Completeness compares what is present with what was expected for a stated purpose. Define the expected records and required fields before calculating a percentage; a fully populated dataset can still contain inaccurate information.
Establish the reporting boundary first
Completeness needs a denominator that someone can explain. A collection of 900 records might be complete for one branch and incomplete for an entire organisation. Start by identifying the population, time period and events that should have produced records. Use an independent expectation where possible, such as an agreed roster or a source-system transaction count.
Do not let the dataset define its own expected size without examining the collection process. If missing records disappear before an export is created, counting the rows in that export cannot reveal them. Record how the expected population was established and any uncertainty around that expectation.
Distinguish absent records from absent fields
A record-level check asks whether an expected case is present. A field-level check asks whether a required value appears in a present record. These checks answer different questions and should not be collapsed into a single quality label.
Consider which fields are required for the intended use. A telephone number may be optional in one workflow while a transaction date is essential for a period-based total. Mark legitimate inapplicability explicitly. Otherwise a missing-value rule may penalise correct records or encourage people to enter meaningless text just to satisfy validation.
Look for concentrated gaps
Break down missing records or values by source, period and relevant operational category. A reassuring overall percentage can hide a complete failure in a small but important part of the population. Review both the size of the gap and its possible effect on the decision.
A pattern does not prove its cause. Missing values from a new branch might result from a delayed submission, a different export format or a collection practice that was never agreed. Ask the source owner to investigate before choosing an automatic correction. Retain the unresolved cases so the question remains visible.
Make the publication decision explicit
Not every gap prevents every use of a dataset. The report owner should consider what could change if the absent information were available. If the missing group is relevant to the decision, a qualified or delayed report may be more appropriate than an apparently complete headline.
A completeness note should state the expected population, observed population, important missing fields and known reasons for absence. Avoid claiming that a high completeness score establishes accuracy. Presence does not prove that an entered date, category or amount matches the real event.
Improve the collection process
Give the issue an owner and a specific follow-up. That may involve clarifying required fields, reconciling an export against a source count or making late submissions visible. Choose a change that addresses the observed gap rather than adding a general instruction to enter better data.
Repeat the check after the change using comparable definitions. Preserve enough history to distinguish a genuine collection improvement from a changed denominator. A quality measure should help people understand the process, not merely make a dashboard turn green.
Illustrative worked example
A fictional scenario to explain the method, not a customer result or a market benchmark.
An illustrative service expects a weekly submission from twenty locations. Eighteen submit a file. Across those files, 360 of 400 expected activity records appear, and 342 of the present records include the required completion date. These are three separate observations: location coverage, record coverage and date-field completion.
The team discovers that two locations have not submitted and one submitted file excluded cancelled activities that the reporting definition still requires. It avoids describing the export as 95% complete using a convenient denominator. The publication note explains which locations are represented and why one part of the population is still uncertain.
After the source owners clarify the export rule, the report is rebuilt and the counts reconciled. The figures are fictional and demonstrate the distinction between levels of completeness; they are not a recommended acceptance threshold or a benchmark for another organisation.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Define the expected population independently where possible.
- Separate record coverage from required-field completion.
- Inspect where missing information is concentrated.
- Publish the limitation and assign a source-process follow-up.
Common questions.
Is a dataset with every field filled in necessarily accurate?
No. Completeness concerns presence. Values can be populated but incorrect, misleading or entered only to satisfy a required-field rule.
Should every incomplete report be withheld?
Not automatically. Consider the intended decision and the possible effect of what is missing, then make an explicit publication choice with an appropriate limitation note.
Further reading & source context.
These primary references provide background. They are not evidence for the fictional example above and do not endorse Accuralytix.
Found an error or a missing qualification? Send an editorial correction.
Check my data readiness.
Use a practical tool to turn this idea into a clearer starting point.
Check my data readiness