Who Owns Data Quality? The Authority Gap Slowing Trials

By John Oncea, Chief Editor, Clinical Tech Leader

Ask James Richardson what happens when a clinical study isn’t executed cleanly, and he doesn’t start with compliance risk or data integrity in the abstract. He starts with money.

“If the study is not executed robustly, then there can be hidden costs associated with that, such as the study going on for longer or having to be repeated,” said Richardson, an industry pharmacist who spent a decade in Bayer’s medical affairs organization and is now moving into a global medical communications leadership role at UCB, a multinational biopharmaceutical company headquartered in Brussels. “And sometimes, especially in a startup environment, these are go-broke activities or go-broke consequences.”

That’s not a hypothetical for Richardson. Reflecting on a study environment he worked in roughly seven years ago, he described sponsors who “just look at what they want to achieve in the study, and they look at the dollar amount of how much it’s going to cost to perform it,” without weighing the downstream cost of a study that has to be re-run because the underlying data wasn’t clean. For a well-capitalized company, that’s an expensive mistake. For a startup running on a fixed runway, it can be fatal.

The fix, in Richardson’s telling, isn’t complicated in concept. It’s a matter of timing and ownership, two things clinical research organizations routinely get wrong.

Fix It Before The Protocol Is Signed, Not After

Richardson was direct about when data-quality and process decisions need to happen: before the protocol is finalized, not once the study is already underway.

“To keep things uniform, it’s best done before the protocol is signed off,” he said. “Otherwise, there might be a chance that, in the early stages of a study, if you’re like deviations identified and there’s a need to correct some things, then there might be an opportunity, but you don’t want to wait until you get to the end of the study and realize you should have done things differently from the beginning.” His guidance was specific: lock in data-quality decisions at the protocol synopsis stage, before the study moves into execution.

That’s a narrow window, and it’s one that’s easy to blow past when protocol development is focused on scientific design and regulatory strategy rather than the operational mechanics of how data will actually be captured, transcribed, and cleaned.

The Person Who Should Own It Isn’t The Person Who Does

Here’s where Richardson’s answer gets more interesting and more uncomfortable for organizations that assume good governance follows naturally from good intentions.

“I’ve got two different answers,” he said when asked who should be responsible for data-quality decisions. “The person who should own it, but they generally don’t have the correct level of voice, would be the stats guys. Because if the stats guys want clean data, then of course this will be the way that they do it.”

In practice, though, authority sits somewhere else entirely: “Ultimately it would be whoever signed off the protocol; it would be the clinical head or the clinical director. I think that is a high enough level that the book stops with them, but also that they have budgetary control of it so that they can decide.”

In other words, the people with the clearest technical stake in clean data are rarely the people with the authority or budget to enforce the standards that would produce it. That mismatch, more than any specific technology gap, is what allows preventable data-quality problems to survive into a locked protocol.

Vendors Could Be Doing More Educating, Not Just Selling

Richardson also pointed to a missed opportunity on the vendor side. Looking back at his own experience selecting an electronic data capture system, he said he would have welcomed a more consultative sales conversation rather than a straightforward product pitch.

“I would’ve loved back in the day if our EDC provider said, look, we’ve got a couple of options for you. You use our product. We’ve also got our product with a plus package,” he said. “I would’ve loved to listen to that pitch because it probably would’ve identified areas that I didn’t know that I didn’t know.” He was explicit that he wouldn’t have read that kind of pitch as pushy sales tactics: “I would’ve really viewed that not as someone trying to sell me something that I don’t want. I would’ve really viewed that as at a minimum a learning opportunity.”

For technology vendors serving clinical research, that’s a specific and actionable data point: sponsors making EDC and data-capture decisions may be underinformed about their own options in ways a well-structured sales conversation could actually fix, not just at the point of purchase but in surfacing quality risks the buyer hadn’t considered.

AI Doesn’t Change Who’s Accountable

The same ownership logic extends to how Richardson thinks about AI’s role in cleaning up this picture. He’s not skeptical of AI as a tool — he described it as useful for flagging outliers and processing data faster than a person could. But he was firm that accountability can’t move with it.

He drew a direct comparison to standards already in place for AI-assisted medical writing: publications can disclose AI use, but the AI can’t be listed as an author, “because to be an author, you’ll have to be accountable that it’s all true and correct.” He extended the same logic to clinical study report sign-off. “You could use aspects of AI to help write the report and to process the data, but it can’t sign off the report because the person has to be responsible for it and AI can’t be responsible,” he said.

That’s the thread connecting all of it: timing, ownership, vendor relationships, and AI governance. Better technology doesn’t solve a clinical research organization’s data-quality problems if the authority to act on what that technology finds still doesn’t sit with the people who understand it. Fix the org chart, Richardson’s experience suggests, and the technology gets a lot easier to use well.