Is this model an AI algorithm?

No. It is a framework of quality characteristics used to organize requirements and evaluation.

Must every project use identical thresholds?

No. A shared framework can support different requirements for different contexts.

Does a quality model generate evidence automatically?

No. Measurements, reviews and other assessments are needed to determine actual quality.

AI Glossary / Governance, ethics and safety / Intermediate

Data quality model

Definition

A data quality model organizes the characteristics used to define and evaluate data quality. It provides a shared framework for deciding which aspects of data matter, such as completeness, accuracy, consistency and timeliness, for a specified purpose.

What it means

Here, model means a conceptual framework, not a machine-learning predictor. It helps a team move from a vague demand for good data to a structured set of concerns that can be translated into requirements and assessments.

Different uses can place different importance on the same characteristic. A historical research dataset may tolerate slower delivery than a real-time operational feed. The quality model identifies relevant dimensions; the application then needs specific requirements, measures and evidence. Adopting a model alone does not establish that a dataset meets those requirements.

How it works

Start by identifying stakeholders, intended uses and consequential data failures. Select and define quality characteristics that address those concerns. Explain overlaps so that teams do not use the same word for different ideas or double-count one defect.

Map each characteristic to requirements and assessment methods. Completeness might be assessed through mandatory-field checks, while accuracy might require comparison with source evidence. A model can also organize characteristics into groups or levels, provided the relationships are clear.

Review the framework when the data or intended use changes. Keep the conceptual model separate from individual test results and implementation tools. A spreadsheet, catalog or validation service may implement parts of it, but the framework remains the agreed structure for reasoning about quality.

Three practical examples

Forecasting project: A team includes coverage, timeliness and consistency because missing regions, delayed records and changing event definitions can distort demand estimates.
Document-learning dataset: Developers consider label accuracy, annotation consistency and representation of relevant document types before selecting checks.
Shared data platform: Several departments use a common set of quality terms while maintaining different acceptance thresholds for their own applications.

Think of it this way

Think of a building inspection framework that separates structural condition, electrical systems and accessibility. It organizes what should be examined but does not itself inspect the building. A data quality model likewise creates a structure for assessment; evidence must still be collected for each relevant concern.

Worked example

Imagine a dataset assessed under four illustrative characteristics. Completeness measures required-field presence; accuracy uses sampled source comparison; timeliness measures arrival delay; consistency checks compatible event sequences.

A release has 99% required-field completeness, 94% sampled source agreement, a median arrival delay of six hours and twenty sequence conflicts. These values cannot be sensibly averaged as written because they use different scales and meanings.

The team instead compares each with its own requirement and investigates failures. If a composite index is later introduced, its transformation and weights must be explained. The example shows how a model organizes assessment without pretending that all characteristics collapse naturally into one universal percentage.

Illustrative case study

Fictional case: Insight Harbour receives a request to certify that a training dataset is “good enough.” The team first defines a quality model around the intended prediction task and maps each characteristic to a responsible owner.

A previous dashboard covered only missing fields. The new framework reveals that outdated category definitions and incomplete regional coverage also matter. The team adds assessments for those concerns and retains unresolved limitations in the handover. This invented case illustrates broader reasoning, not formal certification or measured model improvement.

Common misconceptions and mistakes

A data quality model is not a trained model that automatically repairs data. It does not replace requirements, measurement or judgment about intended use.

A long list of characteristics is not necessarily better than a concise, relevant framework. Overlapping terms can create confusion, and generic dimensions can miss domain-specific failures. Another mistake is treating an adopted framework as proof of compliance with every associated standard. Conformity requires examining the applicable requirements and evidence, not simply using similar terminology in a diagram.

Business use

The framework can improve coordination across data producers, engineers, analysts and model owners. It gives teams a shared way to describe defects, prioritize remediation and explain why a dataset is suitable for one use but not another.

Keep the model understandable enough to guide decisions. Assign ownership for maintaining definitions and map measures back to their intended characteristics. When purchasing data or AI services, ask which quality dimensions were assessed and which were omitted. A single quality score should not hide the framework or its exclusions.

Check your understanding

A team adopts a framework with six quality characteristics but measures only completeness. Has it demonstrated quality across all six?

Answer: No. The framework identifies concerns; evidence is still needed for each claim the team intends to make.

Related terms

Data quality measure
Data governance
Data quality plan
Metadata
Data quality characteristic
data quality requirement
fitness for purpose
assessment framework

Learn next

Draw a small quality model for a specific dataset with four or five relevant characteristics. For each, describe one failure that matters to the intended user and one assessment method. Identify overlaps and gaps. Then separate the framework from the actual acceptance thresholds and measurement results.

Sources

ITU-T Y Supplement 97 (11/2025), term 3.1.34, consulted directly in the official PDF: https://www.itu.int/rec/T-REC-Y.Sup97-202511-I/en. The glossary attributes this definition to ISO/IEC 5259-1. ISO publisher overview consulted: https://www.iso.org/standard/81088.html; the complete ISO standard was not accessed. The examples and workflow are original explanatory material, not prescribed thresholds or a conformity assessment.

Previous: Data quality measure Next: Data quality plan