The SCOPE of Things Podcast

Raj Indupuri on Adoption Hurdles with AI Agents

July 7, 2026

banner-trenches-52

Clinical trial support companies are rapidly adopting AI, but sponsors have been more hesitant. Raj Indupuri, CEO of eClinical Solutions, gives the top three reasons behind the holdup in AI agent deployment. He explains the specific pattern of uptake for these agents, the required human-in-the-loop oversight, and how the agents can improve clinical trial velocity. Host Deborah Borfitz also gives the latest news on a highly acclaimed treatment for pancreatic cancer, filling evidence gaps in drug safety during pregnancy, guidelines for routinely collected health data for research purposes, an innovative portal aiding bidirectional communication with patients, and more.


Show Notes

News Roundup
KRAS inhibitor therapy for pancreatic cancer

  • Study in The New England Journal of Medicine
  • News on the Fred Hutch Cancer Center website

Target trial emulation in rural areas

  • Perspective in the Medical Journal of Australia
  • News on the Griffith University website

AI for improving drug safety during pregnancy

  • Report from the Journal of Medical Internet Research

Guidelines on the use of routinely collected data

New regulatory framework in the U.K.

  • Press release by the Medicines and Healthcare products Regulatory Agency

Participant Engagement Portal

Guest
Raj Indupuri,  CEO of eClinical Solutions


GUEST BIO

Raj Indupuri, Co-founder and CEO, eClinical Solutions
A technologist with over 25 years of industry experience, Raj Indupuri is responsible for establishing the eClinical Solutions vision and future-looking technology strategy. He is deeply passionate about fostering innovation to revolutionize the life sciences industry with ground-breaking technologies that will modernize clinical trials and bring treatments to patients faster. As an industry veteran who has been part of the evolution of life sciences and clinical data management for over two decades, Raj has an astute business vision to realize the digital future and enable progress and potential with data and analytics at the core of the company’s innovative products and solutions. Raj is responsible for the overall direction and management of the company and is a mechanical engineer with an MBA from Boston University who firmly believes data is the new fuel that will drive human progress.


TRANSCRIPT

Welcome And What’s Ahead

Deborah Borfitz

Hello and welcome to the Scope of Things podcast, a no-nonsense look at the promise and problems of clinical research, based on a sweep of the latest news and emerging trends in the field, and what I think is worthy of your 30 or so minutes of time. I'm Deborah Borfitz, Senior Science Writer for Clinical Research News, which means I spend a lot of time with my ear to the ground on your behalf, and a lot of hours every week speaking to top experts from around the world. Please consider making this your trusted go-to channel for staying current on things that matter, whether they give us hope or cause for pause. In just a few minutes, I'll be speaking with Raj Indupuri, CEO of eClinical Solutions, about lessons learned in scaling AI in real-world trials for some big name pharma

Rapid-Fire Clinical Research Headlines

Deborah Borfitz

companies. But first, the latest news, including a new, applause-winning treatment for pancreatic cancer, a trial emulation method for improving rural health research, filling evidence gaps in drug safety during pregnancy, guidelines for using routinely collected health data for research purposes, a new proposal from the UK's Medicines and Healthcare Regulatory Agency, and an innovative participant engagement portal enabling bidirectional communication with patients.

Breakthrough KRAS Drug For Pancreatic Cancer

Deborah Borfitz

Revolution Medicine's new oral KRAS inhibitor therapy for pancreatic cancer drew loud audience applause and a standing ovation at the recent American Society of Clinical Oncology annual meeting when phase three clinical trial results were announced, bringing fresh hope to patients and oncologists. The paradigm-shifting investigational drug, which the FDA has already granted expanded access, almost doubled life expectancy in the metastatic patients who took it compared to similar patients who had chemotherapy. Full approval by the agency is expected. The agent targets the undruggable KRAS gene that's mutated in over 90% of pancreatic cancers, which are generally diagnosed at an advanced stage.

Target Trial Emulation For Rural Research

Deborah Borfitz

Researchers in Australia have come up with a new method to improve health research in rural areas when randomized controlled trials aren't possible due to small populations, limited infrastructure, and workforce constraints. Their answer is target trial emulation using data that has already been collected during routine care to replicate the design of an ideal RCT using real-world data and thereby supporting the development of learning health systems. The approach is especially well suited to enabling telehealth workforce initiatives and point-of-care diagnostics to be researched quickly and rigorously.

AI To Close Pregnancy Drug Safety Gaps

Deborah Borfitz

A new report from the Journal of Medical Internet Research finds that AI can fill evidence gaps in drug safety during pregnancy based on two novel efforts: the Boost HP project using a tree-based approach to data mining, allowing decision pathways to be transparently traced, and the bionic study that combines causal inference and machine learning. The research approach lets AI do the heavy lifting of analyzing large data sets and by identifying safety signals and high-risk subgroups could enable more targeted and ethically designed trials that include pregnant participants earlier in the evidence generation process.

Rules For Using Routine Health Data

Deborah Borfitz

An international research consortium has published first-of-its-kind guidelines on the use of routinely collected health data for research purposes to improve the quality, validity, and transparency of studies based on data such as that from electronic health records and registries. Emphasis was placed on the risk of biased results as well as on the problems associated with missing and erroneous data, and the role of AI-based analytical methods. The study includes recommended actions researchers can take that would help prevent misinterpretations, increase the reproducibility of studies, and strengthen trust in results derived from routine data.

UK Rare Disease Regulatory Rethink

Deborah Borfitz

The UK's Medicines and Healthcare Regulatory Agency has proposed a rare disease therapies framework that introduces significant regulatory innovation to the country's rare disease landscape. At the heart of the proposal is a single investigational marketing authorization designation that combines clinical trial approval with a progressive route to marketing authorization that includes rolling data submissions, modular assessments, and earlier patient access. The guidance supports adaptive and innovative trial designs, accepts that surrogate or patient relevant endpoints may sometimes be appropriate, and acknowledges the potential scientific value of computational modeling, digital twins, and non-animal methods. The agency has urged the pharmaceutical and life sciences industries to weigh in on the proposals by the end of July.

Patient Portal Built For Retention

Deborah Borfitz

And finally, a new participant engagement portal developed by the Alliance for Clinical Trials and Oncology completely flipped the traditional digital health tool framework by focusing on retention and bi-directional communication with patients who are already enrolled rather than recruitment and onboarding. Among the 900 participants in the NCI-sponsored multi-cancer early detection biobank study who opted to use the portal, 84% reported having a positive experience, and 93% agreed to be contacted for future research opportunities. The portal, which has password-free access as well as accessible and multilingual design, was most successful at local community clinics, outperforming larger academic medical centers in patient enrollment rate and participation. The tool now will be expanding across more national clinical trials. As a reminder, links to the articles, studies, and press releases referenced in this month's news segment can be found in the show notes.

Three Real Barriers To AI Adoption

Deborah Borfitz

I am now delighted to bring to the mic eClinical Solutions Raj Indupuri for some practical real-world insights on overcoming adoption barriers with AI in modern-day clinical trials. Welcome to the show, Raj.

Raj Indupuri

Thank you, Deborah, for having me. Quite exciting.

Deborah Borfitz

Yes, yes. Thank you so much for joining us. As many of our listeners likely already know, there has been a huge upsurge in the adoption of AI agents by clinical trial support companies over the last, I don't know, months or years. But from my read, the embrace of these tools by sponsors and CROs has been more measured than frantic. That's probably a good thing. But it also suggests a hesitancy that I'm sure has a lot of sources. Raj, what in your conversations with industry players would you cite as the top three holdups in terms of simply considering deployment of an AI agent for any reason or perhaps some task-specific reasons?

Raj Indupuri

Yeah, the top three barriers from our perspective are data readiness, governance and trust, and operating model change. And to be blank, if your data foundation is fragmented, an AI agent does not solve the problem. It just scares the confusion. And as you are aware, our industry is quite fragmented where you have different sponsors working in a different way. Every CR is different, every trial is different. There's so much data that we have been collecting, right? The modalities and how you bring all this data is different. So it's quite fragmented. And if you don't have strong modern data foundations, AI agents would not be able to work effectively. And that's the biggest barrier from our perspective. And when it comes to the second one that I referred to, governance and trust in real-grid environment environments. Teams need to know exactly how an agent arrived at a recommendation. They need lineage, explainability, audit rails, and it cannot be a black box because at the end, you are going to influence safety and submission quality with these agents. So that's another big barrier. And the third one obviously is the operating model that I was referring to. And this requires a workflow redesign. And also I keep referring to internally with our teams in terms of how we need to change the way we work with these agents. And if I can elaborate on this, if you take data review, which is a big manual task with clinical trials, the real shift is from reviewing everything to focusing more on exceptions or signals, right? This requires an organizational or an operating model redesign and a workflow redesign. So we believe the organization is making the fastest progress are the ones that understand the execution layer is the real work, not the underlying model, not the point solution sitting on top of these fragmented systems.

Deborah Borfitz

Nonetheless,

Where AI Agents Get Early Traction

Deborah Borfitz

we do have a lot of agents out there. There's a whole bunch of them. And and probably we're going to get back to someone you're just talking about when we start speaking about the specific agents that eClinical solutions has as Illuminate agents, I think is how you pronounce it. I think there's like four of them now. Data review, of course, you just talked about data review, but also data mapping and risk-based quality management and study operations, all I guess part of your clinical data cloud. And I think only available starting this year, if I'm not mistaken. So I know these are early days, but but what has been the sort of pattern of uptake with these agents so far? And what specific ways are they initially to be used? Or are there areas where companies are particularly reluctant to give AI a role at all?

Raj Indupuri

Yeah, great question. So the clearest uptake from our site, what we have seen, is where the pain is already measurable. And with our Illuminate Clinical Data Intelligence platform, we provide different capabilities across the clinical development value chain and primarily focused on data review, data mapping insights, risk-based quality workflows. So, as you mentioned, so we we have released agents in Q4, or we started working and then we brought them to production in this year. And still early days, but we have seen significant interest and clear uptake in terms of both data mapping and data review. So, prior to this agentic workflows, so the amount of time it takes to map all this raw data to standardized data or some kind of a model for insight, it's quite cumbersome and time consuming. And our customers see tremendous value in automating data review using this data review agent. That's that's an area where we are seeing good success, and we'll continue to invest heavily into that. And the other big one is the example I gave earlier is data review. The amount of data that we're collecting is only increasing, and it puts an enormous burden on reviewers, medical monitors, clinical scientists to do all this review and also interpret this data and act on this data. So the agent or agents that we're working on, we have seen incredible success where users can use different personas or different stakeholders can use this data review agent or agents and quickly get insights and act on this data for faster decision making. So we have quite excited in terms of adoption with these agents. So our big differentiator with us is our agents they work together on the same governed data foundation. They're not isolated bots, they share the context, data lineage, and also the workflow logic so that the insight from one area can immediately inform action in another. So this is where I think we believe we'll differentiate from others. And also, this is where the difference would show up between experimentation and production, right? A useful agent does not generate an answer, it helps produce an answer that a clinical team can trust, explain, defend, and act on.

Deborah Borfitz

Okay, very good. Good explanation. I appreciate that.

Announcement

Are you enjoying the conversation? We'd love to hear from you. Please subscribe to the podcast and give us a rating. It helps other people find and join the conversation. If you've got speaker or topic ideas, we'd love to hear those too. You can send them in a podcast

Glass Box Governance And Trust

Announcement

review.

Deborah Borfitz

And I want to get back to something you sort of talked about from the get-go here. And this was, I was going to talk about my experience at the recent SCOPEX conference, where some of the most covered topics were the very things we've been talking about: governance, trust, regulatory readiness, sort of things, and the importance, especially, of humans remaining central to oversight and final judgment. You know, that was like a big recurring theme. Presumably, this reflects the prevailing concerns of pharma companies. So, what would you bring to the table on these sort of trending talking points beyond what you've shared already, particularly the human and the loop piece?

Raj Indupuri

Yeah. This is the conversation that actually determines who wins. And this is where I believe you're alluding to govern governance and trust cannot be just check boxes. They're the foundation for the whether AI ever moves beyond pilots in a regulated and regulated environment like ours. So we believe in a glass box governance for every agent and for every AI capability that we build. By glass box, I mean every recommendation has to be traceable. Need to provide visibility into what data the agent used, what rule or reasoning path is followed, and then where a human made the final decision that needs to be logged and audited as well. Another thing I should mention when it comes to our company and our eliminating later intelligence platforms. Again, we have a map. We've been working on this for more than a decade in terms of our data pipeline. And it's this common data that we have access to. But also, we have a lot of experience with our biometric services. So that execution expertise really helps build agents that are more trustworthy and also more regulatory ready.

Raj Indupuri

So that's another an important aspect when we are building agents that need to work in production. That service execution knowledge needs to be encoded into the agents, and that's what we are able to do with our experience. And also when it when when it when we talk about a glass box, we are also striving to provide our customers visibility into the context that has been layered, and also we are enabling to them to layer their own context. Because, again, like I mentioned earlier, every sponsor is different, right? So they have their own layer of context that's that's critical for agents to perform really well in their environment. And also, the other thing is all around evaluations. You'll hear this word e-valves with agents. Again, before you actually use them in production at scale, you need to ensure that these agents work really well. And that's done through a process called evaluation or e-valves. And we are providing our sponsors or customers with an ability so that they can do e-valves on these agents in a self-service way. So that's the reason we refer to this as class box AI. So then when agents are embedded into validated workflows, like what we have with lineage controls and explainable outputs, they can reduce review restrain instead of creating more work. I thought that foundation AI would become another reconciliation problem and a burden for customers, which we are which we are avoiding.

Human Oversight In Validated Workflows

Deborah Borfitz

Okay, I want to get back to the human oversight here, just sort of dig in a bit more there. And like, where would you say that these agents are most needing human oversight, you know, to arrive at something, yeah, a document that's accurate and complete? And has this been witnessed in in real-world trials that you're basing it on on speculation or something that is actually seen and witnessed and know is is the case and a problem or a concern?

Raj Indupuri

Yeah. So the agents that we are deploying are not autonomous agents. Because again, we actually have work in a regular environment. We are dealing with patient data, right? It's very important that there is this human in the loop. So our objective is to have these agents invalidated workflows and automate, help automate as much as possible. But there's always human in the loop for actual judgment, right? Because the context, right, for a trial across data sources is it it there are nuances, right? So we want to make sure that the final accountability for safety and integrity or regulatory submission quality is still trusted with humans. So the goal here is to not to remove humans from clinical trials, right? The goal is to stop wasting expertise or expert human work, right? That machines can do better. And especially with crypto tasks and a lot of workflows that we have already seen that could be automated. So AI should handle this volume of data review, and in this case, again, going back to review pattern recognition or any signal detection, and humans should own judgment, interpretation, and final accountability. This is the this was the operating model shift I was referring to earlier.

Deborah Borfitz

Okay, very good. Yeah. All right. Obviously,

Time Savings From Mapping And Review

Deborah Borfitz

AI agents have a vast potential in terms of improving clinical trial velocity. You know, what sort of time savings can we reliably attach to different tests done in conjunction with AI agents? Is there any case studies you could point to or um just some anecdotal anything?

Raj Indupuri

Absolutely. So I'd be careful in terms of making blanket statements or saving claims because again, the baseline varies enormously by each sponsor or each customer and the study design and complexity and also the maturity of that particular data infrastructure. But it has been really consistent with at least the agents that we already have deployed to production. Again, one, the mapping automation agents and the data review. So, for example, with the mapping, it it takes usually weeks to several weeks, it could be eight to twelve weeks to do the mapping of this raw data to standardized data for either submission purposes or for advanced downstream analytics. And we have seen that with our genetic approaches and this mapping automation agent, we're bringing down it to two to three weeks, right? So that's substantial reduction. And as we continue to innovate, we believe this will be down to literally days, right? So now you're trying from something from eight to twelve weeks down to literally like a matter of days. That's a huge benefit. And especially when trials continue to change, right? So we are in a we are in an environment where there are so many protocol amendments that happen for different reasons, right? It's not all bad. It could be also for doing trials in by learning in real time and making amendments.

Raj Indupuri

So you continuously change the data structures that requires this continuous mapping. So over a period of time during the entire conduct, that leads to significant efficiencies. The second or the two other cases are data review again. The initial feedback I get from data managers who are using this is what used to take hours or weeks, right? Or I would say days when you're looking at certain data sets to interpret. Now it happens in minutes, literally, right? So if you're if you're looking at a visualization or set of charts or long data sets or large data sets, right, which have thousands and thousands of records, using these agents, you can actually do analysis and focus on actions in minutes, literally minutes. So this would all add up when it comes to conducting a trial with these agents. And over a period of time, this could turn into significant efficiencies and substantial compression of cycle times, and not to mention the quality improvements, because majority of the work is done by these agents, right? And the humans would be involved in final judgment, escalation, and accountability. So again, still early days, but but we are seeing significant results. I would I would expect in the next few months, right? This efficiencies or these savings can be more quantified as we continue to accelerate and innovate.

Next Wave Agents And Protocol Intelligence

Deborah Borfitz

Yep, and I will be looking for that. You know, and I'm guessing also that eClinical solutions plans to build a larger team of AI agents beyond the four current ones. And if that's a correct assumption, where can we expect to see some more of these, I don't know, new hires, if you want to call them that in the months ahead?

Raj Indupuri

Yeah, that that's a great way to think about it, right? So the idea here is these agents become your thought partners or team members, like you're referring to new hires, which is really cool. Yes, absolutely. So we are all in into AI and agents, and the unit agents that we assign what that we announced are across four product pillars. We have four product pillars. One is data pipeline, which is a data mapping one I'm talking to, data management analytics, the data review one. The other one is quality and risk management, another around an agent to support risk assessments and then operations. So the strategy and and the roadmap that we have is to build agents across these pillars. But what's exciting is as we're working on agents, we realize that there is a need of a fifth pillar, product pillar, and which we are referring to as protocol intelligence. This is something that we are going to announce very soon, but we have made a ton of progress. The idea here is when you are when you're executing these agents or when when you're building these agents, the agents require significant knowledge and also they need phonetics. And as you're familiar with clinical trials, the protocols have a ton of knowledge, right? So now what we've decided to do is build protocol intelligence as another pillar, and you're building agents so that you can extract intelligence from the protocol and then combine with all other agents, and this becomes this control layer, and it will connect all these pillars so that at the end of the day, they'll all work together, and the outcomes that you are going to achieve will be quite substantial than looking at this in a fragmented way, right? So the upside for PP is not just productivity, it's going to elevate every role in clinical development, data managers, reviewers, statisticians, biometric leaders. It should be spending less time chasing discrepancies or collaborating with different teams and more time interpreting on risk and quality and patient impact. So we're quite excited in terms of where we are headed. And again, we already have this competitive edge where we have this governed data and the validated workflow. And also because of service experience, we have this significant domain context knowledge. And we believe this approach will really help deliver significant outcomes for our customers. And this is where, again, pilots will move into production and the agents will be adopted and implemented at scale in the near future.

Closing Thanks And Scope Europe Invite

Deborah Borfitz

Yeah, sure. We'll be seeing similar stories across the industry. Thank you, Raj, for casting this vision of AI agents as sort of approachable digital teammates rather than threatening competitors who may just need a bit of managing and mentoring to get up to speed on their assigned tasks. It's a new day that soon will stop feeling like it and be the baseline operating standard, right? Good luck to you, any clinical solutions on the journey into this still largely uncharted territory of human-agent collaboration.

Raj Indupuri

Thank you, Deborah. I really enjoyed this conversation. I really loved the way you were phrasing agents as teammates or new hires. I agree. I think this is one of the most important shifts that are happening in clinical development. And I appreciate this chance to discuss where the barriers are unlocking.

Deborah Borfitz

So, I appreciate you taking the time. And as always, a big thank you to everyone out there for listening in. If you're not subscribed to this podcast yet, please consider going to Apple Podcasts and doing so right now so you don't miss your monthly dose of news and perspectives. You'll be hard pressed to find anywhere else. And if you're up for it, I'd also be so very grateful if you'd leave a rating and review while you're there. One more thing before we go. If you like today's conversation, it is only a glimpse of what you can expect from Scope Europe presenters and panelists. Please plan to join us October 13th and 14th in Barcelona, when clinical operations executives will be exploring the latest trends in clinical trial innovation, planning, and operations. Save an additional 10% off any current rate by using the code SOT 10. For more information, visit Scopesummit Europe.com. Bye for now.

Stay Connected

Follow us on Spotify

Meet the Host

Deborah Borfitz

Deborah Borfitz

Deborah Borfitz serves as host of The Scope of Things podcast. She is also senior science writer for Cambridge Healthtech Institute and is the lead contributor to Clinical Research News, Bio-IT World, and Diagnostics World News. Deborah has a long and varied career in journalism, much of it as an independent writer with a heavy focus on healthcare and clinical research. She was introduced to the world of clinical trials 25 years ago by advisory board member Ken Getz and in 2001 co-authored a book with him on the informed consent process.


Learn more

Clinical Research News Online