Data quality framework
Definition
Data quality framework
Data quality framework is the agreed set of dimensions, thresholds and owners that decides whether data is fit for use. It tells you when data is good enough to act on, and it names the person who fixes a measure that slips.
The framework stops at the threshold. It says when data is good enough — it never says what to do with the data, which is a question for a strategy, a product decision or a roadmap.
Most frameworks use the same handful of dimensions: completeness, accuracy, validity, timeliness, uniqueness and consistency. The argument is rarely about that list — it is almost always about where each threshold should sit.
Key takeaways
- A data quality framework defines dimensions, thresholds and owners, then measures against them.
- It says when data is good enough to use; it never says what the business should do with it.
- Thresholds come from the decision the data supports, not from what the data happens to allow.
- Without a named owner per dimension, a quality report becomes a complaint nobody has to answer.
How it works
A data quality framework works by turning a vague worry into measurable dimensions with owners attached. You profile the data, agree a threshold per dimension, publish the misses, assign the fix, then measure again on a fixed cycle.
| Dimension | The question it asks | Who usually owns it |
|---|---|---|
| Completeness | Are the required fields populated? | The team capturing the record |
| Accuracy | Does the value match the real world? | The business owner of the field |
| Validity | Does the value fit the agreed format? | The system owner |
| Timeliness | Is the record current enough to act on? | The pipeline owner |
| Uniqueness | Is the same entity recorded twice? | The master data owner |
| Consistency | Does it agree across systems? | The data architect |
Someone has to run the measurement. A data quality analyst profiles the tables, builds the rule set and reports the misses to the owners who can actually act on them.
The vocabulary is borrowed from manufacturing. Defect density counts faults per unit of output, and data teams apply it per thousand records rather than per software module.
The plainest measure is often the most useful. An error rate per thousand records tells a business owner more than a dashboard carrying twelve weighted indices ever will.
Testing practice transfers too. Quality assurance methods built for software give data teams sampling plans, regression checks and a workable definition of done.
Thresholds come from the use, not the data. Mapping critical to quality characteristics forces the business to say which fields change a decision and which are merely nice to hold.
Profiling comes before targets. Run the measures for a quarter with no threshold attached, because a number set before you know the baseline is a negotiating position — not a standard.
Measurement without consequence is theatre. Each threshold needs an owner who hears about a break, a route to the fix, and a review that notices repeat offenders — the same field failing monthly is a process problem.
Governments publish their own versions. The UK’s Government Data Quality Framework, issued on 3 December 2020, states that civil servants are responsible for establishing whether data is fit for purpose.
Industry has a scoring model too. The EDM Council’s Data Management Capability Assessment Model (DCAM) is pitched as the global standard framework for managing data, and version 3 adds cloud and AI coverage.
Examples
Frameworks look similar on paper and very different in use. What changes is which dimension carries the weight, because the cost of a miss differs sharply between a statistical release, a credit file and a product catalogue.
Government statistics. The UK Government Data Quality Hub published its framework in December 2020 so departments could assess, document and improve quality using one shared approach.
Banking. Credit risk teams test completeness and timeliness on counterparty records before a regulatory return goes out, because a late or missing field becomes a reportable error. The threshold is set by the return, not by the database.
Healthcare. Hospital groups police uniqueness on patient identifiers, since two records for one person splits a clinical history in half. Merge rules are reviewed by clinicians rather than by the data team.
Retail supply chains. Product catalogues are tested for validity and consistency across marketplaces, because a mismatched size or unit arrives back as a return rather than as a data defect.
Related terms
These terms sit beside a quality framework without doing its job. Two name the roles and targets the framework depends on, and three supply the measures and methods it borrows from manufacturing and software testing.
- Database administrator: the custodian of the systems where the measured records actually live.
- Zero defect rate: the target some processes set when any defect is unacceptable.
- Data quality analyst: the person who profiles data and reports misses against the thresholds.
- Error rate: the simplest measure of how often records are wrong.
- Critical to quality: the characteristics that decide whether an output satisfies its user.
FAQ
What is a data quality framework?
It is the agreed set of dimensions, thresholds and owners that decides whether data is fit for use. It covers what gets measured, the level each measure must clear, and who fixes a miss.
What dimensions should it include?
Completeness, accuracy, validity, timeliness, uniqueness and consistency cover most needs. Add a dimension only when a real decision depends on it, because every extra measure costs attention. Six dimensions handled well beats twelve tracked badly.
How is it different from a data strategy?
The framework says when data is good enough to act on. A strategy or roadmap decides what the business does with the data once it clears the bar.
Who owns data quality?
The business owner of each field owns the number, and a data quality analyst owns the measurement. Quality collapses when both jobs land on the same overloaded team.
How often should thresholds be reviewed?
Review them whenever the decision the data supports changes, and at least once a year. A threshold nobody has revisited usually reflects an old system rather than a current need.
Compare providers running data quality work as a managed service across the Outsource Accelerator network before building the function in house.
Independent
