The Business Changed. Your Validation Rules Didn’t.

A validation rule can work exactly as designed and still be wrong.
The logic may be correct.
The code may run without errors.
The control may detect exactly what it was built to detect.
And yet the rule may no longer reflect the business.
That is one of the reasons validation rules should never be treated as something that is designed once and then left untouched forever.
Businesses change.
Processes change.
Definitions change.
Systems change.
And when that happens, the rules that control the data need to change with them.
The first person who should normally recognize this is the Data Owner.
Data Governance cannot be expected to understand every business process in enough detail to immediately notice every operational change.
That is not its role.
The Data Owner is much closer to the business reality.
In many organizations, the Data Owner is also the manager of the Data Steward or at least directly connected to the team responsible for maintaining the data.
That makes the Data Owner the natural first signal.
“The business has changed.”
“This rule no longer reflects what we actually do.”
Once that signal appears, Data Governance should react immediately.
But the first step is not automatically switching the rule off.
The first step is understanding why it has become outdated.
What changed?
When did it change?
Is the change temporary or permanent?
Which data points are affected?
Which downstream reports or KPIs depend on them?
And does the existing control still protect any relevant business risk?
The answer determines what happens next.
Sometimes the business changes completely.
A process moves in a completely different direction and the old validation logic no longer makes sense.
In that situation, keeping the old rule alive may create more noise than value.
It can be switched off while the replacement is designed.
But many changes are not that dramatic.
Sometimes the business logic evolves rather than disappears.
In those cases, I prefer running the old and new validation rules in parallel for a period of time.
A quarter can be useful.
That comparison should go far beyond counting which rule produces fewer errors.
I would look at the number of flagged records.
The number of false positives.
The types of records each rule catches.
And, most importantly, the business impact of the differences.
A new rule may look much better statistically.
It may generate fewer alerts.
It may have fewer false positives.
But if it misses a small number of critical cases with significant business impact, it is not a better rule.
It is worse.
Data Quality controls should not be optimized for attractive statistics.
They should be optimized for the risks they are supposed to control.
And changing the validation logic is never only about changing the rule itself.
If a business definition changes, the surrounding data ecosystem needs to change with it.
The Business Glossary needs to be updated.
Documentation needs to be updated.
Data Lineage needs to reflect the new logic.
Downstream reports and KPIs need to be checked.
Relevant users need to understand the change.
Until all of that is aligned, the new validation rule should not move into production.
Otherwise, the organization ends up with a new rule controlling data according to one definition while reports, documentation or users are still operating according to another.
That creates a different kind of Data Quality problem.
Validation rules also need versioning.
A new rule should not simply overwrite the old one and erase its history.
The organization should be able to see which validation logic was active during which period.
Rule version 1.
Valid from one date until another.
Rule version 2.
Effective from the next date.
Along with the reason for the change, what was tested and what the parallel comparison showed.
This also means that historical data should normally remain evaluated according to the rule that was valid at the time.
If a rule changes in April, January data does not automatically become wrong because it would fail the April logic.
The business rules were different then.
Historical revalidation should happen only when there is a specific reason to apply the new standard retrospectively.
Every new rule should also receive increased attention when it first goes live.
For the first month, I would monitor it more closely than a mature rule.
Too many false positives are a warning.
Missing critical cases is a warning.
Business users saying that the control does not reflect the real process is a warning.
Unexpected impact on downstream reporting is a warning.
Any of these may justify sending the rule back for refinement.
After that initial monitoring period, the rule moves into its normal review cycle.
For many rules, a six-month review is a sensible standard.
But the lifecycle should ultimately be risk-based.
More critical controls may deserve more frequent attention.
Lower-risk rules may need less.
The risk classification itself should not be decided by Data Governance in isolation.
It should be defined together with Data Owners and the business users who understand the actual impact if the control fails.
There should also be another trigger for review that has nothing to do with the calendar.
The Data Owner should continuously monitor the business.
If the process changes tomorrow, there is no reason to wait five months for the next scheduled review.
The rule should be reviewed when the business changes.
That is the real lifecycle.
Regular review when nothing changes.
Immediate review when something does.
A validation rule is not valuable because it has been running for years.
It is valuable because it still controls the right risk today.
The business may change.
The system may change.
The definition may change.
And the rule must be allowed to change as well.
Because a validation rule can be technically perfect and still be wrong if the business it was designed for no longer exists.



Comments