AI Saved Time. Employees Didn't Feel It.

I was excited to attend this year’s Veeva R&D & Quality Summit in Boston. Unfortunately, an ankle injury two days before the conference landed me on crutches and put the kibosh on my travel plans. Not exactly the kind of disruption I expected to be dealing with that week.

I was especially disappointed because I was scheduled to moderate a panel with Anthony Corso, Veeva’s VP of Public Policy, discussing current policy issues, their impact on clinical trials, and how Veeva is advocating for research sites. Fortunately, someone I have a ton of respect for, Moe Alsumidaie, chief editor of Clinical Trial Vanguard, stepped in at the last minute. Thanks, Moe!

I also owe a thank-you to Veeva’s crack Communications Director, Deivis Mercado, who managed to convert two of my planned in-person interviews into Zoom calls. Those conversations were with Angie Badgett, executive director, global project management and strategic solutions at Celerion; and Sweta Pittala, VP of technology portfolio at Velocity Clinical Research.

Although they work in different operational environments, both women are tackling similar challenges: eliminating administrative inefficiencies, implementing AI without creating new headaches, and getting employees comfortable with technology that’s changing how they work. And one of the most interesting things I heard involved an AI implementation that was saving employees time, even though those employees didn’t necessarily believe it.

AN UNEXPECTED AI OUTCOME

Pittala explained that the company had underestimated how employees’ responsibilities were changing. Previously, someone handling an invoice knew where to look, what contract provisions to check, and how to complete the task. With AI performing much of that work, employees were suddenly responsible for reviewing outputs, resolving exceptions, and providing information the system needed to proceed. Sure, the work had become faster, but it had also become unfamiliar.

There were other complications. The underlying data contained missing information, duplicates, and inconsistent terminology across systems. When AI produced incorrect results, employees lost confidence and reverted to their familiar manual processes.

All of this caused Velocity to rethink its implementation approach, putting more emphasis on helping employees understand their changing responsibilities. I found this especially interesting because it challenges a common assumption about AI adoption: that resistance primarily comes from employees worrying about losing their jobs. Pittala acknowledged that concern exists but said it represented a relatively small portion of the difficulties Velocity encountered.

Much of the challenge involved employees learning entirely new ways of working.

Badgett offered a complementary perspective on change management. During Celerion’s implementation of the Veeva Study Training application, the company deliberately assembled a small project team of approximately seven people, representing the operational departments that would use the system.

Those individuals weren’t simply there to provide feedback; they were responsible for communicating changes, preparing colleagues, addressing concerns, and building enthusiasm within their departments. Badgett described it as the company’s most successful implementation in terms of adoption.

AI IS ALREADY CHANGING THE ADMINISTRATIVE WORK

So, when do they make these “transformative” changes?

Simple. When the customer asks for them.

One example is the company’s adoption of Veeva’s eCOA technology. Participants now receive questionnaire reminders through a patient app, while sponsors and CRAs can review responses in real time.

When our conversation turned to AI, Badgett focused on the administrative work that consumes employees’ time. One of her examples involved deviation trending. Celerion undergoes approximately 90 sponsor qualification audits annually. Sponsors frequently want information about deviations, mitigation plans, quality tolerance limits, and trends across studies. Historically, assembling the necessary information could require hours, days, or even weeks of manual work. She said an AI-enabled process can now produce the information in approximately 10 minutes, allowing employees to spend their time analyzing trends and deciding what to do about them.

Pittala described a similar philosophy at Velocity, which operates a global network of more than 70 sites. “The most valuable applications aren’t necessarily the most sophisticated AI models,” she explained. “They’re the ones that eliminate friction from an everyday workflow.”

Velocity is applying AI across three primary areas: patient recruitment, site operations, and financial operations. The first two will sound familiar. But, financial operations? That one caught my attention.

Pittala explained that Velocity is using AI agents to interpret contracts, identify invoiceable activities, and reconcile payments. Across a large site network (or any site, for that matter), missed billing triggers and delayed invoices can leave significant revenue uncollected or slow cash flow.

The company piloted this approach roughly a year and a half before our conversation. Since then, AI agents have handled most accounts receivable operations, with employees intervening when exceptions arise and performing quality checks.

TRUST, VALIDATION, AND WHAT’S COMING NEXT

Trust becomes even more complicated when AI enters regulated clinical research workflows. Badgett recalled hearing a conference speaker say that organizations “only move as fast as your trust in a system.” She wrote it down verbatim. You can have the best change management plan in the world, she said, but if employees don’t trust what the software is doing, it won’t matter.

Pittala explained why traditional validation strains: With traditional software, you know what to expect from a defined set of rules, but with generative AI, outputs can vary even when inputs are similar. That raises new questions around validation, traceability, consistency, and accountability.

Badgett argued that the industry’s approach to validation must evolve, including how it defines acceptable performance. “We can’t continue to live in a world of pass-fail standards in terms of validating a model like AI. We have to start living in a more gray world.”

Don’t worry; she’s not advocating we abandon regulatory expectations or accept unreliable systems. But, it does raise an important question about how companies establish and demonstrate appropriate performance when AI is involved.

Pittala described Velocity’s efforts to address those concerns through an internal AI governance group involving legal, quality, IT security, technology leadership, and other relevant stakeholders. The company has adopted Gartner’s AI TRiSM framework to guide assessments involving security, compliance, traceability, and audits. For regulated workflows, she emphasized the importance of meaningful human oversight. Simply inserting an approval step isn’t sufficient if the reviewer lacks the context needed to evaluate the AI-generated output.

What’s Next?

Looking ahead, Pittala described a future in which coordinators begin their day with a single command center that identifies priorities, monitors outstanding information, and recommends actions. For example, an AI agent might recognize that laboratory results are missing before a participant’s scheduled visit, follow up with the laboratory, and alert the coordinator that the appointment may need to be rescheduled.

Pittala also raised another practical concern: AI projects that look promising during pilots can become considerably more expensive when deployed at scale because of token consumption and other operating costs. Those expenses need to be part of the ROI calculation.

Neither interview provided evidence that AI has solved the longstanding operational challenges facing clinical research. And neither woman suggested the work was finished.

I missed the conversations, sessions, and networking that make an in-person conference worthwhile. But thanks to Deivis, Angie, and Sweta, at least my ankle didn’t prevent me from having two worthwhile discussions about the technology that’s already changing ClinOps.