Jacqueline DeStefano-Tangorra is the president and CTO of DataOps.
Before a single model gets trained, before anyone opens a slide deck about transformation, I ask the room one question. Who owns this data? The answer is almost never clean. People look at each other, and sometimes someone names a system instead of a person. Sometimes someone else names a department that stopped existing two reorganizations ago. That pause tells me more about whether an AI project will succeed than any technical assessment I could run.
I have watched brilliant AI initiatives collapse not because the algorithms were weak, but because no one could answer that question with confidence.
Early in my consulting work, I sat with a leadership team that had invested heavily in a forecasting model. The data scientists were talented, the infrastructure was modern and yet the forecasts kept drifting in ways no one could explain. When I traced the problem back, it was not the model. A supplier field had been redefined by one team while another team kept feeding the old definition into the pipeline. Neither team knew the other existed in that workflow. Both assumed someone else owned the field.
That is the pattern I see across aerospace, defense, healthcare, manufacturing and financial services. The failure rarely lives in the AI. It lives in the ambiguity underneath the AI. When ownership is undefined, every downstream decision inherits that ambiguity and quietly compounds it.
The word governance makes many executives brace themselves. They picture committees, documentation and a slowdown of the very innovation they are trying to accelerate. I understand the reflex, and I want to reframe it. Data governance done well is not a brake on AI. It is the drivetrain that lets AI move at all.
Think of a data flywheel, the self-reinforcing loop where better data produces better models, which produce better decisions, which generate better data. That flywheel only accelerates when the data feeding it is trusted and traceable. Without governance, the flywheel grinds. The models learn from noise. The organization loses faith in the output, and adoption stalls not because the technology failed but because trust failed.
Ownership is the first layer of that trust. When a named person is accountable for the accuracy of a critical data element, you create a chain of responsibility that runs from the source system all the way to the model’s output. When something breaks, you know where to look and who to ask. That single change turns debugging in the dark into a solvable problem.
Ownership is more than a name in a spreadsheet. It requires four things that most organizations underinvest in.
It requires lineage, so you know where data came from, what transformations it passed through and who touched it along the way. When a forecast is wrong, lineage is how you find the root cause instead of guessing.
It requires quality at the point of ingestion. A model is only as reliable as the data it consumes. Validation rules, schema standards and completeness checks at the entry point prevent bad master data from corrupting everything downstream.
It requires clear accountability tied to real people, not abstract systems. Tiered ownership, where the most critical elements carry the most senior accountability, gives you a map of who answers for what.
It requires policy that travels with the data, especially in regulated environments where classification and access control are not optional. In defense work, the responsibility and the tightening compliance landscape have turned data ownership from a best practice into a mission requirement. The organizations that treated governance as paperwork are now scrambling. The ones that treated it as infrastructure are ready.
Here is what I have learned after building governance operating models for large enterprises. The technology is the easy part. The hard part is leadership willing to say, out loud, that we do not currently know who owns our data, and we are going to fix that before we scale AI on top of it.
That admission is uncomfortable. It exposes gaps that have been comfortably ignored for years. But I have never seen a serious AI transformation succeed without it. One senior data executive I worked with put it plainly. The transformation starts by acknowledging where you are. The acknowledgment is usually the bottleneck, not the road map.
Leaders who embrace that honesty move faster in the long run. They stop pouring investment into models that sit on unstable ground. They build the ownership structure first, and then the AI they layer on top actually holds.
Every organization I meet wants to talk about AI capability. Fewer want to talk about the data foundation that determines whether that capability will ever deliver value. The two conversations are the same conversation. You cannot separate the intelligence of a system from the integrity of what feeds it.
So I keep asking the question, in every first meeting, before any project begins. Who owns this data? If the room can answer clearly, we are ready to build something that lasts. If the room goes quiet, we have found the real work. And the real work, every time, is worth doing before the AI, not after.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
