TL;DR
Key takeaways
- Data quality dimensions are the measurable characteristics used to judge data: accuracy, completeness, consistency, timeliness, validity and uniqueness
- Each dimension fails in a different way, so each needs a different check, such as null checks for completeness and range checks for validity
- How good is good enough depends on the use: supply chain may accept 98% accuracy for directional analysis while finance needs a 100% invoice match
- Checks on a single dataset are not enough, because cross-system consistency needs reconciliation against the source
- Quality should be monitored continuously, with thresholds and alerts, instead of cleaned up once
A supply chain team may accept 98% accuracy for directional analysis, while finance needs a 100% match on invoices.
Data quality dimensions are the measurable characteristics used to judge whether data is fit for use. The six most common are accuracy, completeness, consistency, timeliness, validity and uniqueness. Each describes a different way data can go wrong, and each needs its own kind of check.
This guide explains each dimension with an example, how to check it, and how to decide how much quality is enough.
What are the data quality dimensions?
|
Dimension |
Question it answers |
Example of a failure |
Typical check |
|
Accuracy |
Does the data reflect reality? |
A customer's address is wrong |
Compare with a trusted source |
|
Completeness |
Is anything missing? |
Material number is blank |
Null and blank checks |
|
Consistency |
Do systems agree? |
Revenue differs between billing and the warehouse |
Cross-system reconciliation |
|
Timeliness |
Is it current? |
Yesterday's load never arrived |
Freshness and delay checks |
|
Validity |
Does it follow the rules? |
An email without an @, or a negative quantity |
Format, range and enumeration checks |
|
Uniqueness |
Is each thing recorded once? |
The same supplier appears three times |
Duplicate detection and matching |
Some teams add other dimensions, such as integrity and relevance. The six above are the most widely used.
What does each dimension mean in practice?
Accuracy
Accuracy means the value matches the real-world fact. It is the hardest to test automatically, because you need a trusted source to compare against. Reconciliation against a system of record is the usual method.
Completeness
Completeness means required values are present. A blank required field, such as a missing Material Number in inventory, breaks downstream processes. Null and blank checks catch it.
Consistency
Consistency means the same fact has the same value across systems and datasets. It cannot be checked inside one table. It needs comparison across systems, which is what automated data reconciliation does.
Timeliness
Timeliness means data is available when it is needed and is up to date. Pipelines can look healthy while data is late or stale, so freshness and volume checks matter.
Validity
Validity means data follows the expected format, range and allowed values. Typical checks include date formats, numeric ranges and a fixed list of allowed codes.
Uniqueness
Uniqueness means each real-world entity is recorded once. Duplicate customers or suppliers inflate counts and split history. Matching and golden records address it.
How good does data quality need to be?
There is no single number for perfect data. The right level depends on the use. A supply chain team may accept 98% accuracy for directional analysis, while finance needs a 100% match on invoices. Set a threshold for each dimension and each important dataset, and agree who owns it.
Small issues are tolerable until they grow. For example, 200 blank attributes out of 2 million records may be acceptable, but not if the count doubles. Alert on threshold breaches so you notice the change. Read why there is no perfect data, only better data.
How do you measure and monitor data quality?
- Pick the dimensions that matter for each critical dataset.
- Write checks for each one: null, distinct, range, enumeration, date format, duplicates and reconciliation.
- Set thresholds and owners, so a failure goes to someone who can fix it.
- Monitor continuously. Run checks after every load, and use anomaly detection to catch gradual drift that fixed rules miss.
- Trace failures to their cause and fix the source, not only the symptom.
Fixed rules alone miss slow decline, such as a missing-value rate creeping up over months. Combine rules with historical baselines and anomaly detection. See how data observability and data quality work together and the data quality and observability platform. The knowledge hub goes deeper, or you can book a demo.
Frequently Asked Questions
What are the dimensions of data quality?
The six most common data quality dimensions are accuracy, completeness, consistency, timeliness, validity and uniqueness. Each describes a different way data can be good or bad.
What is the difference between accuracy and validity?
Validity means the value follows the expected format or range. Accuracy means it is actually true. An email address can be valid in format and still be wrong for that customer.
How do you measure data quality?
Write a check for each dimension, such as null, range, format, duplicate and reconciliation checks, set thresholds and owners, and monitor the results after every data load.
Which data quality dimension matters most?
It depends on the use. Finance usually prioritizes accuracy and consistency, supply chain often cares about timeliness and completeness, and customer data often needs uniqueness.
How is data quality different from data observability?
Data quality validates data against rules such as accuracy and completeness. Data observability monitors the health of the whole pipeline over time through freshness, volume and schema signals. They work best together.
Can data quality checks be automated?
Yes. Predefined checks for nulls, distinct values, ranges, enumerations and date formats can run after every load, and anomaly detection can flag unusual changes without a hand-written rule for each case.