A company where the same business question receives different answers depending on which system or area was consulted has a data problem, not a communication problem. Monthly revenue calculated by sales, finance and the ERP system rarely coincides when there is no active Master Data Management (MDM). The alignment meeting to discover which number is correct is a recurring, invisible operational cost that also delays the decision itself.
Subcapability 04 of 05 · Data & Analytics
Data Quality & Master Data
Six data quality dimensions and MDM Hub Architecture that eliminate the three versions of the same customer across systems and transform data from a source of debate into a verifiable base for every executive decision.
What is at stake
The CRM system lists 45,000 customers. The ERP lists 38,000. The BI platform lists 42,000. Which number is real? That question consumes management capacity every week and has no answer while three versions of the same customer coexist without a unification process with a formal owner.
What it is, in practice
How we work
MDM Hub with golden record
We implement Master Data Management with Hub Architecture that establishes the authoritative version of each critical entity: customer, product, supplier and financial account. When two systems diverge about the same customer, the survivorship process defines which attribute prevails with an explicit criterion.
Six measurable quality dimensions
We structure a Data Quality Score per critical dataset with six dimensions: accuracy (correct data), completeness (data present), consistency (same data across different systems), timeliness (current data), validity (data within the expected domain) and uniqueness (no duplicate). Each dimension has a defined threshold and continuous monitoring.
Automatic validation in the pipeline
We instrument every pipeline with quality tests via Great Expectations or dbt tests. Data that violates a quality dimension blocks the pipeline before contaminating the consumption layer. Quality shifts from periodic manual review to continuous automatic gate.
Data observability with proactive alerts
We monitor the statistical distribution of fields, null rates, record volumes and schema drift in real time. When the pattern changes anomalously, the alert reaches the engineer before the analyst discovers the problem in the dashboard.
Data profiling before any migration
We map the real quality state of each dataset before any cleansing, migration or AI initiative. Profiling reveals where duplicates, nulls, inconsistent formats and out-of-domain fields exist that will block the project if not addressed first.
Measurable gains
What changes in the result when this subcapability matures.
Percentage of duplicates in critical entities (customer, product, supplier)
MDM with an active deduplication process reduces duplicates in key entities. Each eliminated duplicate improves report accuracy, reduces the cost of redundant communication and increases the confidence of the teams that use the data to decide.
Average Data Quality Score per dataset feeding executive decisions
A score measured by dimension and by dataset makes data quality comparable across quarters and domains. Teams that measure Data Quality Score start investing where the score is lowest, with criteria, rather than treating quality as an abstract problem.
Time between data degradation detection and correction with zero consumer impact
Data observability with proactive alerts and an automated correction pipeline reduces the interval between the moment data starts to degrade and the moment the consumer is affected. Without observability, that interval is measured in days or weeks.
AI-ready data: percentage of datasets with validated quality for model use
An AI model trained on low-quality data produces low-quality predictions. AI-ready data has the six quality dimensions above the minimum threshold, traceable lineage and formal ownership. Data maturity is the most reliable predictor of AI project success.
Frequently asked questions
What is MDM and why do data unification projects usually fail?
MDM, Master Data Management, is the discipline that establishes the authoritative version of critical business entities with a governance process to maintain that version over time. Unification projects fail because they treat MDM as an IT project with a delivery date, not as a continuous discipline with a business owner. When the project ends without a formal owner responsible for maintaining the golden record, data diverges again within six months.
What are the six data quality dimensions?
Accuracy verifies that data reflects business reality. Completeness verifies that required fields are present. Consistency verifies that the same data has the same value across different systems. Timeliness verifies that data is updated within the timeframe needed for the decision. Validity verifies that data is within the accepted domain of values. Uniqueness verifies that there are no duplicates of the same entity. Each dimension has a measurable KPI and an operational threshold.
What is data observability and how does it differ from data quality?
Data quality measures the six dimensions of data at a point in time. Data observability monitors data behavior continuously over time, detecting anomalous changes in distribution, volume and schema before they reach the consumer. The two are complementary: quality defines the expected standard, observability detects when the standard changes.
Does low-quality data really impact AI results?
Yes. An AI model learns patterns from training data. Data with untreated null fields, entity duplicates, inconsistency across systems and out-of-domain values produces patterns that do not reflect business reality. The model learns noise along with signal. In production, the incorrect prediction or classification appears with the same confidence as the correct one, making the error difficult to detect.
How to prioritize which datasets to address first?
Priority follows the financial impact of the decision the dataset feeds. Datasets that feed revenue decisions, such as pricing, demand forecasting and credit scoring, have the highest direct impact. Within those datasets, priority goes to the quality dimension with the largest gap relative to the minimum threshold for the use case. Data profiling makes this mapping before any treatment initiative.
Other subcapabilities in this capability
Data Architecture & Governance
Data Management Body of Knowledge and Data Mesh with federated governance that structure data as a formal asset with owner, traceable lineage and domain accountability that scales without a central bottleneck and enables AI in production.
Data Engineering & Pipelines
Apache stack with orchestration, observability and idempotency that eliminates the artisanal pipeline without monitoring and ensures no executive dashboard ever shows a wrong number with the appearance of a correct one.
Analytics & Business Intelligence
Semantic layer, governed self-service and KPIs wired to the result that replace expensive intuition with evidence-based decisions and eliminate the debate about which number is right in every executive meeting.
Data Products
DATSIS principles and Data Contracts that transform ownerless datasets into products with SLA, defined consumers and explicit accountability, eliminating the central bottleneck no backlog can absorb.
Want clarity on where to invest first?
A complete technology capability assessment with an evolution roadmap connected to financial result.