---
title: "Data quality: definition and example – Carl-Uwe Würfel"
description: "Data quality explained: the six dimensions, how to measure them, why they decide what AI gets right or wrong, and who in a company owns the job."
url: https://carl-wuerfel.de/en/glossary/datenqualitaet
author: Carl-Uwe Würfel
---

1. [Home](https://carl-wuerfel.de/en)
  2. [Knowledge](https://carl-wuerfel.de/en/knowledge)
  3. [Glossary](https://carl-wuerfel.de/en/glossary)
  4. Data quality

# Data quality

Data quality describes how well data fits its specific purpose, measured by dimensions such as accuracy, completeness, consistency, uniqueness, validity and timeliness.

## On this page

  1. The dimensions of data quality
  2. Why data quality decides AI results
  3. How to measure data quality
  4. Fix causes instead of cleaning up after them
  5. Who owns data quality
  6. What privacy law and the AI Act require
  7. Where the term stops working
  8. Frequently asked questions
  9. Sources

The model doesn't check whether a source is still valid. The data has to say so.

_Figure: Comparison using an AI search over internal documents (RAG). Without a date and validity, the model can quote the old price list instead of the new one, just as convincingly; with that metadata, outdated documents can be filtered out. Duplicates with different content produce contradictory answers. Without access rights, content reaches people who should not see it. Scans without readable text drop out of search. Below: fix causes instead of cleaning up after them._

**Example**

A wholesaler rolls out an [AI assistant](https://carl-wuerfel.de/en/glossary/ki-assistent) that answers sales questions from product data sheets. In the first weeks, the assistant repeatedly quotes delivery times from outdated sheets stored in the same folder as the current ones. The company then tags every document with a validity date and owning department, archives older versions and limits the assistant to valid documents.

Data is not good or bad in itself. An address list that works for a newsletter can be useless for invoicing, where every street number has to be right. Data quality is therefore tied to purpose: the dimensions stay the same, and the requirements for each one follow from how the data is used.

The term comes from information and data management. Standards such as ISO/IEC 25012 and the ISO 8000 series, along with the reference body of knowledge from the Data Management Association (DAMA), give business and IT a shared vocabulary. Three terms often get mixed up: [data governance](https://carl-wuerfel.de/en/glossary/data-governance) sets rules and responsibilities, data quality describes and measures the state of the data, and data cleansing fixes individual errors.

AI gives this old topic new weight. A language model that answers questions from internal documents does not check its sources. It repeats outdated or contradictory content fluently and convincingly. For executives, that means the quality of an AI assistant's answers depends on data that was usually created long before the AI project started.

## The dimensions of data quality.

In 2013 a DAMA UK working group described six core dimensions that have since become common language. Accuracy asks whether a value matches reality. Completeness asks whether all required values are present. Consistency asks whether the same fact is represented the same way across systems. Uniqueness asks whether each entity, such as a customer, appears only once. Validity asks whether values conform to the agreed format and range. Timeliness asks whether the data reflects reality as of the point in time it is needed.

The international standard ISO/IEC 25012, published in 2008, goes further. It defines 15 characteristics from two points of view. Inherent characteristics live in the data itself, such as accuracy, completeness and credibility. System-dependent characteristics depend on the technology, such as availability, recoverability and portability. Some, like confidentiality and traceability, belong to both. The ISO 8000 series covers data and master data quality, including the exchange of master data between companies.

For day-to-day work, the six core dimensions are usually enough. What matters is turning them into concrete requirements for each dataset instead of rating them in the abstract.

## Why data quality decides AI results.

Many AI applications inside companies use [retrieval-augmented generation](https://carl-wuerfel.de/en/glossary/rag) (RAG). A search pulls relevant passages from internal documents, and a language model writes the answer from them. The model does not judge whether a source is still valid. If an old and a new price list sit side by side, it may quote the old one, just as convincingly.

Common data errors behave differently here than in a spreadsheet. Duplicates with conflicting content produce contradictory answers. Without metadata such as date, validity or owning department, outdated documents cannot be filtered out. Without access-rights metadata, content reaches people who should not see it. Scanned documents without machine-readable text drop out of search, and the model sometimes fills gaps with made-up details, known as [hallucinations](https://carl-wuerfel.de/en/glossary/hallucination).

With models trained or fine-tuned on company data, the effect is stronger. Errors in the training data end up in the model and are hard to remove afterward.

That sets the order of work. Before an AI assistant gets access to a dataset, the dataset is checked, cleaned and given clear ownership. This step belongs in the project plan and budget, not in the weeks after launch.

## How to measure data quality.

Data quality is measured by checking data against rules derived from its purpose. It starts with the question of what a dataset is used for and which errors cause harm there. That yields rules per dimension, for example: every customer record has a valid billing address. Every customer number appears only once. Every document in the knowledge base has a review date no older than one year.

Each rule produces a metric, such as the share of required fields filled, the duplicate rate, the share of valid formats or the average age of entries. The business side sets a target value for each metric.

Many of these checks can be automated. Data profiling shows value ranges, empty fields and anomalies. Validation rules in data pipelines stop faulty records before they move downstream. Accuracy is the exception: no format check can tell whether a phone number belongs to the right person. That takes spot checks against reality, such as matching records to source documents or asking the customer.

Measurement is ongoing, not a one-off. The trend in the metrics shows whether fixes are working and where new errors appear.

## Fix causes instead of cleaning up after them.

The obvious response to bad data is a cleanup project: merge duplicates, fill in fields, standardize formats. That helps for now. If the cause remains, the same errors are back within a few months.

Most errors start at the point of entry. Typical sources are free-text fields where a dropdown would do, missing required fields, imports without validation, and several systems where the same customers are created independently. What works is validation at entry, clear formats, picklists and a single place where a master record is created.

Problems become most visible during a system migration, for example when moving from a legacy system and scattered spreadsheets to one shared system. Only during reconciliation does it turn out that the same customer exists several times, spelled differently or with conflicting details. This is where master data management comes in: maintaining master data consistently across all systems.

The same logic applies to knowledge bases that an AI uses. Every document gets a date, an owner and a status when it is created. Outdated versions are archived instead of sitting next to the current one.

## Who owns data quality.

Data quality is not purely an IT job. IT provides systems, validation tools and interfaces. But what a record has to mean and which error causes harm is known by the business unit that works with the data.

Data management has settled on two roles for this. The data owner, usually a manager in the business unit, is accountable for a data domain such as customer or product data and decides on rules and targets. The data steward looks after the dataset day to day: maintaining rules, watching the metrics, resolving errors with the people who enter the data and triggering corrections. In smaller companies, one person often holds both roles.

Data governance sets the frame. It defines who holds which role, which rules apply and how conflicts between departments are settled, for instance when sales and accounting record a customer differently. Without that frame, nobody feels responsible for data quality until a report or an AI assistant produces wrong results.

For executives, the question that matters most is simple: for every important dataset, is it clear who decides on its rules?

## What privacy law and the AI Act require.

Data quality is also a legal requirement in the EU. Article 5(1)(d) of the General Data Protection Regulation requires personal data to be accurate and, where necessary, kept up to date. Inaccurate data must be erased or corrected without delay, and under Article 16 people have a right to have their data corrected.

The EU [AI Act](https://carl-wuerfel.de/en/ai-act) sets its own requirements for high-risk AI systems. Under Article 10, training, validation and testing data must be relevant, sufficiently representative and, to the best extent possible, free of errors and complete in view of the intended purpose. Providers of such systems must document, among other things, where the data comes from, how it was prepared and how possible biases were examined. These obligations cover only certain uses, such as hiring or credit scoring. After the July 2026 amendment they apply to high-risk systems under Annex III from December 2027 and to those under Annex I from August 2028 (as of November 2026).

For most mid-sized companies, Article 10 therefore does not apply directly. It still describes well what a careful data foundation for AI looks like: known origin, documented preparation, and gaps and biases checked.

## Where the term stops working

Data quality describes the state of data relative to its purpose. It is distinct from data governance, which sets rules, roles and responsibilities; from data cleansing, which is a single corrective step; and from data integrity, which means protection against alteration and loss. Clean data also does not automatically answer the right question: whether a dataset is fit for a decision depends on the purpose.

## Frequently asked questions

What is data quality?

Data quality is the degree to which data is fit for its intended use. It is measured by dimensions such as accuracy, completeness, consistency, uniqueness, validity and timeliness. Which requirements apply depends on the purpose.

What are the dimensions of data quality?

The most widely used are the six core dimensions DAMA UK described in 2013: accuracy, completeness, consistency, uniqueness, validity and timeliness. The ISO/IEC 25012 standard goes into more detail with 15 characteristics, including credibility, understandability and traceability.

What is the difference between data quality and data integrity?

Data quality asks whether data is fit for its purpose. Data integrity asks whether it is unaltered, completely transferred and protected from unauthorized change. A correctly stored but outdated address has integrity and still has poor quality.

How do you measure data quality?

You derive rules per dimension from the intended use and test the data against them. Each rule produces a metric, such as fill rate or duplicate rate, with a target value. Accuracy also needs spot checks against reality.

Why does data quality matter for AI?

Because an AI model does not check its sources. Outdated, duplicate or contradictory data leads to wrong answers that sound convincing. With models trained on company data, errors also end up in the model itself.

Who is responsible for data quality in a company?

The business unit that works with the data, represented by a data owner who sets rules and targets. A data steward maintains the dataset day to day, and IT provides systems and validation tools. Data governance sets the frame.

What does the EU AI Act require for data quality?

Article 10 requires training, validation and testing data for high-risk AI systems to be relevant, sufficiently representative and, as far as possible, free of errors and complete. These obligations apply only to certain uses, from December 2027 or August 2028 depending on the type of high-risk system. Separately, the GDPR requires personal data to be accurate and up to date.

What software helps with data quality?

Useful tools include profiling features in databases and analytics tools, validation rules in data pipelines and master data management software. Software flags deviations but does not decide which rule is right for the business. That call stays with the business unit.

## Further reading

  * [**AI glossary**](https://carl-wuerfel.de/en/glossary)
  * [**AI knowledge management**](https://carl-wuerfel.de/en/ai-knowledge-management)
  * [**AI governance**](https://carl-wuerfel.de/en/ai-governance)
  * [**RAG: Retrieval-Augmented Generation**](https://carl-wuerfel.de/en/glossary/rag)
  * [**Hallucination**](https://carl-wuerfel.de/en/glossary/hallucination)
  * [**LLM: Large Language Model** Term](https://carl-wuerfel.de/en/glossary/llm)
  * [**Vector Database** Term](https://carl-wuerfel.de/en/glossary/vector-database)
  * [**Fine-Tuning** Term](https://carl-wuerfel.de/en/glossary/fine-tuning)

## Sources

  1. [ISO/IEC: ISO/IEC 25012:2008 Software engineering – Software product Quality Requirements and Evaluation (SQuaRE) – Data quality model, 2008](https://www.iso.org/standard/35736.html)
  2. [ISO: ISO 8000-1:2022 Data quality – Part 1: Overview, 2022](https://www.iso.org/standard/81745.html)
  3. [DAMA International: Data Management Body of Knowledge (DAMA-DMBOK), 2nd edition](https://dama.org/learning-resources/dama-data-management-body-of-knowledge-dmbok/)
  4. [European Parliament and Council: Regulation (EU) 2016/679 (General Data Protection Regulation), 2016](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679)
  5. [European Parliament and Council: Regulation (EU) 2024/1689 (Artificial Intelligence Act), 2024](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689)
  6. [European Parliament and Council: Regulation (EU) 2026/1744 (Digital Omnibus on AI), 2026](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1744)