Data quality dimensions are standardized criteria – such as accuracy, completeness, consistency, and timeliness – used to evaluate whether data is fit for its intended purpose. They provide a measurable framework for assessing, monitoring, and improving data quality across an organization.
A marketing team sends a campaign to 40,000 customers – but 12,000 of those records have outdated addresses. A finance team builds a quarterly forecast on revenue data that, it turns out, counts the same transactions twice. A logistics company relies on delivery ETAs that are 45 minutes stale. Three different data quality issues. One root cause: nobody defined what “good” meant for their data before they used it.
Data quality dimensions are the answer to that problem. We’ve seen the same pattern across dozens of Collibra implementations – organizations that measure data quality against defined dimensions catch issues before they compound. Those that don’t are still arguing about which system holds the “real” number.
Key takeaways
- Data quality dimensions are standardized criteria for evaluating whether data is fit for its intended purpose – there are six core dimensions, and two additional ones increasingly critical in modern data environments.
- No universal standard exists: DAMA recognizes 11, Gartner 6, Wang & Strong identified 15 in 1996. The differences matter less than choosing a consistent framework and sticking to it.
- Each dimension can be measured with a specific formula – completeness ratio, validity rate, accuracy score – and each formula produces a number you can track over time.
- For AI and machine learning use cases, accuracy and completeness are non-negotiable for training data; timeliness and consistency become critical for production monitoring.
- Collibra implements data quality dimensions through configurable, no-code rules that run continuously across your data landscape – no one-time audits.
- The cost of skipping this: Gartner estimated poor data quality costs organizations an average of $12.9 million per year. That number gets worse, not better, as your data estate grows.
What are data quality dimensions?
Data quality dimensions are standardized measurement criteria used to evaluate whether data is fit for its intended purpose.
They give data teams a shared vocabulary for describing data problems – and a framework for building rules, checks, and monitoring that prevents those problems from reaching downstream consumers. Without dimensions, “data quality” stays abstract. With them, you can say “our customer records are 94% complete and 87% accurate” and know exactly what that means and what to do about it.
The concept was first formalized by Professors Richard Wang and Diane Strong in their 1996 paper “Beyond Accuracy: What Data Quality Means to Data Consumers,” which identified 15 original dimensions. Since then, every major framework has refined the list – but the six core dimensions have remained stable across all of them.
How many data quality dimensions are there?
This is the question that causes more confusion than it should. The short answer: it depends on which framework you follow. The practical answer: the core six are agreed upon by everyone; the rest are context-dependent.
| Framework | Number of dimensions | Key emphasis |
| Gartner | 6 | Accuracy, completeness, consistency, timeliness, validity, uniqueness |
| DAMA DMBOK | 11 | Includes conformity, integrity, precision, and others |
| Wang & Strong (1996) | 15 | Grouped into intrinsic, contextual, representational, accessibility |
| ISO 8000 | 4 | Accuracy, completeness, consistency, credibility |
| Murdio / Collibra implementations | 6 | 6 core |
The 6 core data quality dimensions
1. Accuracy: does your data reflect reality?
Accuracy is the degree to which data correctly represents the real-world entity or event it describes.
A customer address formatted perfectly but three years out of date is valid – but inaccurate. Accuracy is the hardest dimension to measure automatically because it requires a reference point: what is the ground truth? In practice, accuracy checks compare data against authoritative source systems, external databases, or manually validated samples.
Business impact: Inaccurate product data generates returns. Inaccurate patient data in healthcare creates clinical risk. Inaccurate financial records create regulatory exposure.
How to measure:
Accuracy Rate = (Records matching authoritative source / Total records sampled) x 100
How to improve: Establish authoritative source systems for critical data domains. Implement integration checks that validate data at point of entry against the source of record. Use data quality checks to flag deviations automatically.
2. Completeness: is anything missing?
Completeness measures the proportion of expected data values that are actually present in a dataset.
A required field left blank, a customer record missing industry classification, a product entry without a valid SKU – all of these are completeness failures. Unlike accuracy, completeness is usually straightforward to measure: either the value is there or it isn’t.
Business impact: Incomplete customer profiles limit segmentation. Missing supply chain data hides bottlenecks. Incomplete regulatory submissions create compliance risk.
How to measure:
Completeness Ratio = (Number of non-null values / Total expected values) x 100
How to improve: Define mandatory fields at data model level. Add completeness rules in your ingestion pipeline so incomplete records are flagged – not silently loaded. Set minimum completeness thresholds per data domain.
3. Consistency: does your data agree with itself?
Consistency means that data describing the same entity has the same value across all systems where it appears.
If your CRM says a customer is “Enterprise” tier but your billing system has them as “Mid-Market,” you have a consistency problem. The data isn’t wrong in isolation – it’s contradictory across systems. Consistency failures force teams into manual reconciliation and erode trust in reporting.
Business impact: Inconsistent data across systems makes unified reporting unreliable and creates operational friction when automated processes rely on data being the same everywhere.
How to measure:
Consistency Ratio = (Records with matching values across systems / Total records cross-checked) x 100
How to improve: Identify your master record for each key entity (customer, product, supplier). Implement master data management practices or use Collibra’s business glossary to enforce standard definitions. Cross-system consistency checks catch deviations before they compound.
4. Timeliness: is your data available when needed?
Timeliness measures whether data is available and up to date at the point it is needed for a decision or process.
Accurate data delivered 48 hours late is useless for real-time decisions. Timeliness failures often surface not as wrong data, but as delayed data – reports that reflect yesterday’s state when today’s state is what matters.
Business impact: Delayed operational data leads to missed opportunities and incorrect responses. In financial trading, even seconds matter. In logistics, 30-minute delays change route optimization decisions.
How to measure:
Timeliness Index = (Datasets updated within SLA / Total number of datasets) x 100
How to improve: Define explicit data freshness SLAs per domain. Monitor pipeline latency. Use data quality automation to alert on SLA breaches before consumers notice them.
5. Uniqueness: are you counting the same thing twice?
Uniqueness ensures that each real-world entity appears exactly once in a dataset – no duplicates.
Three records for “Catherine Jones,” “Cathy Jones,” and “C. Jones” at the same address represent one person, not three customers. Duplicates inflate counts, skew analytics, and waste resources – your marketing team sends three mailers to the same person and your analytics team thinks you have 30% more customers than you do.
Business impact: Duplicate records distort every metric built on top of them. They also create fragmented customer experiences and waste operational spend.
How to measure:
Uniqueness Rate = (1 – Duplicate Records / Total Records) x 100
How to improve: Implement entity resolution logic at ingestion. Define matching rules (exact match, fuzzy match by name and address) appropriate for each domain. Run deduplication as part of your data quality assessment cycle.
6. Validity: does your data follow the rules?
Validity means that data conforms to defined formats, types, ranges, and business rules.
A date field containing “N/A,” a ZIP code containing “London,” a revenue figure of -$50,000 when negative revenue is impossible by definition – these are validity failures. Validity doesn’t check whether the data is true; it checks whether it’s structurally correct according to the rules you’ve defined.
Business impact: Invalid data breaks automated processes, triggers errors in downstream systems, and requires manual cleanup that eats into your team’s capacity.
How to measure:
Validity Ratio = (Records conforming to defined rules / Total records) x 100
How to improve: Document your business rules explicitly per field. Implement rule-based validation at point of entry. Review rules regularly – business logic changes and your validation rules should reflect that.
Two more dimensions for modern data environments
The six core dimensions handle most data quality scenarios well. But two additional dimensions have become increasingly important as data landscapes grow more distributed and regulated.
7. Integrity: can your data be trusted across relationships?
Data integrity means that relationships between data entities are preserved and valid across all connected systems.
If a sales order references a customer ID that doesn’t exist in the customer master, you have an integrity failure – specifically, a referential integrity failure. As organizations connect more systems, integrity becomes harder to maintain automatically and more consequential when it breaks. It’s worth noting that data quality vs data integrity are related but distinct concepts – integrity is one dimension of quality, not a synonym for it.
How to improve: Implement referential integrity checks across connected domains. In Collibra, lineage-aware rules can validate that relationships between assets remain intact as data flows through pipelines.
8. Accessibility: can the right people get the data they need?
Accessibility measures whether authorized users can find, access, and use data when and where they need it.
This isn’t just about permissions. It includes discoverability (can people find the data?), latency (how long does access take?), and documentation (do people understand what they’re looking at?). Inaccessible data gets ignored – teams build shadow datasets instead, creating the consistency failures we described above.
Data quality dimensions for AI and machine learning
Organizations deploying AI models are discovering that standard data quality metrics are necessary but not sufficient. AI introduces specific failure modes that map directly to data quality dimensions.
- For training data, accuracy and completeness are non-negotiable. A model trained on inaccurate labels learns to reproduce those errors at scale. A model trained on incomplete data develops blind spots that surface unpredictably in production. Garbage in, garbage out – but now at the speed of inference.
- For production monitoring, timeliness and consistency become critical. A model trained on Q1 data patterns and deployed against Q4 data is working with a distribution shift it can’t see. Timeliness of training data refreshes and consistency between training and production datasets directly affects model reliability.
- Integrity matters for AI governance. Regulatory frameworks increasingly require organizations to document data lineage for AI systems – knowing which data was used to train which model, and what quality state that data was in. Collibra’s AI governance capabilities build on the same data quality foundation as the dimensions described here.
How to measure data quality: from formulas to scorecards
The formulas above give you a starting point. Turning them into operational measurement requires a framework:
- Step 1 – Identify Critical Data Elements (CDEs). Don’t try to measure everything at once. Start with the data domains where poor quality has the highest business impact. Revenue data, customer master data, and product catalogs are common starting points.
- Step 2 – Define business rules per dimension. A completeness threshold of 95% might be acceptable for marketing data and unacceptable for regulatory reporting. Thresholds are context-dependent. A solid data quality framework defines these thresholds per domain before measurement begins – not after the first incident.
- Step 3 – Establish your baseline. Run your first measurement pass before you implement improvements. You need a baseline to measure progress against.
- Step 4 – Build a data quality scorecard. Aggregate dimension scores per domain into a data quality scorecard that stakeholders can read without a data engineering background. A score of “Completeness: 94%, Validity: 88%” is actionable. A 40-row technical report is not.
- Step 5 – Monitor continuously. One-time audits tell you the state of your data at a point in time. Continuous monitoring tells you when something changes. A data quality dashboard connected to live data is the difference between reactive and proactive quality management.
We’ve seen organizations run manual quality audits quarterly and call it “governance.” They’re always responding to data incidents – never preventing them. Continuous monitoring built into the pipeline is what closes that gap.
How Collibra implements data quality dimensions
Collibra implements data quality dimensions through its Data Quality & Observability product – a no-code rules engine that runs continuously across connected data sources. The key difference from traditional DQ tools: rules in Collibra are tied to assets in the catalog, so a quality score on a dataset is visible to everyone who accesses that asset’s entry.
In practice, a Murdio implementation typically looks like this:
- Rules are defined per dimension using Collibra’s UI – no SQL required for standard checks, custom SQL available for complex business logic.
- Checks run on schedule or on trigger – after pipeline execution, on data arrival, or continuously for streaming sources.
- Scores roll up to scorecards linked to data products, making quality visible at the point of consumption in the Collibra Data Marketplace.
- Issues trigger workflows – Collibra routes quality failures to the appropriate data steward for remediation, rather than silently logging them.
The most common mistake we see during Collibra implementations: teams define rules for accuracy and completeness, declare victory, and stop there. The real value comes from covering all eight dimensions, connecting quality scores to business context, and making those scores visible to data consumers – not just data engineers.
For details on what a Collibra Data Quality implementation looks like in practice, including common rule configurations and scorecard structures, see our dedicated guide.
Signs your data quality dimensions aren’t working
You can have a data quality framework on paper and still have a data quality problem in practice. These are the patterns we see most often – what we call the Measurement Theater trap, where teams track dimensions without acting on what they find.
- For Data Managers: Your quality scores look fine at the domain level, but individual business units are still building their own extracts to “fix” the data before using it. The dimensions are measuring the wrong things, or at too coarse a granularity.
- For Data Governance Managers: Dimension scores aren’t connected to business impact. Nobody can answer “what does 87% validity mean for our monthly close?” If your CDO asks what a score means for regulatory risk and you can’t answer, the framework isn’t operational.
- For CDOs: Quality scores exist but don’t change behavior. Data producers don’t see their scores. Data stewards don’t have SLAs tied to thresholds. The framework is decorative. Real adoption happens when quality scores affect priorities – when a business unit sees their domain in the red and escalates to fix it.
Poor data quality improvement programs usually fail not on the measurement side but on the accountability side. Defining dimensions is week one. Wiring them to responsibilities and incentives is the rest of the project. For the full set of operational recommendations, see our guide to data quality best practices.
Defining data quality dimensions is the straightforward part. The challenge is making them operational – connecting scores to responsibilities, thresholds to business outcomes, and measurements to change. We’ve built and rebuilt this infrastructure across dozens of Collibra implementations, and we know exactly where programs stall and why.
If you’re at the point where dimension definitions exist but adoption hasn’t followed, or you’re starting from scratch and want to avoid the common traps – let’s talk.
It depends on the framework. Gartner uses 6 core dimensions; DAMA DMBOK defines 11; Wang and Strong originally identified 15. There is no universal standard. In most Collibra implementations, we work with the 6 core ones – because these cover the most common failure modes in enterprise data environments.
Accuracy (does the data reflect reality?), completeness (are all expected values present?), consistency (does data agree across systems?), timeliness (is data available when needed?), uniqueness (are there duplicates?), and validity (does data conform to defined rules?) are the six core examples. Integrity and accessibility are two additional dimensions relevant to modern data environments.
Each dimension has a corresponding formula. Completeness is measured as (non-null values / expected values) x 100. Validity is (records conforming to rules / total records) x 100. Accuracy requires a reference source to compare against. Scores per dimension are aggregated into a data quality scorecard per domain.
Dimensions are the categories of quality you care about (accuracy, completeness, etc.). Metrics are the specific numbers you track within each dimension (accuracy rate: 94.2%). Dimensions define what to measure; metrics define how the measurement is expressed and tracked over time.
For AI training data, accuracy and completeness are most critical – errors in training labels or missing values propagate into model behavior at inference time. For production monitoring, timeliness and consistency matter most – distribution shift between training and production data is a leading cause of model degradation.
Collibra’s Data Quality & Observability product lets teams define rules per dimension using a no-code interface, run them continuously against connected data sources, and surface results in scorecards linked to catalog assets. Quality scores are visible to data consumers directly in Collibra, not buried in a separate tool. Murdio helps organizations design and implement these rule sets as part of Collibra Data Quality implementations.
In Collibra, a rule violation triggers a configurable workflow – routing the issue to the relevant data steward, generating a ticket, or escalating based on severity. Without automated routing, quality failures get logged and ignored. With it, they get fixed. The difference is accountability, not tooling.
