Data quality dimensions: why 'good enough' destroys trust
IBM famously estimated that poor data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → costs the US economy $3.1 trillion per year. Gartner puts the average cost to an individual organization at $12.9 million annually.
These numbers sound abstract until you trace specific costs to specific data quality failures: a bank sending loan offers to deceased customers. A manufacturer shipping products to wrong addresses because of duplicate postal codes. A hospital administering the wrong medication dosage due to a unit conversion error in the EHR system. A retailer's demand forecast 40% off because historical sales data included returns.
Data quality isn't a technical concern. It's a business risk with a measurable price tag.
The six dimensions of data quality
The DAMA-DMBOK defines six dimensions of data quality. Every data quality problem maps to at least one of them:
1. Accuracy, Does the data correctly represent the real-world entity or event it describes? A customer's address is accurate if it matches their actual address. Accuracy failures: "Paris" recorded as "Parsi" due to a typo; transaction amounts rounded incorrectly; product weight recorded in the wrong unit.
2. Completeness, Are all required data elements present? A customer record missing an email address is incomplete if email is required for the business process. Completeness failures are often systemic: optional fields that should be mandatory, APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → integrations that drop fields, migration scripts that don't transfer all attributes.
3. Consistency, Is the same data represented the same way across all systems? A customer's date of birth recorded as DD/MM/YYYY in the CRMCRMCustomer Relationship Management: software and strategy to manage and analyse customer interactions throughout their lifecycle.View full definition → but MM/DD/YYYY in the billing system is an inconsistency. Consistency failures are the most common root cause of "which system is right?" debates.
4. Timeliness, Is the data available when it's needed, and does it reflect the current state? A customer's credit score updated monthly is timely for annual mortgage reviews but not for real-time lending decisions. Timeliness failures often emerge as "we knew about the issue but the data hadn't refreshed yet."
5. Validity, Does the data conform to defined formats, ranges, and business rules? A phone number with 11 digits in a field defined for 10 is invalid. A transaction date in the future is invalid. Validity failures are the easiest to detect and fix at the point of entry.
6. Uniqueness, Is each real-world entity represented exactly once? Duplicate customer records are a uniqueness failure. Deduplication is one of the most common data quality initiatives, and one of the most underestimated in complexity.
Data Quality Management | DAMA DMBOK Chapter 13 Explained
Knowledge check
1. A customer's date of birth is stored as DD/MM/YYYY in the CRM but as MM/DD/YYYY in the billing system. Which data quality dimension does this failure primarily violate?
2. A lender uses a customer's credit score for real-time lending decisions, but the score only refreshes monthly. What dimension of data quality is most at risk here?
3. What is the core message behind framing poor data quality with a dollar cost to the economy and to organizations?
4. Select ALL scenarios that are examples of a VALIDITY failure (data not conforming to defined formats, ranges, or business rules).
Select all the correct answers.
5. Select ALL statements that correctly describe how the six DAMA-DMBOK dimensions relate to data quality problems.
Select all the correct answers.
Measuring data quality: from subjective to quantitative
"Our data quality is pretty good" is not a measurement. "Our customer master data is 94% complete, 87% accurate, and 91% consistent across CRM and billing systems" is.
Data quality scoring: For each data domain, define a quality score as a weighted average of the six dimensions. Weight dimensions according to business importance. For customer data used in marketing, completeness and accuracy of contact information might weight heavily. For financial data used in reporting, consistency and timeliness matter most.
Profiling: Before you can score, you need to profile. Data profiling tools (Great Expectations, Informatica Data Quality, Ataccama) analyze datasets to reveal: null rates, value distributions, format violations, duplicate rates, referential integrity failures. Profiling a new dataset for the first time is almost always revealing, and often alarming.
Quality dashboards: The CDO should have a real-time data quality dashboard showing quality scores by domain, trending over time. This dashboard should be shared with domain data owners, making quality visible is the first step to making it accountable.
The business case for data quality investment
The ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition → calculation for data quality is straightforward:
Current cost of poor quality: Quantify at least one measurable cost, rework time, error corrections, customer complaints, regulatory fines, missed opportunities from incomplete data.
Cost of quality initiative: Tool licensing + implementation + ongoing stewardship + profiling and monitoring.
Target state: What quality score will you achieve, and what business outcome does that enable?
A European telecom calculated that 12% of their marketing budget was wasted targeting inactive customers due to poor data completeness and accuracy. A €50K data quality initiative that reduced waste by 50% delivered a €3M annual saving on a €25M marketing budget, a 60x ROI. That's the language CFOs understand.
Frequently asked questions
What are the six dimensions of data quality?
The DAMA-DMBOK defines six: accuracy, completeness, consistency, timeliness, validity and uniqueness. Accuracy asks whether data matches the real-world entity, completeness whether required fields are present, consistency whether the same fact looks identical across systems, timeliness whether it is current when needed, validity whether it respects defined formats and ranges, and uniqueness whether each entity appears exactly once. Every data quality problem maps to at least one of them.
How much does poor data quality actually cost a company?
Gartner estimates the average cost at $12.9 million per year per organization, and IBM put the cost to the US economy at $3.1 trillion annually. These figures only become actionable when traced to specific failures: a bank mailing loan offers to deceased customers, a demand forecast off by 40% because returns were counted as sales, marketing spend burned on inactive contacts. Quantify one of those locally and you have a number a CFO will act on.
What is the difference between data accuracy and data validity?
Validity means the data respects the defined format, range or business rule; accuracy means it describes the real world correctly. A phone number with ten digits in a ten-digit field is valid even if it belongs to nobody. Validity can be enforced automatically at the point of entry, while accuracy usually requires comparison with an external source of truth.
How do you measure data quality instead of just saying it's "pretty good"?
Build a quality score per data domain as a weighted average of the six DAMA dimensions, weighting each according to business use. "Our customer master data is 94% complete, 87% accurate and 91% consistent across CRM and billing" is a measurement; "our data is pretty good" is not. For marketing contact data, completeness and accuracy carry the weight; for financial reporting, consistency and timeliness matter more.
What does data profiling reveal, and which tools do it?
Profiling analyses a dataset to expose null rates, value distributions, format violations, duplicate rates and referential integrity failures. It is the prerequisite to any quality score, since you cannot weight dimensions you have never measured. Great Expectations, Informatica Data Quality and Ataccama are commonly used for this. Profiling a dataset for the first time is usually revealing and often alarming.
How do you build the ROI case for a data quality initiative?
Three figures: the current cost of poor quality, the cost of the initiative, and the target quality score with the business outcome it unlocks. A European telecom found 12% of its €25M marketing budget was wasted on inactive customers because of incomplete and inaccurate records; a €50K quality initiative cut that waste in half and saved €3M a year, a 60x return. Quantifying one measurable cost, rework time, error corrections, regulatory fines, is enough to start the conversation.
What to do, from this lesson
These actions are compiled in the role's Playbook.
- Profile datasets and publish real-time quality dashboards by domain
Related articles
Recent articles from the blog that build on this lesson.
- DataAI slop detectors make your training data worse before they make it betterFiltering AI-generated text from training datasets sounds like straightforward hygiene, but a recent experiment shows the cure can degrade model performance more than the contamination itself. CDOs who treat this as a tooling problem will miss the governance question underneath it.
- DataHow Salesforce learned to make master data stickSalesforce spent years selling data quality to its customers while quietly struggling with fragmented customer and product records across its own acquisitions. The way the company addressed that internal contradiction holds practical lessons for any CDO trying to move MDM from a slide deck into operating reality.
- DataData observability: catching bad data before it reaches decisionsBad data doesn't announce itself. This playbook shows CDOs how to build detection mechanisms that intercept data quality failures before they corrupt reports, models, and the decisions that follow.