Data Quality
What a Data Quality Rule Needs to Say to Be Auditable
A rule that only says 'field is blank' cannot be audited. An auditable rule says what, why, how severe, at what grain and who fixes it.
Dave Welensky, Director · Published November 2026 · 5 min read
Most data quality rules are written to run, not to be explained. When an auditor, a steering committee or a data owner asks why a check exists and what a failure costs, the rule has to answer. Identity, logic, severity, business reason, fix, grain and state are what turn a check into a control.
The rule everyone writes first
The first rule on almost every data quality programme is 'customer name is blank'. It is easy to write, easy to test and easy to approve. It also cannot answer the questions that matter once the number goes in front of someone senior. Why does this check exist? What breaks downstream when it fails? How serious is a failure? Who corrects it, and where?
A check that cannot answer those questions produces a count. A rule that can produces evidence. The difference is what makes a data quality result something an auditor, a steering committee or a data owner can act on instead of argue with.
What an auditable rule says
Identity: a stable identifier and a plain-language name, so a finding can be traced to the same rule across every run. Logic: the exact test, expressed in a form the business can read, whether that is a blank check, a list of allowed values, a cross-field condition or a duplicate test.
Severity: assigned by policy, not inferred, so the same failure is never serious on Monday and minor on Tuesday. Business reason: why the check exists and what it protects, written for the person who will read the finding. Fix: what the data owner does about it, including the SAP transaction where the correction is made.
In Ixia DQ every rule carries these attributes, and the failure file for a domain names the record, the field, the current value and the transaction to correct it. A finding that cannot be traced to a named rule, a reason and a fix is a number, not evidence.
Grain: the quiet way scorecards go wrong
Extracts are usually denormalised. One material can sit on eleven plant rows. A rule that does not declare its reporting grain will count eleven failures for one defect, and the scorecard will look eleven times worse than the data is. The first time someone notices, the whole measurement loses credibility.
An auditable rule declares the grain it reports at. One material is one failure, however many rows it appears on. The count then matches how the business thinks about the problem.
State: say so when a rule cannot run
Rules are not only enabled or disabled. A rule can be retired, kept for history but no longer run. And a rule can be blocked: its data is not available in this extract, so it cannot be evaluated.
A blocked rule is the important case. If it quietly passes, the scorecard overstates quality. If it quietly fails, the scorecard understates it. The honest behaviour is to report the blocker as a blocker and keep it out of the failure count. Anything else mixes missing evidence with bad data.
Rules across objects, in the business's words
Some of the most valuable rules are not about a single field. A material extended to a sales area must have a delivery plant. A payment term must be allowed for the company code. A transport zone must be valid for the customer's country. Each record can be valid alone and invalid as a set, which is where object-by-object checking quietly fails.
These rules should be written in business terms, by or with the people who own the data, in a tool that does not need a developer for every change. Where the standard logic types cannot express a rule, a custom function covers it, but that should be the exception, because the business can read the standard rules and cannot read the code.
Key takeaways
- A check produces a count. An auditable rule produces evidence.
- Every rule needs an identity, logic, severity, business reason and fix.
- Declare the reporting grain, or one defect is counted many times.
- Report blocked rules as blocked. Never let missing evidence pass or fail silently.
- Write cross-object rules in business terms, so the data owner can read them.
Recognise your programme in this?
Written by the people who deliver the work. The conversation works the same way.
