Video transcript: Kelly Hogan: [0:01] Hello, and thank you for attending today's webinar, From Pilots to Practice: Scaling AI Across Clinical Trials, presented by Medidata and Fierce Pharma. My name is Kelly Hogan, conference producer with Fierce, and I will be your moderator. Before we begin, here are a few housekeeping items. [0:16] To learn more about our speakers, click beneath each of their names to read a detailed bio. In the resources window, you'll find the 2026 State of AI in Clinical Trials report and Medidata Plus fact sheet available for download. If you accidentally close either of these windows, you may reopen them by clicking the icons on the left. This webinar is being recorded and will be available on demand. And finally, please submit your questions as you think of them via the submit questions window. [0:53] Now, the industry is no longer deciding whether to use AI. It's already seeing real value and now grappling with how to scale it responsibly and turn successful pilots into practice. Today's discussion will explore that shift through the second edition of Medidata's State of AI in Clinical Trials report conducted with Everest Group and drawing on responses from senior clinical operations leaders. The report offers a clear view of where the industry stands today and how far there is still to go. We'll use those findings alongside Worldwide Clinical Trial's own AI adoption journey as the backbone of our conversation. [1:34] It is now my pleasure to introduce our speakers. Joining us today are Nik Morton, chief strategy and transformation officer at Worldwide Clinical Trials. Nisarg Shah, vice president at Everest Group, and Tom Doyle, CTO of Medidata. Now let's get started with the reality of clinical trial AI in 2026. Nisarg, this is the second year of this research. What's the biggest shift you saw year over year and where does the industry actually stand on the adoption curve versus where it thinks it stands? Nisarg Shah: [2:10] Yep. So thank you, Kelly. As you said, this is the second year we looked at ... We surveyed around 200 participants across pharma CROs globally. So I think the biggest shift is the conversations going from, where can we use AI to where can we scale AI? And we do see that this is a conversation which almost everyone is asking nowadays. [2:35] I think a few points will stand out. One is the adoption curve is also accelerating, which means that there are more pharmas and CROs evaluating AI to deploy in their environment. I think the 2025 data point was around 20% evaluating AI, now it is 30% evaluating AI. And similarly, if you look at the enterprises using pilots in their environment, it again goes from 22% to 28% in 2026. [3:08] So I think both parameters point to a positive change in the right direction, and it is also a conversation about, which is the next shiny object to ... Which are the use cases which will likely yield value for us in the near to long term? The data also points that there is still a lot more ground to be covered because only 28% enterprises are actually thinking about or using AI in their environment. And those are some challenges ... Or those point to some challenges which need to be solved for. Governance and compliance being the biggest one, followed by the scalability of AI going to ROI and to some extent also the integration complexities which they present in large scale deployments. [4:04] So we feel that a lot of these conversations are also about making it real for everyday users and also how to take the needle from what we have right now to, what is the art of the possible in the future as well? And I just want to touch a bit on, what are the kind of use cases which we saw coming out from the results. So a lot of the use cases in 2025 were about data interpretation, data analytics, those still continue to be a good portion of deployments, but now they're also looking at workflow based automation. They're also looking at insights and reporting. [4:48] There are some highly complex use cases which are also coming up the curve, which we'll also discuss in a few minutes. Trial simulation, protocol design, authoring and so on and so forth. So net-net, there is tremendous progress happening in the field of AI and also the industry recognizing that it's no longer just the next shiny object. It's something which has real materialistic value for everyone out there in the space. Kelly Hogan: [5:23] And Nik, does this picture match your experience? Where did Worldwide Clinical Trials see itself on that curve when you started this journey? What barriers felt most real to you? Nik Morton: [5:34] Thanks, Kelly. Honestly, the report matches our experience surprisingly closely. I was really interested to read the report and see that most organizations are on that journey between exploration pilots, early implementation, and that's exactly the journey that we have been taking over the past couple of years at Worldwide. [5:57] When I think about it, there's never been any shortage of enthusiasm to get up the curve as quickly as possible, and there's not been any shortage of opportunity. Every week there seems to be a new AI company with a new solution or recommended pilot opportunity that was going to solve all of our problems. So that wasn't the challenge. The challenge was in the opposite direction, figuring out which AI actually mattered most to our company. At Worldwide we probably reached that inflection point about 18 months ago. Our growth was increasing, our clients were becoming more interested in AI, and internally our teams were really enthusiastic about checking out these technologies and how they could help our study execution. [6:47] And at the same time, we were seeing this flood of outreaches from different vendors and different tools. The real barrier is not the capability of the AI. The real barrier is the fragmented data, the disconnected systems that acquire integrations, the governance and associated financial decisions, and then operational change management. And that's what came out in the report. [7:15] You quickly realize the most impressive AI capability will struggle to generate meaningful value if the underlying data sits across multiple systems, multiple processes, multiple teams. I think that was one of the biggest lessons for us was operational outcomes matter far more than the technology. Therefore, you have to shift your thinking and say, "What's the key problems we are trying to solve and how do we create the environment where AI can help solve that repeatedly and at scale and not focus your thinking on the next copilot that you want to take on?" Tom Doyle: [8:00] I agree a lot with Nik's experience and what he's saying. Clearly there hasn't been a gap in excitement, a gap in investment, but there has been, and what we continue to see is a readiness gap that's largely in the infrastructure, what Nik pointed to around the data infrastructure and preparedness to really take the full advantage of this opportunity in front of us. We can get by in the short term with incremental improvements without having large, let's say investments or cleanup of the data infrastructure, but we'll never fully unlock the full value of AI, of agentic operations, the agentic workflows that Nisarg was referring to, without ensuring that we've invested properly and we've created the right semantic layer, et cetera, to expose our underlying data most effectively to agents. [8:54] And that we start to see in the bifurcation of success that our customers are having, that our industry is having. Those who have well-prepared for this moment, who have made investments in their underlying data infrastructure are able to move more quickly and realize more gains, while those who haven't are seeing incremental improvements, they're seeing some success in POCs, but as Nisarg was pointing to, they're still wondering, "How do we get scale? How do we really get that true return on investment?" Kelly Hogan: [9:20] And Nisarg, the early adopter data is significant, but why is the performance gap largest and the hardest KPIs and what does that tell us about how AI value actually accrues? Nisarg Shah: [9:41] Sure. There were approximately 20 to 25% of respondents who had deployed AI for more than 18 months in their environments. We asked all the respondents to rate, are they seeing improvements in various KPIs? And those are the usual suspects, time, cost, quality, and then you can have a lot more say protocol error rates, FD hours saved and so on and so forth. So I think as you said, the data presented a divide of sorts. The early adopters rated significant improvements in all the KPIs compared to those who were earlier on in the curve, so less than 18 months or so. And that's because of a few reasons, we've already discussed some of those. [10:33] One is some of the data may be siloed, which means that it's difficult to orchestrate across say end-to-end deployments and get results at scale. And the other is also about the learning curve, the governance and compliance conversations as well. So what we're starting to see is a few things. One is those who have been deploying have already come up with a set of use cases and compliance measures to ensure that they scale in their environment. They've also been able to figure out a way to unify all the data silos which exist in the organizations. [11:18] I tend to put it this way, the more silos you have, the more hurdles you'll likely run into when you're trying to scale AI because pilots or siloed pilots can succeed, but the moment you try to scale them, they will eventually hit a wall because you're trying to solve for a lot of factors, which were not foreseen during the pilot itself. It's like trying to drive a Ferrari on a potholed road. I've said this a lot of times. The non-linear results can be seen once some of the top challenges can be worked out and the top two are the governance and security conversations, and the second one being, how do you solve for integration? [12:02] So the early adopters, they've found a way to ensure that they solve for these two at the very least. And that's where we are starting to see conversations go from, "All right, we have early success in this. How do we translate it into say another use case where learnings and best practices can be shared and cascaded, and not just individually, but through the entire ecosystem working in unison?" Kelly Hogan: [12:31] Thank you. Now you talked about moving from pilots, and I think this next question is something that a lot of people are wondering about. Let's go with Tom here. As AI moves from isolated pilots to core trial workflows, how can sponsors maintain appropriate oversight and who ultimately holds responsibility when an AI model output incorrect information? Tom Doyle: [12:55] I think that's the major question facing us today across industries, and you see a lot of this as we move to more autonomous workflows, autonomous technology that is really all around us. In life sciences, of course, that's paramount because of the risk there is to patients, the risk there is to trials, and we've always been very focused on who really is managing and accountable for that risk. [13:22] Regulators has been fairly consistent throughout time, through the inception of technology, that really we are largely technology-agnostic, when it comes to risk. The risk and accountability lies with sponsors and with the PIs, with the care teams, because they're in the best position to understand what the impact is to patients. That isn't to say that then the entire ecosystem around them doesn't bear an increased responsibility to make sure that we deliver on that accountability. And you see that a lot with technology providers now being more focused on how they provide more effective oversight, more effective visibility into the decisions that are being made. [14:08] In life sciences, we've never been a black box type industry. In clinical research, we have always been very focused on evidence. That is the nature of our business, so to speak. And so it is natural to then expect that that same expectation would come to agentic AI, to AI broadly, around making sure that we are transparent in why a model is making a decision that it is, how we are deriving the inference we're deriving, and full capture of all of the activities, the audit trail, if you will, of everything that was performed. So that those who ultimately have the responsibility of reviewing that and making sure that is in the best interest of patients and following, for example, a trial protocol. That they're successful in being able to do that. [14:53] We of course need to deliver now more capabilities to enable and support those who have to make those key decisions because the scale and speed that we aim to run under, means the old ways of providing that level of governance and oversight simply won't work. And that's really the moment that we are in now. You see as much of an acceleration in adoption and excitement in what AI can bring us in terms of bringing new insights and trial acceleration, for example, ultimately with the goal of bringing new therapies to patients faster. [15:25] We similarly see a lot of new advancement and new focus on how we're providing new mechanisms to govern and to oversee those processes. And I think it's the coming together of both of those things that will really unlock the true potential. [15:46] Nik Morton: [15:46] Tom, we've been talking about this a number of times recently. I think our industry's comforted by the human in the loop in 2026, 2027, maybe beyond. But we do expect these models to learn from those interactions with the humans. We do expect that over time we might get to a situation where we are convincing ourself that we can be more trusting of the models and the recommendations. And one of the things that you and I have been talking about is, how do we make those probabilities transparent and almost be coaching ourself as we go forward? [16:23] The audit trail is the key data for us to collect, but I'm also keen to collect the behaviors of the humans and when are they accepting the recommendations? When are they overruling the recommendations? What does those probabilities look like? How do they improve over time? How does the behavior change over time? There's a lot for us to learn, whilst we're in this comfort zone of the human is making the decisions and we're sort of holding that as our principle for now. I'm fascinated to see what we can learn from the audit trails that we collect. Tom Doyle: [17:00] Yeah. Nik, I couldn't have said that better. I think you're absolutely right. Automation has always led to the risk of cognitive atrophy. So the more we enable technology, the more we trust it, the less we are building the muscle to continue to understand how it works. In the past, the risk profile looked one way, and now as we look into the future, we can see a much bigger risk profile, when we think of what the possibilities are in really providing clinical decision support, moving into things closer and closer to patients. And it really does test then our ability to deliver new ways of working, as you're referring to. [17:35] Looking at the behaviors of those that we entrust to govern these processes and really help them be successful with the transparency of what has occurred, why it occurred, but then also making sure that we have the right ability to see that they also are taking their role in that overall governance process in the right way. Kelly Hogan: [18:06] Excellent. Well, from the conversation so far, it's clear that some organizations are excitedly already moving ahead with AI, while others are still taking more of a wait and see approach. But the window to catch up may be narrowing and with the industry expected to increasingly separate into AI-enabled leaders and followers over the next two to three years. [18:20] For those who haven't made the leap yet, let's talk about what may help accelerate AI integration and close that gap. So Nik, staying with you, take us back to Worldwide Clinical Trials inflection point. What was happening at Worldwide Clinical Trials that made you realize the traditional ways of running trials wouldn't get you where you needed to go? Nik Morton: [18:50] Yeah. Thanks, Kelly. So I already talked a little bit about our journey with AI and the realization of the opportunity there, but there was a second trajectory that was impactful as well. For some time as an industry, we've been looking for ways to reduce the complexity of trials and to decrease the volume of data we were collecting. However, the reality has been that clinical trials have continued to become more complex and data volume has continued to increase, whilst the expectations around quality and transparency and speeds continue to be there and in some instances rise. [19:34] So at the same time that we're thinking about AI, there's almost this conversation about, "Well, can we give up on our aims to reduce complexity? Do we just embrace the fact that we can deal with data volume now?" And I think for us, we though that the two have to be considered together. So the question really becomes, in a world of more complex trials and more data, how do we enable highly skilled professionals to operate at a completely different level of effectiveness? [20:12] And then AI becomes part of the solution, not the solution. So that became a framing at Worldwide, and that was helpful because it moved us beyond the thinking that AI could primarily be a cost reduction tool and started to make us think a little bit more strongly about, how can it be a capability multiplier in an increasing complex environment? So if we break that down, you can think about the traditional roles. We can have a debate about whether we should be thinking beyond the traditional roles, but if you think about the traditional roles, how do we help a data manager review more data more effectively, more quickly? How do we help a clinical operations team review data and identify risks earlier? How do we help study teams make better decisions faster? How do we move the study timelines forward faster? And that's probably about removing a lot of administrative friction that allows us to move at pace. [21:10] So those questions ultimately lead to more broader transformational discussion because then you start to realize it's not just about picking an AI tool, the data matters. What access data do we have? The operating model matters, the technology platform matters, and that's what really pushed us beyond isolated pilots and towards a more strategic approach. So that was really our inflection point then. Tom Doyle: [21:39] You know what occurred to me or occurs to me, Nik, is the approach that you're talking about leans into vision and being an AI forward CRO, which you talk a lot about, but also the practicality of making sure that you're making those investments and we're making those investments, not to build the shiny objects that Nisarg was referring to in the beginning that sometimes can be quite distracting, but they're having real practical impact. They are driving a better experience for everyone on the trial team. They are helping us really run faster and better clinical trials. [22:10] And as a CRO, you have really tangible ways to measure that. We've appreciated your partnership in helping us and challenge us to make sure that we are not just focused on the technology impact, but also how does that commercialize? How do you adopt it? A lot of that is behind Medidata Plus, for example, as a way for our industry to rethink how we commercialize, we enable the use of new AI-based technologies. You see a lot of that, and Nisarg I'm sure can share a lot more industry insight about that. You see this shift going from ... [22:47] Early on, a lot was given away for free to create that excitement, and now we're seeing a lot of shift to consumption-based pricing. And what I'm hearing a lot is leaders like Nik saying, "This is putting me in a tough position where I want to move faster, I want to be more aggressive, but I also have to manage a token budget." With new models like Medidata Plus, we try to avoid that. We get really focused on, what is the business outcome we're driving, what are the business artifacts that we're creating together and we're creating better and faster? And we're not just measuring, what is the unit of work the CPU, for example, is doing? This is going to be a big unlock in us equating investment to value as well in what is a growing more complex space. Nisarg Shah: [23:39] And I agree. I think, Nik, when you mentioned AI may create a few new roles in the organization, what I'm also hearing is it is also empowering and elevating existing roles in the organization to do more. So faster clinical trials, more with less, and so on and so forth, but allowing the ecosystem and the community to do .... Or reach the North Star, which is impactful therapies for patients who need it the most. Kelly Hogan: [24:12] Nisarg, the report shows the real gains come from scaling AI across multiple trial workflows. What should organizations look for in a vendor to know they can actually support that journey and not just a single use case? Nisarg Shah: [24:28] Sure. I was at a conference a few days ago and almost every other provider claims to showcase a use case, be it an agentic workflow or an AI-enabled workflow. I think the key takeaway is it is very easy to create point solutions or AI use cases today because I think foundational models have become so great at doing that. But I think what differentiates a good provider from the rest is, are they just showcasing a mix of ... Or throwing a few use cases on a slide or are they actually having the domain expertise to also deploy it in the client's environment for them? [25:10] The second is also the depth of partnerships, which the provider brings to the table. To really scale AI, it is difficult to go at it alone. You need a lot of partners, be it hyperscaler partners, data partners, and so on and so forth, to really deploy AI and see the gains of it in a few months' time. Leading providers also look at not just depth of the use case, but also the depth of ecosystem. And finally, it's also the interoperability conversation, which both Nik and Tom have spoken about. So how easily can the provider integrate AI into the core workflows, into the core system of record at hand as well, so that it's not just about complex integrations and customizations? It's about a smooth interface and easy to use use case, which can really bring out the best of both worlds. [26:16] So in my opinion, leading vendors try to do three things differently. One is have a good or great partnership ecosystem. Two is think about an interoperability as a built in conversation. And three, think about how to make it real for the client by showcasing not just a technological mindset, but also a domain experience combined or married with the technology expertise they bring to the table. Kelly Hogan: [26:53] Nik, anything to add there? Has that been your journey as well? Nik Morton: [26:59] Yeah, I think ... Tom mentioned already the partnership that we have with Medidata. As we think about the journey and the comments Nisarg made, it's probably ... A part of the journey that many are going down and I see people taking different tracks, as to what partnerships that they're developing. I thought I'd make a few comments on that because we looked at virtually every option you could imagine, and I think most organizations have done this. [27:30] We've evaluated the really established specialist frontier AI companies for partnerships. We looked at emerging startups, we explored building our internal AI development team. We looked at things on a point solution for a specific workflow. We looked at overall personal productivity solutions, and then we also looked at the broader platform approach. And what became increasingly clear was that the most ... Most of those discussions came back to not who's got the capability to build a good AI solution, but how is that going to be best informed by the best data set? [28:18] And when I talk about the best data, we're talking about quality and consistency, we're talking about accessibility and volume, and we're talking about that being a journey that continues to feed us with more data because that's where we think the winning AI solutions are going to come from. That was one of the reasons that led us to our excitement about a partnership with Medidata, because of that data access that they've been collecting for decades now. But we also were thinking about the pilots we were going through and we were thinking about the situation, where you want to try and connect all these new ways of working in a new operating model with access to that high quality data and those solutions and the interaction between them. [29:16] When you think about those workflows and those use cases, that's ultimately where the platform philosophy also became pretty compelling for us. Not because AI is about platform, but because we want to move beyond pilots and really be talking about scaling AI. That's where the platform comes in and the more we discussed the various options, the importance of data access and a platform capability came to the fore, and that's where we've really enjoyed the collaboration with Medidata. Tom Doyle: [29:52] I think that's right, Nik. A lot of our discussions have been, of course, technology focused, that's generally why people talk to me to begin with. But more than that, we've been doing a lot more discussions around process reinvention, as Nisarg was talking about, in a world where we're leaning heavier into what's possible and delivered by this technology. How do we reimagine our process to be more AI forward, but also in the supporting processes around quality and compliance. [30:25] You and I have spoken a lot about the evolution that we take now in a more probabilistic world versus the sort of historical context of deterministic computer systems validation, as an example, and getting quality and compliance teams around that is incredibly important. Being able to not just evaluate technology and see for what it can be, but recognizing the changes that the organization undertakes in order to fully get the value out of it is incredibly important. [30:55] Doing that tool by tool is very costly and will never yield the larger outcome that partners like Worldwide are after. And I think that's why they're rightfully pursuing a platform approach, but recognizing that it's important that that platform plugs into a bigger ecosystem because no one person is ever going to be able to bring ... No one provider will bring all of the technology to bear for something as sophisticated as clinical research or life sciences and healthcare more broadly. Nik Morton: [31:29] And that's where you want to go beyond partnering to strategically partnering. What is on the roadmap from your partner and where is there gaps that your business needs to resolve for in a shorter period of time? And then how do you collaborate with innovators in that space? But how do you think about also the future where that can be enabled as part of the platform play as well? [31:58] It's a constant changing of the horizon scan, from this year to a couple of years out to beyond that, which is a very exciting time for us all to be thinking in that way with the pace of the innovation that we're seeing right now. Kelly Hogan: [32:17] And to that point, Nik. As part of your overall roadmap with Medidata Plus, looking ahead to things like AI assisted study builds, CRF form design, and AI-based edit checks. How do you think those capabilities will transform how Worldwide Clinical Trials runs trials and what do you envision this partnership will help achieve in the next few years? Nik Morton: [32:41] Yeah, thanks Kelly. There's a level of excitement about automating and enabling those individual tasks, but what's more interesting to me is really about creating a more connected and intelligent way of launching new trials. As an organization, Worldwide is growing, we're launching more trials, but as our customers also benefit from AI and drug discovery, we also hope that they'll be in a position to be launching more assets into the clinic as well. [33:14] When we talk about AI assisted study build, CRF design, the automation of edit checks, there's obviously an efficiency play there and there'll be gains that we can make. But I think the bigger opportunity is improving consistency, quality, decision making about how we launch studies and taking some of the time lags out of that. Historically, many of those activities have been performed in sequence, often involving handoffs and rework and lots of manual reviews cross-functionally. [33:41] Now, the humans are still going to be in the loop and the expertise is still required because every protocol will be different. And we've already talked about the complexity levels increasing, but what AI has the potential to do is to help us connect those activities more effectively, draw down on previous studies or ongoing studies, think about and deploy standards where it makes sense. But also, the AI helping us accumulate knowledge across our workforce in a way that is hard for humans to share all their experiences and help the next team make the right decisions. [34:31] That's one of the themes that I think is a benefit from AI, that will come when we're implementing at scale. So it's not really the value of those specific ... Features and isolation is the bigger opportunity, I think. Tom Doyle: [34:54] Yeah. Nik, I agree. What gets talked a lot about in AI, especially in agentic AI, is around the automation and the acceleration. What probably needs more airtime is this flywheel effect that you're referring to. The knowledge created in one study, for example, or by one team being transferred in a much more seamless way to another team. [35:16] For a long time, what we've aimed to do is cover that using training, using SOPs and work instructions. We tried to codify that knowledge, if you will, into technology or into documents, and we've achieved some success in that, for sure. There's been incredible advancements in how drugs are brought to market, but I think we can all agree that we are simply scratching the surface of what's possible now in agents not only taking more action on your behalf, working alongside you in a semi-autonomous ... Maybe even in a more autonomous way, being governed by a human in the loop or on the loop, but also that it is creating a larger corpus of knowledge and understanding that helps make better decisions, do better trial simulation. Really set up for research of tomorrow, not just run the research of today faster. Nik Morton: [35:32] Hundred percent. Kelly Hogan: [36:12] Thank you. It's fascinating to see just how much progress has been made from say the first annual report to now, and I'd like to take a look at what comes next. Nisarg, the report makes five predictions for 2026 to 2030. Which one should this audience pay the most attention to and why is protocol design and optimization flagged as a breakout use case? Nisarg Shah: [36:44] Sure. I think of course ... We all spoke about agentic AI, that's the elephant in the room, as the next big thing. And also if you think about how enterprises are already deployed, less than 10% of enterprises have thought about it or are thinking about actively deploying it right now. So 90% of the audience is still figuring out the ways, which points to codifying the knowledge, more how to guide and so on and so forth. [37:16] So our belief is that agentic AI will start off with say the narrower workflows first, before expanding into end-to-end orchestrated workflows. The second is governance being the largest enabler for taking most of what we're doing right now towards a scaled deployment. We're also talking about simulating clinical trials in the report of course, and that's also something which is very exciting for the industry. And we do anticipate that most of the early adopters will likely see light of the day for most of the use cases as foundational models mature, how to guide ... Process expertise also matures in this space. [38:03] I think the breakout use case, which you mentioned, which is protocol design and optimization is definitely flagged as a high impact and high value use case. Almost 60 to 70% of the enterprises mentioned wanting to invest in the use case because A, it's a highly complex use case and B, if done right, it has significant impact downstream in terms of say, reducing costs, improving clinical trial timelines. And also improving patient experience along the way. Most of us know that an amendment can probably impact approximately half a million dollars, in terms of the clinical trials cost. [38:47] We believe that ... One foundation models are mature right now. There is historical data in terms of clinical trial performance. What if analysis, which could potentially impact how this use case gets deployed further down in the environment. I think what the industry has mostly struggled with is, they have the scientific rigor in place, they have the site know-how, et cetera, but how do you marry both together? And that's where AI comes into the mix and the potential for this is again, profound. [39:24] So we do believe that this is one of the most ... Or going to be one of the most important industry events if there's an ecosystem, which is willing to bring in the foundational model, plus data, plus the process into one piece. Kelly Hogan: [39:45] Fascinating. And Tom, Nisarg mentioned simulation in clinical trials. How do trends like simulation and digital twins map onto what Medidata Plus can do today and where it's headed? Tom Doyle: [40:01] Yeah. Nisarg is right, the real focus is not just how do we do things faster, but it's how do we do things better? And that's why there's so much focus on the protocol design, the optimization and the simulation of that. If you look across industries, we can simulate airplanes flying, before they're ever built. We can simulate a car before, it's ever driven. We can in many ways simulate surgery, before they're ever performed. These are exciting inventions or innovations that have helped propel entire industries, created better outcomes for consumers, created better outcomes for patients. And for sure that is where we are headed in clinical research as well is in the simulation of how a trial will run before a single patient is enrolled, before the protocol is completed, because it simply helps us be more efficient but also more focused on the overall care and outcome of the patient. [40:57] How that materializes itself into the Medidata platform and into Medidata Plus, as Kelly, you asked, is that we are committed to bringing that simulation into the platform, bringing together the foundation models, the best of the intelligence to surface that simulation together with the data that's necessary in order to power such a complex simulation scenario, whether that is data that's curated within the Medidata platform in clinical research or is augmented from data outside of the platform. From the ecosystem of knowledge we know around how patients will perform under certain therapies, how certain interventions affect the patient experience and the experience of sites. These are all things we need to bring together, in order to really optimize a protocol and to get towards that trial simulation. [41:49] And that's what Medidata's approach to AI and Medidata Plus is all about, how we continue to put that into a single platform, so that you're not just trying to bolt on tool after tool after tool to hopefully achieve the outcome, but rather every step that you take is deliberate in driving you to this sort of promising future of better simulation of trial activity, before we start really making massive investments in time and in money in activating sites, enrolling patients. [42:23] As an industry, we're just really focused on, how we can be better, how we can be faster, and ultimately what are all the meaningful steps that we can take that help us get novel treatments to patients faster? Nik Morton: [42:36] I think that's the really exciting point as we look forward the next two, three years. I think we've got a lot of reasons to be confident that we'll be able to look back in a few years and say, "Yeah, we were able to more reliably launch studies. We were able to more quickly clean the data and be in better shape to get to database lock more efficiently and with less time." And then from there, the analysis and the reporting that has been automated and AI enabled and that takes time out of the equation. [43:13] Those are tangible things that we're all working towards. But what I would hope for is this piece that Nisarg and Tom have been talking about is that we could actually use simulations and the data that we're accumulating to design better protocols that match more with the patients in the real world and we know where those patients are and we know how to engage them with the trial and we take what is still and always has been the biggest challenge of our industry, which is enrolling the patients to match the protocol. I would hope that we're moving the needle on that. That would be something for us all to be very, very excited about. Kelly Hogan: [44:03] Agreed. Such an important part of the lives of clinical trial leadership, that enrollment piece, and exciting to think about the possibilities that this technology affords. Tom, I want to emphasize one statistic here, 90% of the industry is aware of agentic AI, but barely anyone has hands-on experience, close to 10%. Where will autonomous AI show up first in clinical trials from your point of view? Tom Doyle: [44:34] We're seeing a lot of that already, and it kind of goes to what Nik was referring to before. A lot of the focus has been on the acceleration, the operation side of trials. It's an area of large spend. It's an area that is very time-consuming, and so it's a natural area to start, but also in many ways it's more tangible to see, how do you see through the other end of that? Already it's very real that we can predict the outcomes or the operations outcomes at various sites and various protocol designs. We can know with better and better certainty how certain trials will enroll. That isn't to say that we've solved all of the challenge with it, but we've definitely shown there is opportunity there and that's where a lot of investment today is going in terms of agentic AI. [45:21] The other is in the very time-consuming tasks, the drafting of documents and artifacts that are necessary for filing or for activation. That's certainly an area that is ... What many would think of as being fairly low-hanging fruit to go after. And that really starts to prepare us for what ultimately are those bigger use cases we were referring to around the clinical simulation, the outcome simulation, better enrollment, better patient engagement. [45:50] Those that feel less tangible or much harder today will become closer and closer as we get better, both in terms of building better data, building better trust, building better technology, building better processes. All of that will start to feel a little bit more real. And so what I like to think of is the work that we're doing today is impactful today, but is also building the muscle in order for us to really unlock those opportunities of tomorrow. So today, largely focus on operations. Tomorrow, we largely focus on outcomes. Kelly Hogan: [46:29] Very well said. We're going to be taking some audience questions here in just a minute, but I have one last question. Kind of going back to the top of the hour to Nisarg, for organizations watching this who are still in exploration or pilots, what should they do in the next 12 months to avoid falling on the wrong side of the leader, follower divide? Nisarg Shah: [46:53] Sure. I think the most important change would be to have all the data silos unified under one environment, have a single source of truth for truly scaling or deploying AI in an end-to-end manner and get the most out of it. The second is also to have the governance and compliance ironed out and fleshed out with respect to, where and when do humans come into the loop and what's the way to govern consumption and so on and so forth? And the third is also the due diligence on the provider or the partner ecosystem in terms of, who has the capabilities to truly work with them and scale the AI capabilities to the ... I'd say to the next big leap because everyone can claim to have an AI use case or an AI offering. It's not something which is difficult to come by nowadays, but those who have validated workflows and governed access would ideally be the ones to support them and get them to the right side of the deployment journey. Kelly Hogan: [48:10] Tom, agree, disagree, anything to add? Tom Doyle: [48:10] No, I think that's right. I mean, the reality is it doesn't take much to stand up a POC as Nisarg referred to earlier. What does take a lot is to scale that and to show that it does day over day deliver the level of performance and quality that is not just a desire of our industry, but is really an expectation. We're not in an area where close is always close enough. We need to be very confident in the decisions that we're taking, the investments that we're making, and that's the difference between standing up a solution in days and really bringing something that can offer enterprise scale, that can really help patients, help sites, help sponsors, help CROs be much more effective in this very complex space that we call clinical research. Kelly Hogan: [49:07] Well, thank you for an excellent ... Oh, sorry. Go ahead. Nik Morton: [49:12] Yeah. No, I was just going to add to that. Worldwide, we're trying to ban the word pilot for this reason, because I think even before AI comes along, most of us have been involved over previous decades of piloting something, a new technology, a new process, a new way of working. And many of them succeed because they're run by a small group of very enthusiastic people working on an exciting new topic with a lot of focus, a lot of management support, extra investment. The pilot's successful, but then we can look to and point to failed implementation. And it's the question of scale that we've been talking about through this whole discussion. [49:55] We talk now about proof of concept versus proof of value, and that proof of value is where we look to gain the insights that really give us the opportunity to be successful in implementing at scale. So that's where we're going beyond that small group of enthusiastic people. We're involving more people. We're collecting not only the success of the outcomes, but the effort involved and the level of satisfaction or frustration that it generates within the workforce because we know that's what you have to overcome to implement at scale. So you want to learn more than just the technology output, you want to learn the human experience. [50:37] I think if you're then transparent with that level of experience from that proof of value and you bring that into the business case for your implementation at scale, that can help you generate the trust within the organization, but also be ready for the real life experience that those larger number of people and a larger number of projects and processes will experience and you need them to be able to engage with that change management. Ban the word pilot. Kelly Hogan: [51:09] I really like that summarizing statement you offered. We've moved from proof of concept to proof of value, which really is kind of the through line of the conversation. So again, thank you for a great discussion so far. We're going to go ahead and transition to some audience Q&A. Thank you to our audience for already submitting some great questions. Feel free to continue to send those in as you think of them. This first question I'll send over to Tom. How do you structure oversight to satisfy regulatory standards and build trust? Tom Doyle: [51:49] It's a great question, and certainly the regulatory environment is ever evolving, and so we spend a great deal of time making sure that we are staying in front of that. That comes not only from making sure that we're understanding and partnering with regulators and thought leaders around the world, but also we're working with industry partners around, what is behind a regulation or what is an intention, which often goes beyond the language of the text. And I think that's the important thing. [52:24] Our industry has long been predicated on continuing to raise the bar on everything, including our position to say regulatory adoption or regulatory requirements. And that's an industry commitment to just continue to be better, to make sure that we're delivering safe and effective medicines to patients. That has to be continued to build into our technology. A part of that we've talked about today around building more transparency and ensuring trust in the models we use, the data that powers them, how decisions are made, how insights are produced. [53:06] All of that is incredibly important and will continue to be incredibly important into the future. But I also suspect that we will see new challenges, especially as we get more focused on the more semi-autonomous and autonomous operations, that are certainly available to us and will be part of the future, but how we're making sure we provide the right governance and oversight of those< will as much be a regulatory and process challenge as it will also be a technology challenge to bring more capability to bear on that. [53:41] Some of that is actually more agentic technology itself, that is looking for anomalies, for example, or signals in process and in data that would suggest that something is drifting or is not as you would expect it to be, but not all of it will be solvable that way. There is still more work that we need to do as an industry to make sure that we can be as fast in our thinking, as technology is proving that it can be in the innovation cycle. Kelly Hogan: [54:13] Excellent. Thank you. Now this question is for Nik. What's been the hardest part of preparing Worldwide Clinical Trials for AI and what would you do differently if you were starting over? Nik Morton: [54:26] Great question. I mean, I think the hardest challenge is change management. I think the lesson learned is that as we approached AI, we thought, "This is a new wave. This is a different type of innovation that we're dealing with and we have to think differently." And we thought very carefully about all the complexities that AI brings, the data that's required, the legitimacy of the modeling, the trust that we can put in it, what can be autonomous versus needs to have the human decision. [55:04] We focused early on the unique questions that were new with AI, but I think ... What I reflect on is ... Those were simpler to wrap your head around and get alignment around. And what remains is the challenge that we always face in the organizations, when it comes to change management, which is the value proposition, the reason why we're driving for some change needs to be broadly understood. We need to generate the trust and engage the people in a way that they can see the opportunity ahead of them. They can see the efficiency gain, the quality gain, the timeline gain. They can be excited about that and then they can support and drive the change at the scale that you want. [55:58] Those are time-honored change management challenges that we all know well. I think that's my reflection is to hold those core and recognize that yes, those are the new questions to ask ourselves because of the innovation within AI, but the old questions are still the important ones to focus on. Kelly Hogan: [56:21] Thank you. Next question. What's the one question organizations should ask to truly vet a partner's AI maturity and ability to scale? Nisarg, any thoughts on this? Nisarg Shah: [56:33] Sure. I think one question which could probably short circuit a lot of the ifs and buts could be, have you done this with another client? Can you show us ... Did they see proof of value or proof of scale? And ideally that would also lead to a know-how in terms of, what are some challenges which enterprises or CROs may commonly face, and this is potentially what you could do to overcome it. So the intent is not to replicate exactly what someone else is doing, but to at least get an understanding of, what were some of the challenges that most industry participants face, what did they do to overcome it, and what was the ROI realized in those situations, which could potentially lead to a similar outcome? If not even better if done right. Don't shy away from asking your partners, "Have you done this before and can you show me or share with me a few examples of a similar situation?" Kelly Hogan: [57:44] Thank you. Nik, anything to add? Nik Morton: [57:44] Yeah. Well, two things. One, I don't think one question is going to be sufficient to truly vet a new AI solution, so I think there's going to be a number of questions. Nisarg's definitely identified a very key topic of conversation. I think I would supplement that with questions around the data that has informed the models within the AI solution. Where is that data? What does that look like? How is that accumulating over time? How much will the solution rely on the data that I have to bring versus the data that the solution's bringing? That's an area of focus for me. Kelly Hogan: [58:32] Thank you. Tom, what's the biggest operational change that converts a successful pilot into scaled adoption across trial workflows? Tom Doyle: [58:44] In some of that, it's a lot of what we've been talking about on this webinar. I'll come back to the last one that Nisarg and Nik were just talking about. Our goal is to create experiences, is to create proof points that demonstrate not just what was possible once, but what can be possible under various different circumstances. So using some of the examples we used, protocol optimization, it's one thing to show that we have done this before. We have worked with a client to do a protocol optimization in one therapeutic area or one disease or one protocol, but why do we believe that we can do that more broadly? What is the conviction and evidence behind it that would suggest it is generalizable and scalable? I think is an important operational question. [59:34] Nik touched on that, in terms of the evergreen-ness of the knowledge, how we continue to reintegrate or train models on new data, for example, that helps make sure that they're ever current. That is certainly one operational challenge. The other is just how we approach them though, is the thinking that goes into new innovation is different than perhaps in the past of software development. It's, "What is the problem we're trying to solve? What is the value that's being created?" And then a real introspective evaluation of, what is actually feasible today? [1:00:09] Our ambitions certainly today exceed what is possible, but that isn't to say that there aren't steps along the way we can make. Coming back to the simulation example, operation simulation is very feasible today. Lots of great examples of it. We've got lots of great partners that we're doing that with. Safety is becoming a new area where more opportunity is proving itself. We've got some ... To Nisarg's point, we've got some proof points there that would suggest we can go further. And we just need to keep evaluating, how much can we keep pushing the envelope in terms of what's possible and scale versus where do we still need to do more work, collect more data, build more experience before we're ready for this very demanding industry? Kelly Hogan: [1:01:02] Thank you, Tom. And thank you to all of our speakers for a great conversation. Thank you to our audience for your active engagement. If you were not able to answer your specific question today, we will do our best to follow up after the webinar. As a reminder, a recording of this session will be available on demand. And on behalf of Fierce Pharma, we look forward to seeing you next time. Thank you so much.