent defects, don't
                             just correct them; measure and improve continuously

The best place to *enforce* quality is in the database definitions (constraints), not in application code — defect *prevention* beats defect *correction*.

Governance, stewardship and the CDO

Data governance is the set of high-level organizational groups and processes that oversee data stewardship — it provides the *authority* to enforce standards and guides data-quality, architecture, integration, MDM, and warehousing efforts. It needs both top-down commitment (executive sponsorship) and bottom-up participation (data stewards). A data steward is accountable for the quality of data in a specific subject area but does *not own* the data. A chief data officer (CDO) is the C-suite executive accountable for all data activities — signaling that data is a strategic asset.

Master Data Management: the single source of truth

Master Data Management (MDM) ensures the currency, meaning, and quality of *reference data* (master data) — the common shared entities nearly every application references: customers, products, employees, locations. When five systems hold five inconsistent versions of "customer," any decision is incomplete. MDM determines the "golden record" for each entity so all applications reference one authoritative version. Its three architectures parallel the three Chapter 9 integration approaches:

- Identity registry (≈ *federation*) — master data stay in their source systems; a registry knows which system holds the authoritative value for each attribute. - Integration hub (≈ *propagation*) — changes are broadcast (often asynchronously) to all subscribing databases. - Persistent / consolidated (≈ *consolidation*) — one consolidated golden record that all applications draw from; the most work-intensive but the clearest single source of truth.

MDM is *not* a data warehouse (no history, no transactions) and *not* an ODS (not an operational store) — it is infrastructure for consistency. This ties the whole course together: the integrity rules from Module 4 and the integration approaches from Module 9, lifted to the enterprise.