Data-quality dashboards can provide reassuring precision. They tell us that a dataset is 94% complete, 91% valid, or 86% accurate. They classify issues according to familiar dimensions such as completeness, validity, accuracy, consistency, timeliness, and uniqueness.
These dimensions are valuable. They help organizations identify how data is defective and establish a common language for monitoring it. DAMA UK similarly describes data-quality dimensions as indicators for measuring and communicating data quality. The problem begins when these metrics are treated as the ultimate measure of data quality.
What does an 86% data-quality score actually mean for the business? Does it indicate that production is at risk? That invoices cannot be processed? That procurement decisions are based on unreliable information? Or does it simply mean that thousands of optional fields, which are rarely used, remain empty?
A percentage can tell us how many records failed a rule. It cannot tell us whether those failures matter.
When the average hides the real problem
Imagine two datasets.
The first contains 1,000 invalid values. Perhaps secondary contact details are missing, descriptions use inconsistent abbreviations, or optional classification fields have not been maintained. The data-quality dashboard turns red. The number of open issues looks alarming. Yet the immediate effect on the organization may be limited.
The second dataset contains only one invalid value. Statistically, it is almost invisible. Operationally, it is catastrophic. An incorrect planning parameter prevents a critical component from being replenished. The material is missing when production starts, the line stops, and every hour of downtime adds further cost. The dashboard may still show 99% data quality, but the business is no longer producing.
Which dataset has the greater data-quality problem?
If we look only at failed records, the answer appears to be the first. If we look at operational impact, the answer may clearly be the second.
This is the weakness of aggregated data-quality metrics. They treat defects as if their importance were distributed evenly across a dataset. In reality, data risk is highly concentrated. Thousands of minor issues can create less exposure than one defect at a critical point in a business process.
The average does not necessarily reveal this. In some cases, it conceals it.
A valid value can still be dangerous
Technical dimensions also tell only part of the story at the individual record level.
A supplier bank account may be complete, correctly formatted, and technically valid while still being incorrect. A material classification may conform perfectly to the required format but lead to the wrong purchasing, customs, or planning decision. A delivery address may pass every validation check and still refer to the wrong location.
Conversely, an empty or technically inconsistent value may have no meaningful consequence if the field is not used in a relevant process.
This does not make validity or completeness unimportant. It shows that data quality cannot be assessed independently of its purpose. The same defect can be insignificant in one context and critical in another. Its importance depends on the process it supports, the decision it influences, and the consequence if it is wrong.
Traditional dimensions answer an important diagnostic question: In what way is the data defective?
The business needs an additional answer: What could happen because of this defect?
Without that second question, organizations risk creating highly sophisticated measurements of the wrong problems.
The hidden cost of treating every issue equally
When dashboards prioritize issues primarily by volume, data teams and business users can spend significant effort correcting large groups of visible but low-impact defects. Meanwhile, less frequent issues that affect production, payments, compliance, customer deliveries, or an ERP transformation may remain buried in the same backlog.
This creates a second problem. Employees begin to see data quality as an endless cleansing exercise rather than a way to protect and improve business performance.
A dashboard lists thousands of errors. Business users are asked to correct them. However, nobody can clearly explain what the issues are costing, which processes depend on them, or why one task should take priority over another. Over time, attention declines because every issue appears urgent and none appears meaningfully connected to an outcome.
The answer is not to stop measuring completeness, validity, or accuracy. The answer is to connect those measurements to business context.
From defect volume to business exposure
A more meaningful data-quality dashboard would still show the number and type of failed records. But it would also show which business process is affected, the potential operational or financial consequence, the effort required for correction, and the person accountable for the decision.
This does not mean that every data issue needs a detailed financial business case. The level of analysis should be proportionate to the potential impact. For many issues, a simple classification such as low, medium, or high business criticality may be sufficient. Important issues can receive a more detailed estimate of cost, risk, or operational disruption. For the most critical cases, organizations may use process analysis or simulations to understand the potential consequence more precisely.
The purpose is not to calculate a perfect monetary value for every incorrect field. It is to make better prioritization decisions.
Once this context is available, a lower data-quality percentage does not automatically mean greater urgency. A high percentage does not automatically mean that the business is protected. Management can instead focus resources where unreliable data creates the greatest exposure.
The goal is not perfect data
Organizations will never correct every data issue. Nor should they.
Some defects are too minor to justify the effort. Some fields are no longer relevant. Some inconsistencies can be tolerated without affecting a process or decision. Pursuing perfect data across every system and attribute is usually unrealistic and economically questionable.
The more useful objective is reliable data wherever the business cannot afford failure. That requires a shift in how data quality is communicated. Instead of asking only, “How complete is our data?” organizations should ask, “Which missing values could interrupt an important process?” Instead of reporting only the number of invalid records, they should identify which invalid records create financial, operational, regulatory, or reputational exposure.
Completeness, validity, accuracy, consistency, timeliness, and uniqueness remain relevant in many cases. But they are diagnostic dimensions, not business outcomes.
A data-quality score tells us how much data appears to be wrong. Business-impact and risk metrics tell us what to do about it.
Because when one incorrect record can stop production, an average of 99% is not reassuring. It is irrelevant.