The clinical trials industry has spent a decade debating digital transformation. What it has spent far less time debating is who pays the price when transformation stalls — and who holds the accountability when AI gets it wrong. At a recent roundtable convened during Veeva’s annual gathering, three of the company’s most senior voices on regulatory strategy sat down to work through the questions the keynote didn’t have time for: the unintended consequences of the FDA’s guidance rollback, the opacity of EMA’s AI adoption, the stubborn dominance of paper in a supposedly digital era, and whether Europe’s fragmentation makes its competitive ambitions credible. What emerged was less a product briefing than a frank account of what it looks like to build technology for a regulatory environment that nobody — sponsors, agencies, or vendors — fully controls.
Moe: How should accountability be assigned when both sponsors and regulators are relying on AI to evaluate the same submission?
Senior Director, Government Strategy, Veeva Systems
Crystal Allard: Our customers — sponsors of an application to a health authority — understand that they hold all of the responsibility for submitting correct data to a regulatory agency, and they take it very seriously. From Veeva’s side, we have to clearly define the way our AI will be working as we roll it out, whether it is verifiable, and how you’ll maintain compliance with those requirements. Those ten guiding principles were very high level. There are a couple of other guidances that are much more detailed — the drug discovery guidance out of FDA, for instance — so we take a careful review of those to ensure we’re remaining compliant as we’re building. We have plans to reach out to both FDA and EMA and get answers if we have any questions about whether our systems are meeting those requirements, and we do now have direct ways to communicate with them.
As far as what FDA is doing — they also know that they are fully responsible for making a determination of safety and efficacy on an application they’re reviewing. They are absolutely exploring AI tools. You probably saw that their chief AI officer just recently left, so they’re in a little bit of a question about the future of their AI tooling. We should be assuming that health authorities, especially the bigger ones, are going to be using AI — at a minimum, they’re going to be automating process with AI. So we need to be submitting AI-ready information for them to use, which means we need to rethink data format. Maybe the two-dimensional standards we’ve been using for years aren’t entirely necessary in the way that FDA is going to use that information. But they also have to help us understand how they’re going to use that information. We all have a lot to learn in this process, but we’re staying connected to make sure we’re working with health authorities to give them information they can use the way they want to.
Moe: How do you explain the difference in transparency between FDA and EMA on AI adoption?
Crystal: The EMA has staged their path forward recently, but FDA, to their credit for the last year and a half or so, has also been very open about what they’re trying to do with AI. It’s not as easy to predict where it will land, but they’re giving information. EMA is not putting out a lot of information about how they’re using AI. We know they are, but they’re not openly telling us what they’re implementing and how they’re using it. Whereas FDA openly said they’re going to use it for filing, which is pretty useful. So yes, there are major differences in the amount of information that we know — and in innovation and adoption.
Senior Director, Clinical Strategy, Veeva Systems
Werner Engelbrecht: I agree. We had a technology roadmap meeting with EMA a couple of weeks ago, and when it came to AI functionalities used internally, it was honestly not very clear where exactly things are coming in. One area they are looking at is UCTR and CTIS submissions — they’re looking, two years from now, to have advanced AI capabilities utilized for the initial review. But I have not yet seen much transparency about potential real-world use cases beyond that.
“We need to be submitting AI-ready information for health authorities to use — which means we need to rethink data format. Maybe the two-dimensional standards we’ve been using for years aren’t entirely necessary in the way that FDA is going to use that information.”
— Crystal Allard
Moe: How do you define meaningful progress in regulatory digital transformation — as opposed to just migrating the same processes onto a new platform?
Crystal: We’re approaching it in three different ways. The first is working directly with standards organizations to get them to converge on their own standards. HL7 has created duplicative standards — one for FDA and one for EMA — and I would like them to stop doing that, because this over-complexity makes it harder for us to build technology that’s actually useful. We can build technology that can output in any format you want, but it takes more time, even with technology, and it’s not as good an experience for users as it would be if it were simpler.
We’re also working with standards organizations to ensure that as they’re building new standards, they’re building standards that will work with AI. I don’t think we’re ever going to get away from standards entirely — the serious use cases rely on that standardization to make AI work well. But the standards will look different. They’re not going to be these two-dimensional flat files. We need semantics, ontologies, and more digital-native data. And they need to be developed for use with a technology exchange platform.
We should be assuming that the future of data exchange will be digital and will look different. It won’t be this over-the-fence process where we do everything here, validate it, send it over there, and they validate it and do exactly the same thing. If we’re sharing space in the middle where standardized data can exist and be used and pulled, we need to rethink the way we’re developing these things. So we’re working with standards organizations, we’re working with platform partners, and we’re working directly with health authorities to help them understand the value of this for their own public.
Vice President, Public Policy – Clinical Trials, Veeva Systems
Anthony Corso: Standards used standardly would make all of our lives a little bit easier. There’s so much paper — AI can read it, or you can scan it with character recognition, but somebody’s got to print it, get it out, scan it, and not make any mistakes. If they made an error on the paper, now you have to fix the paper and scan it again. That’s the trial landscape in a nutshell, globally. Tufts University did a study and figured 67% of consents are still done on paper. And then the fragmentation in healthcare — you have EHD with a target of 2029, and if you happen to be in a member state that has one EHR, that helps, but that’s not the case for most of the EU. All of these disparate health record systems are going to have to figure out a standard to feed into the EHD. And even third-party standards organizations are creating duplicate standards to meet needs instead of working toward harmonization. That’s just what we need — from research through healthcare — to make any of this really work.
Moe: Why are sponsors delaying trials rather than accelerating into a lighter-touch regulatory environment in the US?
Anthony: For sponsors, late-phase trials are so expensive and so invested in — and then you don’t have that guidance, or there’s uncertainty. Pharma has generally expressed a decline in trust in FDA specifically. They’re not in the uncertainty business. They work with numbers, discrete data, defined endpoints. They put time, money, and people into getting a medicine to market. So I think the rollback had unintended consequences. Whether it’s legislation or regulation, you need to be smart about it — and when you just wholesale remove things, people are going to pump the brakes because you don’t know. As far as FDA itself, they are hiring a lot of people right now. There are a lot of job postings out there. So I think there was an oversteer, and they’re course-correcting to get those experts and professionals back and rebuild that trust. We talked a little about this: the US has seen fewer trials, less access, less growth — and that’s a concern across the board. Everybody wants to keep trials in their market. It’s a big part of the economy. When you rock the boat too much, people get scared about their business.
Crystal: We do have customers in the regulatory space who have made decisions to pivot and look at different markets instead of defaulting to FDA first. We’ve absolutely seen that happening, anecdotally. I worked at FDA during the first Trump administration, when there was a one-to-one guidance executive order — for every one you issue, you have to remove one. Removing a guidance is actually a very burdensome process. It takes a lot of time and resource. So what happened was we stopped writing guidance. We didn’t remove one to write the new one — we just stopped writing them altogether. So again, it’s this unintended consequences piece. It sounds like a good idea, and there probably are guidances that are outdated and more than are necessary. But I do think AI is actually a great use case here — finding outdated guidance and regulatory documentation and using AI to update it. You can also use AI to review public comments when you send guidances out. I am hopeful that all health authorities will be able to get more administratively efficient using AI. I know that was FDA’s intention. But it’s really hard to do that in three years. It just takes a long time.
“We just stopped writing guidance altogether. We didn’t remove one to write a new one — we stopped writing them. Again, it’s this unintended consequences piece.”
— Crystal Allard
Moe: How do you assess the EU’s competitive pitch for clinical trials given how fragmented the region remains operationally?
Anthony: If you look at the EU Biotech Act, the second page has a graph — US, EU, China. They wrote an entire act to be competitive. No matter what country you’re from, pharmaceuticals and devices are such a boon for your economy. Our customers and sponsors want to be in every country — that’s the biggest ROI. But then there are countries that specifically want to be the entry point into the global economy. So everybody’s trying to figure out what they’re doing in real time. The EU has the Biotech Act; the US has its own approach. AI is the shiny cure-all to speed — that’s true on both sides. But there’s so much paper in the middle. It would be so much more beneficial to standardize all of that technology, how it communicates, pull from healthcare, and accelerate EHD. That would be, in my personal opinion, the quickest way to staying competitive in a global market.
Crystal: Most of our customers are global companies. They want their products on the market everywhere — it’s more about where to go first so they can learn from that application. One of the things we’re working on is with the Accumulus platform, where you can submit at the same time to 70-plus different countries. They’ve run pilots where they sent something to EMA, took the approval letter, put it into Accumulus, and got approvals from those countries in 24 to 48 hours. So yes, we are looking at things differently locally, but our companies are still very global, and collaborative review is still a better path for them.
Anthony Corso is Vice President, Public Policy – Clinical Trials at Veeva Systems. Crystal Allard is Senior Director, Government Strategy at Veeva Systems. Werner Engelbrecht is Senior Director, Clinical Strategy at Veeva Systems.
Moe Alsumidaie is Chief Editor of The Clinical Trial Vanguard. Moe holds decades of experience in the clinical trials industry. Moe also serves as Head of Research at CliniBiz and Chief Data Scientist at Annex Clinical Corporation.
