Who owns data quality?

Most companies have a data-quality program of some kind — tests, a catalog, a monitoring tool, maybe a dashboard that grades freshness and completeness. What they usually do not have is an answer to a much simpler question: when a number is wrong, who is responsible for making it right — not just this once, but so it stops happening? Data touches everyone. Engineering moves it, analysts shape it, operations enters it, leadership decides on it. And because everyone touches it, no one owns it. Quality that is everyone's job is no one's job.

There is a one-question test for the gap. Pick a number the business actually runs on — pipeline, churn, margin, headcount — and ask who is accountable for its correctness. Not who built the report. Not who would scramble to fix it during a fire. Who owns the definition, the standard for the data underneath it, and the authority to say 'this is the number.' If the honest answer is a shrug, or a committee, or 'it depends,' you have found the real reason the quality program is not working. The ownership gap almost always shows up in one of six forms:

1. 1.Engineering owns the pipes, not the meaning. The data team moves data reliably and on time, and is measured on uptime and latency — not on whether the numbers reflect what actually happened in the business. They can deliver a wrong number perfectly.
2. 2.Analysts own the report, not the source. They clean, caveat, and reconcile downstream because they cannot change what arrives. Every dashboard rests on a quiet layer of analyst judgment that no one has written down or made anyone accountable for.
3. 3.Operations owns the entry, not the standard. The people who create the data — in the CRM, the ticketing system, the workbook — follow no shared definition of what each field means. Two people record the same event two different ways, and both are sure they are right.
4. 4.Leadership owns the decision, not the definition. Executives consume the metric and steer by it, but no one at that level owns what the metric counts. When two reports disagree, the meeting argues about whose number is right instead of which definition the company actually chose.
5. 5.The tool owns the test, not the call. Monitoring flags an anomaly and pages a channel. But a test that fires with no one accountable to act on it is just a louder way to not fix the problem. Alerting is not ownership.
6. 6.Everyone owns a piece; no one owns the number. This is the gap itself. Data quality gets run as a relay — each function responsible for its own leg, no one responsible for the baton arriving intact. The handoffs are exactly where quality is lost, and the handoffs belong to no one.

None of the six are solved by buying another tool. Tests, catalogs, and observability are real and useful — but they decay back into noise without someone accountable for acting on what they surface. An owner is the thing that makes every other data-quality investment compound instead of rot. Without one, you are funding instruments that no one is required to read.

Why this matters before you build AI

A model is the most literal data consumer you will ever have. It inherits whatever quality the data carries, applies it at scale, and removes the human who used to quietly compensate — the analyst who knew to drop the duplicate region, the ops lead who remembered the definition changed in March. An unowned dataset becomes an unowned model: confidently wrong in exactly the places no one was accountable for, and now wrong automatically, thousands of times a day. The ownership gap that was an annoyance in a quarterly report becomes a systematic error in an automated decision — and the model, unlike the analyst, does not know to flag it.

What an owner actually does

Ownership is not a person who personally fixes every error. It is a role accountable for four things on a specific dataset: a definition everyone agrees to, a standard for the data that feeds it, a correction path faster than working around it, and the authority to declare the number final. You do not need this across all your data at once. Name the owner for the handful of decision-critical datasets first — the three or four numbers the business truly steers by — and let the rest follow. The owner does not have to be a full-time hire, either. What has to be singular is the accountability: one named role that answers for the number, whether that role sits inside the company or is a fractional, external partner brought in to hold it.

Data quality is not a tooling problem. It is an ownership problem wearing a tooling costume — and no tool answers a question no one is accountable to ask.

So before the next data-quality tool or the next AI pilot, answer the simpler question first: for the numbers that matter most, who owns the answer? Name them, give them the definition, the standard, the correction path, and the authority — and the tools you already bought start to work, because someone is finally accountable for what they say.