---
title: "Data quality: dimensions, metrics and where to start"
description: "The DAMA data quality dimensions, how to control them with rules in tools such as Microsoft Purview, which metrics to track and the four controls to start with, keeping it simple."
url: https://bidatia.com/en/ideas/calidad-del-dato-dimensiones-metricas
lang: en
date: 2026-05-12
updated: 2026-11-04
author: "Santos Vicente"
keywords: ["data quality", "data quality dimensions", "dama data quality dimensions", "kqi", "data quality metrics", "microsoft purview data quality", "une 0079", "data quality rules"]
publisher: Bidatia
---

# Data quality: dimensions, metrics and where to start

> The DAMA data quality dimensions, how to control them with rules in tools such as Microsoft Purview, which metrics to track and the four controls to start with, keeping it simple.

## Executive summary

- Data quality is the degree to which data meets the requirements for its use, and it is managed through a model of policies, processes and accountable people, not with one-off clean-ups.
- DAMA dimensions: accuracy, completeness, consistency, timeliness, uniqueness, validity and integrity, each with its own rule type.
- Rule results generate quality indicators (KQIs) with targets, thresholds and severity (blocking, major, warning).
- Four checks are enough to start (duplicates, freshness, mandatory fields and formats), applied first to data used by AI systems and to authoritative sources.
- Includes example rules per dimension, the incident workflow, a KQI record, a path through maturity levels and tools per layer (Purview, Great Expectations, Power BI).

Data quality is the degree to which data meets the requirements for its use. It is not an abstract ideal: it is the difference between a dashboard that senior management trusts and one that nobody looks at, or between an AI assistant that gets it right and one that makes things up.

## A quality model, not a one-off clean-up

Cleaning the data once is pointless if the source keeps generating errors. A quality model brings together policies, processes and people accountable:

- **Who defines the rules**: the data governance team; the data governance lead defines them and the custodians implement them.
- **Who validates**: owners and stewards receive the indicators, validate them functionally and report impacts.
- **How problems are detected**: automatic alerts from the tools plus functional checks by those who know the data.
- **How they are resolved**: the governance lead prioritises and tracks the incident; the custodian fixes it; the steward confirms it is resolved.
- **How it improves**: a prioritised improvement roadmap, delivered in cycles.

## The quality dimensions

| Dimension | Question | Example check | Rule type |
|---|---|---|---|
| Accuracy | Does the value reflect reality? | Matches a source of truth | Custom |
| Completeness | Are the mandatory values present? | Mandatory fields not empty, counting disguised nulls | Empty fields |
| Consistency | Are values coherent with each other and across systems? | The end date is later than the start date | Data type or custom |
| Timeliness | Is it up to date? | The dataset is refreshed at the expected frequency | Freshness |
| Uniqueness | Does each thing appear only once? | Primary key with no duplicates | Unique values or duplicate rows |
| Validity | Does it match the format? | Email, national ID or postcode format | Format match |
| Integrity | Do the relationships hold? | The key exists in the master table | Table lookup |

## Metrics: from rules to indicators

The results of the rules become each dataset's **quality indicators (KQIs)**:

- Percentage of empty values in mandatory fields.
- Percentage of duplicates.
- Percentage of records with a valid format.
- Percentage of referential integrity.
- Compliance percentage: rules passed out of the total.

They are published on the data governance dashboard and reviewed by the committee, with thresholds that trigger alerts.

## What tools can and cannot do

Catalogue tools with quality features, such as Microsoft Purview Data Quality, let you define rules, run them, generate alerts and assign actions. But their rules are mainly **simple validations on a single dataset**: blanks, ranges, lists, counts or duplicates. Checks with complex business logic, or that compare several systems, are implemented in the data processes themselves or with specialised tools such as Great Expectations or Pandera.

## Where to start

The usual mistake is trying to measure everything. It works better to start with four checks that the tool configures directly:

1. **Duplicates by key**, to guarantee uniqueness.
2. **Freshness**, to check that datasets arrive when they should.
3. **Mandatory fields**, to detect blanks.
4. **Basic formats**, for validity.

And apply them first to two types of data: data used by **AI systems** and **authoritative sources** (the single version of each entity). The rest is extended as the model matures.

## One example rule per dimension

For a customer dataset, this is how the dimensions would translate into specific rules:

| Dimension | Rule | Severity |
|---|---|---|
| Completeness | The tax ID (NIF) cannot be empty, or contain “0”, “000000000” or “SIN NIF” | Blocking |
| Uniqueness | There cannot be two active customers with the same NIF | Major |
| Validity | The email address matches the format; the postcode has five digits | Warning |
| Consistency | The end date is later than the start date | Major |
| Integrity | The municipality code exists in the municipalities table | Major |
| Timeliness | The dataset has been loaded in the last 24 hours | Blocking |
| Accuracy | The total of active customers matches the source system (±0.5%) | Warning |

Severity determines the response: a **blocking** rule stops the load and notifies the data owner; a **major** one only lets it continue with explicit approval; a **warning** is logged and processing continues.

## The workflow for a data quality incident

1. **Detection**: a rule fires, or a data steward spots something in a report that does not add up.
2. **Logging**: the incident is raised through the ticketing tool with its dataset, affected dimension, impact and urgency.
3. **Analysis**: the steward helps assess the functional impact; the custodian looks for the technical cause.
4. **Correction**: at source whenever possible. Fixing it only in the data platform is a patch that will recur with the next load.
5. **Validation**: the steward confirms the problem is resolved.
6. **Prevention**: if it did not already exist, the rule that would have detected it is created.
7. **Reporting**: owners periodically receive the status and trends.

## How to describe a quality indicator

Like any other indicator, a KQI needs a clear record:

| Element | Example: completeness of customer NIF |
|---|---|
| What is measured | Percentage of active customers with a populated and valid NIF |
| How it is measured | Active customers with a valid NIF divided by total active customers |
| Unit | Percentage |
| Thresholds | Target ≥ 99.5%; alert below 98% |
| Interpretation | A drop usually points to a change in the registration form or in the load |

## Quality and maturity

In the data governance maturity assessment (UNE 0080), quality management is one of the processes that carries the most weight. A realistic path:

- **Level 2**: first rules on critical data and accountable people identified.
- **Level 3**: corporate quality model, with dimensions and processes defined for the whole organisation.
- **Level 4**: indicators with targets and thresholds, a dashboard and root cause analysis.
- **Level 5**: continuous improvement and managed quality to get the maximum value from data.

## Tools at each layer

- **In the catalogue** (Microsoft Purview Data Quality, OpenMetadata): simple rules, alerts and tracking visible to the business.
- **In the load processes** (Great Expectations, Pandera, validations in Fabric notebooks or pipelines): complex rules, cross-system comparisons and blocking loads.
- **On the data governance dashboard** (Power BI): the evolution of indicators by domain and product.

## Common mistakes

- **Measuring without an owner**: an alert that nobody receives is useless.
- **Always fixing at the destination** and never in the source system.
- **Rules that do not count disguised nulls**: a “999” or a “01/01/1900” passes any empty-field check.
- **A hundred rules in the first month** and none maintained by the sixth.
- **An overall quality percentage** that hides the fact that a critical data item is wrong.

## Quality starts in development

Every data product should include, from the moment it is built, checks on formats and values, completeness, duplicates and consistency across datasets. And every problem found should become an automatic rule, with an owner and a severity, so that it does not reappear. We explain this in more detail in [biases and data problems that make AI fail](/en/ideas/sesgos-y-calidad-de-datos-en-ia).

## Balancing quality with data protection

The Spanish Data Protection Agency (AEPD) points out that technical quality should not lead to disproportionate demands that clash with data minimisation: the aim is not to collect more data so that it is “more complete”, but to ensure that the necessary data is correct.

At Bidatia we implement data quality models with Microsoft Purview, OpenMetadata and automated validations, aligned with DAMA and UNE 0079. For the full framework, see [data governance: roles and organisation](/en/ideas/gobierno-del-dato-roles-y-organizacion) and [master data and authoritative sources](/en/ideas/datos-maestros-y-fuentes-de-autoridad).

## Frequently asked questions

### What are the data quality dimensions?
The most widely used, according to DAMA, are accuracy, completeness, consistency, timeliness, uniqueness, validity and referential integrity.

### What is a KQI?
A key quality indicator: it measures compliance with the quality rules of a dataset or data product across one or more dimensions.

### Can Microsoft Purview be used to control data quality?
Yes, for simple validations on datasets: blanks, duplicates, formats, ranges and freshness. Complex or cross-system business rules are implemented in the data processes or with specialised tools.

### Which checks should I start with?
Four: duplicates by key, freshness, mandatory fields and basic formats, applied first to the data used by AI systems and to authoritative sources.

### Which standard covers data quality management?
The UNE 0079 specification, together with UNE 0077 (governance) and UNE 0078 (management). UNE 0080 makes it possible to assess the maturity of those processes.