Six Tests of Data Quality
Meera once made a decision from a dashboard that turned out to be wrong: she moved staff to Kolkata because on-time looked terrible there, and later learned the courier feed had been a week out of date. She did not ask for "better data" afterwards. She asked a sharper question: how do we know a number is good enough to act on? Data quality gives that question six concrete tests.
The rule
Fit for the decision: relevant, timely, accessible, interpretable, accurate, coherent.
Quality is not perfection; it is fitness for a use. The same data can pass for a yearly strategy review and fail for this morning's shift plan. Each test is a question:
| Test | The question | A failure at Weftwise |
|---|---|---|
| Relevant | Does it measure what the decision needs? | "On-time" was measured at dispatch, but customers care about arrival |
| Timely | Is it fresh enough for the decision's clock? | The courier feed updated weekly; shift leads plan every morning |
| Accessible | Can the right people reach it, and only them? | The file lived on Ishaan's laptop, while the salary sheet was open to all |
| Interpretable | Will readers understand what it means? | A column called "OT%", with no definition and no units |
| Accurate | Are the values correct? | 30 deliveries were dated before they were dispatched |
| Coherent | Does it agree with itself, other sources and past periods? | The dashboard shows 960 delivered orders; Finance's ledger shows 955 |
The first four are about use: the right data, in time, reaching and making sense to the right people. The last two are about truth: correct values that tell one consistent story. A dataset can be perfectly accurate and still useless because it arrives too late, and a beautifully timely feed can be wrong.
The shortcut
Run the six questions in the order a decision meets the data: Is it the right thing? Is it fresh? Can I get it? Can I read it? Is it true? Does it agree?
Right thing?
On-time was measured by dispatch date. Changed to "delivered by the date promised at checkout", because that is what customers feel.
Drills
Answer fast — trust the rule.
Some orders show a delivery date earlier than the order date. Which test fails?
The courier data arrives a week late, but shift leads plan each morning. Which test fails?
Nobody can say whether the dashboard's "Revenue" includes GST. Which test fails?
Marketing judges customer happiness by the number of emails it sent. Which test fails?
The dashboard's March revenue differs from Finance's ledger for the same month. Which test fails?
Remember
- Fit for the decision: relevant, timely, accessible, interpretable, accurate, coherent.
- Ask in order: right thing, fresh, reachable, readable, true, agreeing.
- Trap: treating quality as perfection instead of fitness for a specific use.