Sponsored content
This interview was produced in partnership with Veeva Systems.
Veeva Systems
The clinical data world has been promised transformation before. With advancements like eSource, risk-based monitoring, and paperless trials, each arrived with genuine momentum, and each stalled somewhere between pilot and scale. Drew Garty, Chief Technology Officer for Clinical Data at Veeva Systems, has spent three decades watching that pattern repeat, and he thinks he finally understands why. It is not a technology problem. It is a timing problem, a change management problem, and, in a specific and fixable way, a silo problem. In this conversation, Garty maps the three structural waves he believes are now converging, explains how Veeva’s Study Builder Agent is designed to collapse the cross-functional misalignment that turns protocol amendments into database delays, and makes a case that the most consequential gains of the next five years will come not from what AI does to data, but from what it allows people to stop doing.
Veeva’s Drew Garty on why three converging waves, not one AI breakthrough, will finally move clinical data off its 20-year plateau.
Why is this moment in clinical data innovation structurally different from earlier cycles that also promised efficiency?
Drew Garty: I like to think of it as a constructive interference moment, where enough different activities are happening in parallel that they start solving some of the structural challenges we’ve had.
A good example is eSource. The site’s paramount concern is seeing patients and providing care as seamlessly and efficiently as possible. Clinical research structurally asks something different of them, and historically the solutions sponsors provided weren’t built for sites. They were designed for the collection of data for sponsors.
Things have changed on that front. More solutions are available that are built specifically for sites and the concept is simple. Sites have their own tool set, their own place to record their ISFs, CVs, and all of their content for a trial. They should have their own CTMS to process the business side of clinical research. And as it relates to patient data, when you look at patient privacy laws internationally, sites need to be in control and protect patient identity. That means they need to hold those records in their own ecosystem. Sponsors can’t have them.
There has always been friction around how a sponsor lets sites into their tools while protecting patient privacy. We’ve handled that through transcription, delivering a site copy, then a sponsor copy. By giving sites their own environment, they can take in the trial’s data collection definition, add their own identifying information, and then only the data the sponsor asked for via the case report forms gets automatically sent out. Sites manage patient privacy. Sites manage their own clinical trial activities within their own process.
How did leading EDC adoption at scale teach you about the gap between a technology being available and an organization actually changing how it works?
The way I think about it, in surfing, what’s challenging isn’t springing up on the board. You can learn that on the beach. You can build balance, build a core. What’s really challenging, and what you can only do out in the water, is understanding the timing. The timing of the energy of that wave, and when to take advantage of it.
In our industry, there have been three key waves. Two we’re deep into. The third is upon us.
The first wave was digitizing data. Instead of paper case report forms, we moved to tools like EDC. IRT and randomization tools (RTSM) are another example. What made that possible was personal computers becoming reasonably priced, and then the internet enabling the exchange of information, so network costs kept going down. Sites could enter data digitally. We could ask real-time queries. We improved quality faster.
The second wave came as a result of digitizing so much data. We then had to deal with it en masse. That wave caught a lot of us off guard. Tools like a Data Quality System can bring all that data together and automate key processes, including aggregation and cleaning. That gives data management back the time to focus on the actual state of the data.
In my career, it’s always been roughly a decade where new technology first starts emerging to the point where the industry is the majority of the way in and receiving real value.
When I look at the third wave, leveraging LLMs and agentic workflows, a lot of people in the industry were saying about three years ago that the time was now. Jump on board. I said it was too early. Regulatory still needed to think about it and give some perspective. The technology needed a more robust framework around it. And critically, you can’t just add complexity to something complex and hope for simplicity and quality. You have to actually remove things.
Removing processes people need to do frees them up for the critical thinking that is the real evolution of data management. But you need time for that. If roles are getting wider, you must shape that, train people, hire for it. Even when you know the vision and have the technology, it takes years to pivot all of the stakeholders in a clinical trial process. It is a long journey.
“You can’t just add complexity to something complex and hope for simplicity and quality. You have to actually remove things.”
Veeva recently announced a new Veeva Study Builder Agent. How does the agent address cross-functional misalignment when it still requires a finalized protocol as its starting input?
Drew: One of the challenges Veeva Study Builder Agent is focused on is that today we run trials in distinct silos. Those silos are organizational. Data management, clin ops, stats, and medical all operate as independent teams. They each start with the final protocol and ask, what does this mean for my team? What do all the footnotes mean? How do they apply to my work?
The idea behind Study Builder Agent is to transform when and how the study team comes together. Instead of working in a silo, you disagree and commit to the design up front and then use that metadata to automate the downstream system configurations. We’re starting with clinical data, but the vision is to go much broader.
What this addresses at a practical level is the downstream misalignments. When two teams configure something differently, you discover it later in testing. It’s really about all of those small cycles, the turbulence that goes on after each team creates something independently.
With shared metadata, we can store and utilize that information in ways that benefit the whole study. Industry initiatives like CDISC’s Unified Study Definitions Model (USDM), and TransCelerate’s Digital Data Flow (DDF) are focused on bringing together a standard that describes the protocol. But a protocol doesn’t describe an entire study. It’s an abstract. It has business intent that isn’t about configuring all the systems. Each system will have its own further requirements downstream. The real value is in that upfront alignment. From alignment, we get first-time quality.
By automating the configuration and build of systems and removing the bottlenecks to that goal over a few years, systems can potentially be configured in a day. Teams can then load them with synthetic or test data, simulate the analysis, the TFLs, the tables and listings and figures, and provide meaningful feedback for the design of the protocol from the very first draft. Instead of a waterfall where you finalize the protocol and then build all the systems to collect, you’d configure the systems, load them up, see how you’d want to optimize the protocol, make changes, and do it again. Fewer poorly designed protocols. Better study integrity, and hopefully, better scientific outcomes.
How does human oversight work in practice for AI-generated configurations, and what does the validation story look like ahead of the December early adopter release for Study Builder Agent?
Drew: The decisions that come from bringing people together still need to be recorded. You need a system of record for that. That is not AI. That is an application. Applications still have their place. In fact, they’re critical for AI success. They represent the organization. They represent the quality of the process in an enduring way.
So it’s really the coupling of a workflow and system-of-record application with the LLM doing the configuration.
A lot of people ask: how can you use a nondeterministic or probabilistic method and yield high quality? The reality is people make mistakes. Agents will make mistakes too. So, it’s really about the review process.
As you shift humans away from doing all the manual setting, you still need an efficient and consistently high-quality way to review all those configurations.
One of the things we talk about is standards first. There’s a philosophical debate in the industry. Does AI remove the need for standards, or does it increase the value of standards? We follow the latter. We believe strongly that standards are agreements among people about what right looks like. A data collection standard is based on a data standard the stats team expects to receive so they can run their analysis. As we’re assembling configurations, we’re using more intelligent building blocks based on human agreement, as much as we can, and protecting those from nondeterministic alteration without awareness.
We’re using the AI to create new things. And new things have always had to be reviewed and agreed upon. That’s how we build quality into the process. It’s a workflow, with user experiences around it, and we use an application as the enduring system of record.
Why will eSource and participant engagement be the most consequential automation opportunities for sites and patients in the next three to five years?
Drew: I don’t see EHR coming all the way into this space and solving for clinical research in the next three to five years. What I do see is a lot of solutions in the industry that exist to help sites move data from source, collect additional source data necessary for the trial, and flow that data seamlessly all the way through. We call it straight-through processing, going straight from source to the sponsor’s environment. But that is an industry initiative. We all must work together, just like we have to figure out agentic clinical data together.
From a patient perspective, we’re also looking at the transition from sponsor motivations toward doing what’s right for participants. A good example is the shift alongside eCOA, where we were really collecting direct patient data. The focus now is on participant engagement. Helping participants to navigate the study, understand what’s happening to them, understand what’s next, and stay connected with their research sites. Not all sites have the same tools available to support that.
There is a long-held goal in the industry of giving data back to participants in meaningful ways that help them understand what’s happening when the time is right. When you focus on the human engagement, you end up with a better relationship. This helps retain participants or even attract them into a new trial. We have a significant attrition problem, in both sites and participant populations.
eSource and participant engagement are two areas I think will make the biggest difference in the next three to five years. And on top of that, looking at the problems LLMs can help solve, if we build the right framework for it.
“It’s not just about technology. This is about the human endeavor. With a focus on that, we can get there together.”
When a data manager opens a study that Study Builder Agent configured, what should feel concretely different on day one?
Drew: I think it’s truly breaking away from the silo. From having to wonder what the other teams are interpreting. A lot of people have to do this in a vacuum. Read a protocol, go create the forms, and never have the conversations about the footnotes or how to interpret them. You end up producing a lot of rework.
We’ve proven in other parts of EDC design that bringing people together to disagree as quickly as possible and then commit on the decision is transformative. One of the core tenets of our EDC builds is exactly that. How do you go from 12 weeks to 6, down to 4, down to 1? We now have a customer who’s doing their EDC builds in 1 week with agreement from the study team, because they all come together as early as possible.
I know there’s a lot of focus on LLMs and AI, but sometimes it’s just bringing people together in a meaningful way that makes the biggest difference.
It is change management. All of the advancements and technology, they absolutely require people to change. And that means this isn’t fundamentally about technology. It’s about the human endeavor. With a focus on that, we can get there together.
Drew Garty is Chief Technology Officer, Clinical Data, at Veeva Systems.
Moe Alsumidaie, MBA, MSF, is founder and Chief Editor of Vanguard Publications, which publishes Clinical Trial Vanguard, Pharma Vanguard and BullScope, and Head of Research at CliniBiz. He has two decades in clinical trial operations and data science, with earlier roles at Genentech, Abbott Vascular and Stanford University Medical Center, and is a guest lecturer in clinical trial sciences at Rutgers University.
