Article 8 of 10 · Data & Decision Intelligence · 10 min read
Data Quality: The Silent Tax on Every Decision
You can agree on what every number means and still make the wrong call — because the data feeding that number was quietly wrong. Bad data doesn't announce itself. It renders beautifully, reads confidently, and charges a tax on every decision it touches.
A retail company decides to double down on its best-performing customer segment — mid-sized business buyers in the north. The dashboard is clear: that segment grew 34% last year. Budget shifts, a salesperson is reassigned, marketing spend follows the data. Six months later the return is a fraction of what was modelled, and no one can say why.
The why was in the data all along. The "34% growth" was inflated by duplicate customer records — the same twelve companies entered two or three times, under slightly different names, each counted as new. The segment hadn't grown by a third. It had grown by maybe 8%. The decision wasn't wrong because the logic was bad or the definition was unclear. It was wrong because the numbers underneath were dirty, and nothing on the screen said so.
In the last article we solved the single source of truth problem — getting everyone to agree on what a number means. This is the next layer down, and it's unforgiving: even one agreed definition, calculated in one governed place, is only as trustworthy as the data flowing into it. Garbage in, gospel out — the report looks authoritative no matter how flawed its inputs.
Why bad data is invisible
A broken machine makes noise. Broken data is silent. A chart built on duplicated, stale, or mislabelled records looks identical to a chart built on perfect data — same colours, same confident lines, same polished export. There's no error message, no red flag. The failure only surfaces later, downstream, disguised as a decision that didn't work out. By then it's nearly impossible to trace back to its cause.
A broken machine makes noise. Broken data renders beautifully — and you only meet it later, wearing the mask of a decision that failed.
That invisibility is exactly why it's a tax rather than a one-off cost. You pay it whether you notice or not: in decisions quietly skewed, in analysts spending 60–70% of their time cleaning data instead of interpreting it, in the slow erosion of trust once people learn a report can't be taken at face value. It's a levy on every decision the data touches, collected in the dark.
The six ways data goes bad
"Bad data" isn't one thing. It fails in six recognisable ways, and naming them is the first step to catching them.
The six dimensions
- Completeness — Missing values. Half your customers have no region recorded, so any regional analysis silently ignores them.
- Accuracy — Values that are simply wrong. A price typed as 100 instead of 1,000 doesn't look wrong in a table — until it moves an average.
- Consistency — "RO", "Romania", "România", "ROU" — the same thing recorded four ways, so one country becomes four rows.
- Timeliness — Data that's true but out of date. Last quarter's stock levels driving this week's reorder decision.
- Uniqueness — Duplicates. The same customer, invoice, or transaction counted more than once — the north-region trap above.
- Validity — Values that break the rules. A delivery date before the order date; a negative quantity; an email with no "@".
Each one is mundane on its own. Together they're why a report can be built with perfect logic on a flawless definition and still mislead the person reading it.
Where the mess comes from
Dirty data is rarely anyone's fault in particular — it accumulates from ordinary operations. Manual entry with no validation at the point of capture lets typos and blanks straight in. Multiple systems holding overlapping records drift apart the moment one is updated and the other isn't. Integrations between systems create seams where fields get truncated, misaligned, or silently dropped. And time itself degrades data — a customer moves, a supplier closes, a price changes, and the record keeps insisting on the old truth. None of these is dramatic. All of them compound.
The economics: the 1–10–100 rule
There's a well-worn rule of thumb in data management, and it's worth internalising because it dictates where to spend effort. It costs roughly €1 to prevent a bad record at the point of entry, €10 to find and fix it later in cleaning, and €100 once a decision has already been made on it.
The numbers are illustrative, but the ratio is the point: prevention is an order of magnitude cheaper than correction, and two orders cheaper than the wrong decision. This is why "we'll clean it in the report" is a losing strategy. You're choosing to operate permanently at the €10 tier, and occasionally paying €100, to avoid a €1 fix at the source.
What to actually do
1. Fix it at the source, not in the report
The single highest-leverage move. Every fix applied downstream — a filter in a dashboard, a cleanup script, a corrected spreadsheet — has to be redone every time the data refreshes, and only fixes the one report that person happens to touch. Fix the record where it lives and every report improves at once.
2. Validate at the point of entry
The cheapest data quality control is the one that stops a bad value from ever being saved. Required fields, dropdown lists instead of free text, format checks on emails and dates, range limits on quantities and prices. Most dirty data is preventable at the keyboard.
3. Deduplicate — and keep it deduplicated
Duplicates are the most common and most decision-distorting fault. Match on the fields that actually identify an entity, merge deliberately, and put a check in place so the same duplicates don't quietly return next quarter.
4. Profile your data regularly
You can't fix what you don't measure. Periodically scan the key tables for the six faults above — how many blanks, how many duplicates, how many invalid values. A short monthly data-quality check surfaces drift before it reaches a decision.
5. Give the important data an owner
Data with no owner degrades by default. Assign responsibility for the quality of each critical dataset to a named person, the same way you'd assign responsibility for the number itself. Ownership is what keeps quality from being everyone's job and therefore no one's.
The payoff
Clean data isn't glamorous and it never wins a demo. But it's the difference between a decision-support system that earns trust and one that quietly betrays it. Companies that treat data quality as ongoing maintenance — not a one-time project — stop paying the silent tax. Their analysts interpret instead of scrub, their reports get acted on without a second guess, and their decisions rest on ground that actually holds.
Find out how clean your data really is.
Upload a dataset into PrismIQ and an AI agent profiles it for you — flagging blanks, duplicates, inconsistent labels, and out-of-range values before they reach a decision — then produces a clear report in business language, no technical skills required.
One Number, Three Answers: The Single Source of Truth Problem · Why Data-Driven Cultures Fail