AI Agents, Patient-Centered Design, and the Future of Clinical Trial Execution: Q&A with Krishna Cheriath, Thermo Fisher Scientific

As artificial intelligence (AI) moves from pilot projects into core clinical operations workflows, the question is no longer whether the technology works—it's whether organizations are structured, staffed, and process-ready to make it work.

To explore this further, Applied Clinical Trials spoke with Krishna Cheriath, VP and head of clinical research digital data and AI at Thermo Fisher Scientific, about what separates successful AI adoption from stalled implementations, how patient-centered digital strategy plays out at the site level, and what a realistic agentic AI future looks like for clinical trial roles.

ACT: How are clinical operations teams using AI-driven tools to improve case intake, and what separates a successful implementation from one that stalls?

Cheriath: When you look at the opportunities in case intake, there is a range of use cases that are active today. AI is really great at knowledge synthesis—being able to summarize information, frame recommendations for incorporation—but at the end of the day, a human still needs to own the results and outcomes around it. So knowledge synthesis, summarization, and recommendations is a standard pattern. The second is what I would call the applied intelligence aspect, where AI is able to detect patterns and provide recommendations that support human judgment across those cases.

To your question on what it takes to successfully implement—and this is not just related to case intake, but across the board—what I find is that tech is no longer the barrier for progression. Tech is complex, and we need to do it, but the two biggest factors that drive successful adoption are, number one, the willingness to reimagine business processes and operational workflows to be AI by design. And second is upskilling the talent to be able to use this appropriately and wisely. Those two things look great on PowerPoint slides to say, but they're harder to do.

The way I look at it is: bad business process plus AI still equals bad results. So how do you take a step back? What you classically see in a digitally native company—one that doesn't have the accumulated history of how things have been done—is the ability to start from an AI-first position. The critical success factors are how much attention can be put to the fundamental reimagination of workflows and business processes with AI, and the attention to upskilling the workforce so that people trust it, can use it, and use it appropriately.

ACT: What organizational factors most determine whether AI adoption in clinical operations delivers real efficiency gains?

Cheriath: To me, you need both moon shots and micro shots. You have the top-down objectives around clinical operations—significant cycle time compression, significant quality gain, big productivity increase. Those are what I would call moon shots, big bold bets. That is needed. But at the same time, you need to build confidence bottom up to use AI appropriately and get the workforce using it properly as well.

So the first thing is making sure that the priority use cases you pursue strike a fair balance between moon shots and micro shots. The second is that across biopharma, biotech, CROs, and others, all of us are busy with a ton of things to do. Is there enough time in the day to step back from the hamster wheel and say, could I do this differently? Could I do this better? The space for innovation is becoming a problem across many industries. The core subject matter experts who really know how things are done today—do they have the space and time to step back from the day-to-day busyness and look at the art of the possible?

The third is that to really transform, you need a combination of expertise. You need people who really know the business processes of clinical operations workflows, and you need people who understand the potential of AI capabilities like agentic AI. You need this combination to be able to step back and reimagine. That's hard, because you may have subject matter expertise in clinical operations but not enough tech translators who can connect and empathize with clinical operations. There are not enough boundary spanners—people who can speak both clinical operations and AI tech. The more you can build that kind of combo team, the more that will predicate your success.

ACT: How do you define a patient-first digital strategy, and what does it look like in practice at the site and participant level?

Cheriath: I've gone through my own direct family's healthcare journey, which has given me a sharper focus on the patient and caregiver. I've been working in the tech space for 30-plus years, all of it at the intersection of digital and healthcare. Some of these things can feel remote in terms of impact until you feel it yourself.

If I take the enrollment of patients in clinical trials as an example—I am a digital geek at heart, and I have firm conviction that AI-based capabilities to locate patients, match patients, and help give a friction-free enrollment experience are going to really help improve enrollment rates. The experience management of the patient and caregiver enabled by AI, so that they have the information where they need it and when they need it, makes a real difference. But I'm not convinced that is enough, because there are a lot of other factors that drive patient experience beyond just the tech side of the equation.

I was recently with a patient advocacy organization talking to patients, and the conversation was a stark reminder that there are social determinants of health that also need solving. The best AI systems don't solve for the fact that a patient has an accessibility and affordability problem. If a patient has to give care to kids or elderly family members and can't get to a nearby site in time, enrollment is still going to be a struggle. For those of us working in tech transformation around the patient, having a little humility and empathy around what we need to address from the non-tech side is number one.

Second is how we make it a human-centered experience around AI. One of the biggest opportunities to improve patient attention at the site is actually indirect—using AI to reduce all the non-patient-centric activity and smooth that out. When a site conducts a clinical trial, they have so many system entries to do, so many data follow-ups, so many interactions with sponsors and CROs. What if we lower all of that and augment it with AI agents and automation, so that when you look at the stack of work that happens, the site's predominant focus is on the patient? Those are under-appreciated aspects of what we need to get right. But in my humble opinion, it needs to start with the humility that there are a lot of non-tech things to address if our goal is to have more patients find trials, access trials, and stay on trials.

ACT: Where do patient-centered digital approaches have the greatest impact on trial timelines—recruitment, retention, data collection, or somewhere else?

Cheriath: If I think about where the biggest opportunities are from digital and AI transformation in clinical trials, it needs to start with smarter trial design. Thinking about the patient experience and patient enrollment after the protocol is locked is no longer the way to go. As you're designing the trial, how can we infuse AI-augmented intelligence around what a given trial design means for patient burden, site burden, enrollment potential, and operational implications? A smarter trial design is step number one, and from our standpoint, that is one of the biggest areas of investment and focus—helping sponsors make the best trial designs possible so that challenges are anticipated and solved before they become challenges.

Second is enrollment. AI's ability to scan through data, identify potential patients, match them more effectively, and generate high-quality referrals to the site—so that the site is focused only on the most likely eligible patients—is a huge area of opportunity. But it needs to be matched with patient-centered AI capabilities: a friction-free connection to the site, a smooth handoff between the patient's physician and the investigator, connected systems that give the patient and caregiver the information they need, where and when they need it.

After that, it is all about patient data collection. The bulk of the work that happens in clinical trials is collecting data, making sure that data is trusted, and being able to analyze it fast. I think we only scratch the surface of what we can do—from the ingestion of data from various sites to analysis with AI agents playing a significant role in augmenting human capacity. The potential for dramatic cycle time compression is significant. I am very confident that in my lifetime I will see cycle time compressions across trials by factors of 30% and 40%. It may not happen overnight, but the opportunity is there.

One final point—a not-often-talked-about patient opportunity with AI is in rare and ultra-rare disease. Areas like oncology and GLP-1 are data rich, so you can identify patients and the question is how fast you can analyze. But that's not the case with rare and ultra-rare data. One of the newest areas I'm intrigued by is the potential for synthetic data and digital twins of patients to act as a proxy for early-stage operational decisions where data itself is a rarity. I think that's an interesting angle on AI acceleration with patient centricity.

ACT: How do you see agentic AI realistically reshaping trial design and execution over the next few years?

Cheriath: I've been in this industry for 30 to 35 years and I've seen the tech waves come and go, some fads go by the wayside. On the spectrum from AI optimist to AI pessimist, I am an AI realist—and I believe this agentic AI wave is here to stay.

The future of this work is going to be a combination of human plus digital workforce. What I would encourage anyone in this space to do is critically look at the work being done and the different roles we play—clinical research associates, clinical data programmers, biostatisticians, pharmacovigilance specialists, all of these various roles—and ask: where will these roles be from an AI augmentation standpoint six months from now, one year from now, two years from now?

The way I think about it is a scale. At the bottom is augmentation level zero—current state, where work is done predominantly by a human using tools and assistance. Level one is where work is done by a combination of human and AI, but a human is still verifying and validating every output. Level two is where the majority of work is now being done by AI, but a human is still validating the output—the human becomes more of a verifier and certifier, but AI is doing the bulk of the work. Level three is where AI is doing the bulk of the work and a human is not verifying every output, but instead auditing and sampling—like an assembly line in a chocolate factory where you don't sample everything, but you occasionally random sample to see if it's still within quality guardrails. Level four is full autonomous AI agents.

I don't expect clinical trial roles to be in that level four category anytime soon. Maybe some administrative and low-compliance-risk tasks will get there. But I believe all roles in clinical trials will sit somewhere in the scale from level one through three over the course of the next two years, and that is significant.

This means we need a strategy around AI agents very similar to how we have a human workforce strategy. When you recruit somebody, they come in at entry level, get trained, become mid-level, then senior experts. We have to think about AI agents the same way—entry level AI, mid-level AI, expert-level AI across the whole range of tasks. This is not something futuristic. This is here and now, and it is part of the workforce planning we need to do today. The potential is huge, but it needs to be done with the right guardrails, because at the end of the day we are dealing with patient-centric data. If you do that, the opportunity in terms of total value from scientific progression is immense.