AI Glossary / Governance, ethics and safety / Beginner
Data quality plan
Definition
A data quality plan specifies how an organization or project will achieve its data quality objectives. It sets out relevant practices, processes and resources, turning quality intentions into an actionable arrangement for the data concerned.
What it means
A project may agree that its training data must be complete, correctly labeled and available on time. Those objectives are not yet a plan. The plan explains what work will be done, by whom, with which resources and how results will be assessed.
The distinction from data quality planning is simple: planning is the activity of deciding; the plan is its documented outcome. A useful plan connects to a particular dataset, life-cycle stage and intended use. It should be detailed enough to guide work without becoming a document that nobody maintains.
How it works
Identify the quality objectives and the requirements they imply. Specify the checks, reviews, collection improvements or remediation activities needed to meet them. Assign responsibilities and allocate time, people, tools and budget.
Define how exceptions are recorded and resolved, what evidence is retained and which conditions require escalation or reassessment. Include the relevant data versions and dependencies so that teams know what the plan covers. For continuously arriving data, describe ongoing monitoring rather than only a one-time cleanup.
Review progress against the plan and revise it when circumstances change. The exact format can vary, but the connection between objectives, activities, resources and evidence should remain clear. These practical elements are explanatory guidance, not a claim that a particular template satisfies every applicable standard.
Three practical examples
- Training-data preparation: A team schedules label review, duplicate detection and coverage analysis before a dataset is approved for model development.
- Supplier feed management: A data owner defines arrival-time checks, incident contacts and a process for handling missing daily files.
- Dataset migration: Engineers plan reconciliation between old and new systems, including record counts, field mappings and review of exceptions before the migrated data is used.
Think of it this way
Think of a maintenance plan for a shared building: it identifies inspections, responsibilities, resources and follow-up work. The plan itself does not fix a broken system. Likewise, a data quality plan is valuable only when its activities are performed and its findings influence decisions.
Worked example
A fictional project has twelve thousand labeled records and an objective to assess label accuracy before training. Its plan allocates two reviewers to independently inspect a selected sample, an adjudicator to resolve disagreements and time to investigate recurring error patterns.
Suppose each reviewer can inspect one hundred records per hour and the chosen sample contains six hundred records. Independent review requires twelve reviewer-hours in total: six hours for each reviewer. Adjudication and preparation require additional time.
The calculation helps resource the plan, but it does not establish that six hundred records is a statistically sufficient sample. Sample design must match the claim, population and uncertainty the project needs to assess.
Illustrative case study
Fictional case: Atlas Forms prepares documents for an extraction model. Its first schedule includes model training but no budget for checking labels. When reviewers discover inconsistent date annotations, the team cannot resolve them within the planned release window.
The revised data quality plan assigns an annotation owner, reserves adjudication time and requires a documented decision on unresolved categories. Training can proceed only with an explicit understanding of those limitations. This invented case demonstrates resource planning, not a real project's delay or recovery.
Common misconceptions and mistakes
A data quality plan is not merely a list of desired percentages. Objectives need associated activities and resources. It is also not a retrospective report saying what happened; reports provide evidence about execution.
Do not assume automated checks cover every quality concern. Source accuracy, ambiguous labels and population coverage may require human judgment. A plan copied from another project may miss important differences in data, users or consequences. Finally, completing planned tasks does not automatically mean objectives were achieved: the results may show that more work or a different use decision is needed.
Business use
A good plan makes the cost of quality work visible before model development depends on the data. It clarifies ownership at handovers and helps managers decide whether to improve collection, repair existing records, narrow the application or defer use.
Keep it proportionate to the project and link it to operational schedules. Track both completion of activities and the quality evidence they produce. When requirements change, update resource allocations and acceptance decisions rather than silently maintaining a plan designed for an earlier purpose.
People also ask
Is a data quality plan the same as a quality policy?
No. A policy gives broader direction; a plan specifies how objectives will be achieved for a defined scope.
Can a small project use a short plan?
Yes. The plan should be proportionate while still making responsibilities, work and evidence clear.
Does finishing every task prove the data is suitable?
No. The assessment results must still be compared with the intended-use requirements.
Check your understanding
A project lists a target of 99% completeness but assigns no owner, checks or remediation effort. Is that an actionable quality plan?
Answer: No. It is an objective without the specified work and resources needed to pursue it.
Related terms
- Data quality planning
- Remediation
- data quality objective
- data quality management
- acceptance criterion
- data steward
- quality evidence
- dataset release
Learn next
Take one data quality objective and write a one-page plan linking it to an activity, owner, resource estimate, evidence output and review point. Include one likely exception and how it will be handled. Check that the plan distinguishes completing work from meeting the objective.
More in Governance, ethics and safety
- Data quality claim - A data quality claim states how well data meets a specified quality requirement.
- Data quality measure - A data quality measure is a variable used to express a measured aspect of data quality.
- Data quality model - A data quality model organizes the characteristics used to define and evaluate data quality.
- Data record - A data record is a collection of related data items handled as one unit.
- Data security - Data security is the protection of data against unauthorized access, disclosure, alteration, destruction, and loss of availability.
- Debiasing - Debiasing is the process of identifying and reducing a specified unwanted bias in data, representations, models, or their use.
Sources
ITU-T Y Supplement 97 (11/2025), term 3.1.35, 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 5259-3. ISO publisher overview consulted: https://www.iso.org/standard/81092.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 model Next: Data quality planning