How Agentic AI Is Transforming Clinical Research Site Operations

Jimmy Bechtel sits down with Brooklyn Montgomery, Business Development Director at Deep Intelligent Pharma (DIP), to explore how agentic AI is reshaping the way clinical research sites operate. Brooklyn breaks down where AI delivers real value for physician-led research, what meaningful adoption actually looks like at the site level, and the three questions every site should answer before spending a dollar on AI tools. The conversation also covers how mentorship and technology work together to close the gap between academic research and clinical execution, and where sites are headed as they move from execution partners to upstream contributors in trial design.

Disclaimer: The content of this article reflects interview perspectives and is intended for general informational purposes only. It does not constitute any commitment, representation, or guarantee regarding the performance or outcomes of any product or service. Actual results may vary depending on factors including, but not limited to, study design, data quality, regulatory requirements, and implementation conditions.

Transcript

Jimmy Bechtel 0:00

Well, hello, and welcome everyone to SCRS Talks provided by the Society for Clinical Research Sites. I'm Jimmy Bechtel, the chief site success officer with the Society, and today I'm joined by Brooklyn Montgomery, who is the business development director with an organization called Deep Intelligent Pharma. Brooklyn is here to talk with us about, a very current and hot topic, agentic AI, and how we can use this to empower some of our research sites, and a little bit of, insight into what's being done in this space. But before we get into those questions, Brooklyn, I'd love to learn a little bit more about you and maybe a little bit about the organization as well.

Brooklyn Montgomery 0:38

Absolutely. Oh my gosh, thank you so much. I'm very excited to be here. I love this opportunity, truly. So just kind of diving in, my background is a bit interesting, not gonna lie. going through, life, just college, ended up joining the military, and became a data analyst in sorts, through aviation and helicopters in the Marine Corps. And when I got out of the military, I wanted to do bigger and better things, and at the same time help people. And I got an amazing fellowship at a small CRO in San Diego, moved through everything with their CEO, and went through and did business development, got my MBA, and kind of just learned on the job and saw on the job how a lot of CROs are unable to help in these processes and move the pin in areas where you're bridging the gap in, money, in funding, in foundations, in getting trial to patient care. And it just really brought me into this passionate community of wanting to do more, but not having the tools. And now with DIP I feel like I have the tools. So that's just kind of very brief on me. always on LinkedIn, so if anybody ever needs to talk to me, I'm here.

Jimmy Bechtel 1:47

Awesome. Thanks, Brooklyn, and I'm really looking forward to talking about, this subject today. We know that physician-led research particularly has faced a lot of, specific and unique barriers. So in that space, maybe starting our conversation off, where do you see AI making the biggest difference in helping sites and investigators move faster?

Brooklyn Montgomery 2:06

Oh, yeah. I love that question honestly because-- I mean, to start it off, we really are in the simplest way, we are pharma-specific AI workflow partner. We don't wanna replace anyone at the sites and/or other areas. We want to be an amazing tool. so really just for those long-standing barriers for physician-led research, et cetera, where is AI actually going to make that difference and not just become something creating more problems? So here's the thing about really physician investigators. They usually have the best ideas, the deepest understanding of their own patients, and that's never the problem. The problem is what happens next. So really turning their insight into that protocol, into the regulatory strategy, and a study that you can actually run, how do you do that? So that can take months of fragmented manual work, and that can really create its own barrier that's-- has never been a shortage of insight. It's really just a shortage of structured support around that insight. And that's exactly where I kind of plug in, where AI makes that big difference. So it collapses that early document-heavy stretch. It can structure, the concept, review the literature, compare endpoints, catch the inconsistencies, draft protocols, connect all of that forward movement to CRF design and statistical planning. It's areas that people joke about, when did this have to be a meeting or did this have to be an email? I wanna get a mug now that says, "This meeting and email could have been AI." Because truly, we're moving along in the process, and we're wanting to make things faster, but not only faster, but human QC'd and better and having actual movement to get to the patient. So really the e-expertise does reach an executable study far, far faster. So to really be clear, AI does it for you. It's just time to clarity. So you surface the design and feasibility issues early, and then we have, that cheap fix in a way instead of late when they blow up the whole team's timeline or make everything super expensive because it's been months or sadly, sometimes years. I would really just say like all, especially for acute, for physician-led and academic origin work, it would be incredible to move more into those areas. And the win isn't automation, it's just early issue detection, so problems can be caught when they're small and not as big. So it's am-amazing anecdote, and I think that it's moving into clinical trials faster than we think. But there's going to be growing pains and, Deep Intelligent Pharma has been doing this since twenty seventeen, and we've been past our growing pains, and we just want to s-share the love and move the trials along for great and passionate companies.

Jimmy Bechtel 4:44

It's an interesting perspective, Brooklyn, not one we talk a lot about, approaching AI. So what does... moving on, what does meaningful adoption look like at the site level, and how do we know when it's, pulling on some of those threads, how do we know when it's working like it's supposed to and maybe in some of the ways that you described?

Brooklyn Montgomery 5:01

meaningful adoption almost never actually looks like an AI transformation. It's much quieter than that. it's more like one high friction or lower risk workflow, and it's just done noticeably better. so if we could even start in smaller examples, it would be protocol and amendment comparisons, eligibility extraction, feasibility first drafts, document summarizations, consistency checks. pick one that hurts and do it better using AI and having the human always QC and always make the decision on how to move forward. So that's how you know it's working, is when you can actually measure it against your old manual baseline. You review the hours saved, faster turnaround, more inconsistencies caught, So like I said, human QC never removed. So the goal was never more AI activity, the goal is better, more reliable site process, and that's just the whole name of the game truly. so just kind of picking what hurts, doing it better, and then seeing the outcome. And I mean, DIP, we can also do the an- analytics on those outcomes as well if they were ever needed, just to make sure that you are moving the pin enough and spending, you know, what little money you do have to spend in a CRO or a vendor or a support situation to make it better

Jimmy Bechtel 6:19

Yeah, Brooklyn, it seems like a pretty straightforward and, logical approach to understanding this, is when you can measure and see the difference that you're targeting going about. to summarize what you said there, it's kind of this making sure that we implement the systems and the logical, methodical approach to what is the problem, what are we trying to, or, and/or what are we trying to solve for, what does good look like, right? What is the outcome? What is the goal around that? And then how do we effectively measure that when we implement said process or tool or AI, you know, agentic AI system to work on that solution, is really what, y- this, what I took away as the kind of the summary for what you were saying there.

Brooklyn Montgomery 6:56

Absolutely. And honestly, It's not just an Excel sheet or a protocol document or drafting. It's basically we know that AI could look impressive, but it's if the coordinator's process is faster and more consistent and still fully under human control, is where we're winning, basically.

Jimmy Bechtel 7:13

So then what should sites be thinking about before they dive in, and what are some of those mistakes that you see often, especially when we're talking about what you just shared, this sort of methodical approach to it?

Brooklyn Montgomery 7:27

So, before a site even spends a dollar, I would have them answer just their, I would say, three top questions. One, which single workflow hurts the most every week? So what is constantly a pain? Because I know a lot of sites go through the process of telling, their sponsors what is hurting and get ignored or pushed to the side. Or say, "Don't break the wheel," kind of process. So basically, which one is hurting and keeps hurting? the second one would be two, who is the named human reviewer? So who is the person you would go to for these reviews or assigned? And then also three, what's that one success metric that tells you whether it actually worked? And then if they do get those three kind of hammered out along with their team and they've dodged most of the pain out of there, it's really because the common mistakes are just the mirror image of those questions. So trying to transform everything at once, dropping the sensitive data into general purpose tools, which never a good thing. and then also treating AI output as the final annotation, not great. Always a human's QC always a smaller and shorter process of reviewing, but still a human decision. And then also automating that process is, it's poorly defined to begin with because- you just are going to create a faster mess, and you skip that audit trail when you don't ask yourself these questions, and when you don't want to do things piece by piece instead of all at once. So none of these are moral or, you know, incorrect or any failings by the site in any way at all. These are normal early stumbles, and the fix is almost never more technology. It's just sequencing and governance. And sequencing and governance can be a large, and should be the largest portion of AI. So I think in those areas, I would say if you're a site, ask yourself those three questions, jot it down on a sticky note, put it on your computer screen, and see if it's the same problem every week, if you're going to the same person every week, and see how you can create solutions using small pieces of AI, small pilots with their own firewall, secure ways, not just through online open AI. and see if you can alleviate the process in that way, and, cheaper pilots are just a great way of doing it with great companies right now.

Jimmy Bechtel 9:51

a lot of great points there, Brooklyn, right? Taking it one step at a time, identifying the low-hanging fruit or the easiest way for you, to implement that. But you know, start with the highest impact and I it's also balancing that with something that's high impact but isn't gonna be this giant daunting process, right? Which you alluded to in that We wanna address something as a pain point or as a challenge or maybe a time suck, but not something that's going to be this, massive insurmountable task, right? As a place to start, we need to strike that balance as we look to identify those problems, and that's probably a group activity, right? Like, as a site team. You know, talk through those things with your colleagues and work on how do these things affect globally, and what types and aspects of that work are we maybe not considering when we go to implement some sort of different solution,

Brooklyn Montgomery 10:39

No, I completely agree. I love how you put that because honestly, it's, there's so many gaps where people try to have AI create this transformation project. But when you have AI create your grocery list, you're not able to go grab all the food at once. When you have AI create a workout regime, you're not able to get the results the next day just because it was AI generated. You have to prove the value would be there and then earn each step. So that's kind of what those three big questions would really help for a site internally reviewing on that process.

Jimmy Bechtel 11:11

Excellent. And so, Brooklyn, there's a gap between academic research and clinical execution, and we can talk about that all day. But, focusing on the topic at hand here, how can mentorship and technology work together to close that gap or, or at least align them more effectively?

Brooklyn Montgomery 11:31

Oh, absolutely. So I love all types of mentorship. I think that it can be handled and held in every area of the world and workforce. And honestly, the gap really just comes down to academic excellence and clinical execution are two genuinely different skill sets. So you can have a brilliant investigator with deep, hard-won tactic knowledge, and that same person can hit a wall trying to turn it into a protocol or a regular to- a regulatory strategy, and a study site can actually run. So it's not a knock on them, 'cause I have amazing PIs that I would still call to this day. they are the workhorses of this industry. They move the pin, they move things along, and they feel all the struggles, and it's just honestly a different muscle, creating these, these areas that truly I feel like they should not have to spend all of their time creating. They have other tools and avenues of getting things done, and I want more of their time in reviewing data and patients and input, and especially when sites often have multiple clinical trials. So for technology, I would say that AI does accelerate the research and the drafting. And then for mentorship, experienced people help decide what's scientifically meaningful, what's operationally realistic, and then what's actually worth the advancing. So you really honestly need both. So AI without that judgment just gives you fast noise, and then mentorship without AI just stays slow and stays at the baseline that we're at right now or earlier on in time. So the combination, it just moves things along and does create strength together and makes it a joined skill set

Jimmy Bechtel 13:14

I couldn't agree more, Brooklyn. A lot of really interesting points, around what that means and, how we use each other, right? Lean on each other and also implement technology to get to a better place ultimately for the patients that we're trying to treat. And, because in the end, when that disconnect exists or we have challenge in that space, That's where the struggle really starts to become apparent

Brooklyn Montgomery 13:36

I couldn't agree more. at a few conferences that I've gone to, I've noted some of the harder spaces that you put yourself in in this industry, going to gene therapy summits, or going to rare disease foundation gatherings and networkings, and you realize that those parents are the CEOs, and they are fighting for a trial that is, the only thing they're up against is time. And I think that if we're not working hard enough to make that timeline smaller for the patients, especially the ones that don't have that time, then we need to be working harder at that. So really it's just A- AI will accelerate the work, and mentors make sure it's the right work. So that combination is really how the good idea in a hospital or a foundation or, at home with your child that's been recently diagnosed becomes that study that a site can actually run without hitting so many walls and hardships,

Jimmy Bechtel 14:31

And another excellent point, Brooklyn. So I wanna conclude our conversation today. and, learn a bit, a little bit about what you all are building and how it speaks to everything we've been talking about, and where you see the site's role ultimately then heading as all of this continues to evolve.

Brooklyn Montgomery 14:48

Yes, so honestly, just if you put all of this together, for what is DIP actually building, since 2017, where are the ties, like, get that conversation all together from everything we were talking about overall. And basically just kind of tying in everything overall. So when we do that, we want it to make sense, correct? So the sites are moving from being mainly execution partners to being earlier design and evidence generation partners, and that makes sense, but the sites understand the patients, the recruitment realities, the workflow constraints, et cetera. So pulling this together for Deep Intelligent Pharma DIP, as l- our little acronym, we want to build that workflow and mentoring layer that lets sites contribute at that earlier stage, and then that goes through all the areas we've spoke on, protocol design, feasibility, investigator led trials, continuous evidence generation, But we want it to be global, we want there to be access, we want there to be, security for each trial to its IP and sponsor. But at the same time, we want them to be able to move the timeline along, and be able to reinvest the money that they would be spending in that time in themselves, in their trials, in their sites, in their patients. So- This will always be our number one goal, is just moving that pin with AI. So just connecting back, we have niche access. We have ph- physician-led, Japan and global demonstrated success, the proof that it works, and we're moving into that forward global movement and sites as that upstream partner. And honestly, what the sites should be doing today is, of course, document your workflows, improve the data governance, train your staff on responsible AI use, and pick a measurable pilot. bring it up to your sponsors. See if it fits into your budget or ask yourself those three questions and bring them up at every meeting that you can. it's sad that the PIs and investigators, et cetera, have to be their own advocates, but I promise you, DIP is advocating for you. When we do have these meetings, we want it to be easier. We want it to be cheaper. We want it to run better without removing the human touch. So really just from shaping it at the site side, it's completely doable, and it can be successful. It's just another step to the day, and we wanna make that part as easy as possible with, our mentorship but then also our agentic AI model in its most general terms.

Jimmy Bechtel 17:17

Well, thanks, Brooklyn. I think that's a really excellent place for us to conclude our conversation. Thank you for sharing the insights. It's really interesting to learn not only about what you're doing, but also some, unique perspectives on the future of where we're headed and some things that we need to consider as teams at the sites to be able to bridge some of these gaps and move forward effectively. So thank you for sharing those with us and being with us, here today.

Brooklyn Montgomery 17:42

I truly, truly appreciate it, and Deep Intelligent Pharma, they were extremely excited about this, so appreciative, and we love being in areas where we can touch back to our roots and our passion in it, which is with the sites, with the patients, and moving timelines. So very much so if you're listening, if you even have a question, please reach out. Our BDs, business developments, are extremely capable of giving you any answers and linking you with the right people, even if you're curious of how to move forward in the smallest thing as a pilot, because we want the sites to be able to have those tools, and we love the sites. So always here to work and partner and always here to support.

Jimmy Bechtel 18:20

Thanks again, Brooklyn, and, thank you to, DIP. For those that are listening, make sure to check out other site-focused resources, like additional podcast series and opportunities to hear from our partners through webinars, news articles, and SCRS events, like our Site Solutions Summit on our website, myscrs.org. And again, for those listening, thanks for tuning in, and until next time.