Your Trial Plan Doesn’t Fail in Execution. It Fails in the Handoff.

Hello, and thank you for attending today's webinar, How AI Is Accelerating Clinical Decision Making from Protocol Design Through Study Startup, presented by Medidata and Fierce Pharma. My name is John, and I will be your host. Before we start, just a few housekeeping notes for everyone. To learn more about our speakers today, just click beneath their names to read a detailed bio. Available downloads are in the Resources window on your screen. If you accidentally close any of these windows, you can reopen them by clicking on the icons on the left side. This webinar is being recorded. It will be available on demand. And finally, please submit your questions as you think of them via the Submit Questions window. It's now my pleasure to introduce our speakers today. Joining us are Dan Braga, General Manager of the Study Experience, and Ken Hamill, VP, Trial Management Solutions. All right, let's get started. Dan, over to you.

Thank you very much for the introduction, and thank you again to everybody for joining today. We're really excited to talk about these topics. It's something we as an organization are very excited about, and we know the industry as a whole is very excited about. So, thank you all for making the time today to go through this information with us. Before we begin, I thought it'd be a really good idea just to ask you a few questions, just to kind of level set on where you are on your journey with AI in the clinical research industry.

So you should see a poll question popped up on your screen now. Really simply, how are you currently using AI in clinical trials? I would love to learn more. If you're not using AI at all, this is something that you have zero usage of across your organization. Maybe you're at that stage where you're kind of investigating: should we use this? Should we not? You know, where is it best used, where's it not? Does our organization support this? Do we have the team members who are able to implement this? Would love to also see, maybe you're further along in the process. You're evaluating solutions, the use cases that you want to focus on. And then lastly, hey, we're all in on this. We've got tons of initiatives going forward. So anyway, just take a few moments to step through that. Would love to see some of the results as they come in. So we'll pause here just for a few seconds and let you vote. And again, we're just trying to better understand where you are on your current AI journey, all the way from nothing to "we're all in, actively using it across the organization."

So, seeing a lot of the results pulling in here, it looks to be fairly even and kind of spread across the board. We'll push this out to you so you can all see the results before we move on in the presentation as well. All right. It looks like the results have kind of stopped coming in. Everybody should see that the results on the screen look pretty even. Some people are not currently using AI at all, and so we'd love to kind of shed some light on AI in clinical research today during this webinar. The vast majority of you, or almost 50% of you, are just investigating whether to use it. This is the perfect discussion, the perfect webinar for you. Some are further along in the process, actually looking at individual use cases and evaluating potential vendors. And then almost one in five of you are at that stage where, hey, we're actively using AI in clinical trials. Maybe not for everything, maybe not for every use case, but there are use cases that you've found within your organization where you are actively using AI. So that's great. I think there are going to be nuggets in today's webinar for all of you, depending on where you are in your journey. So it's great to get that feedback. Thank you for sharing.

I wanted to continue on and first talk about the promise of AI in clinical trials. We're in a world today where you hear about AI for everything, every part of your life. There's AI this, AI that. It seems like some of it is really cool, and some of it, they're stamping the AI label on things that maybe don't add a ton of value to you, whether that's in your personal life or your professional life. But AI seems to be stamped on everything. You hear a lot of terms in the industry today: agents, generative AI, and all these different things. So what we want to do is just kind of step through the challenges that we're seeing and what we're learning from the industry, what we're seeing, and what we're doing as an organization to solve those challenges. So first off, increasing trial complexity. We all know trials today are much more complex than they were in the past.

They'parser to design protocols for, they're harder to find patients for, and they're harder to find sites that have the patients for the trials. There's everything in between. Trials are much more complex. So how can AI be used to perhaps simulate some of those outcomes before the trial ever begins? It's a concept that we're really big on: simulate first, execute second. I mean, you wouldn't dream of executing a clinical trial without first simulating the clinical trial. So, kind of simulate the outcome before you ever begin. And you're going to learn a lot of things during simulations. We'll talk about that later on.

Another challenge that we see is so much manual data entry and manual build time and manual everything. To get things started in a trial, there's a lot of manual, error-prone build time. Again, AI: why don't we take a protocol, digitize the protocol if it wasn't structured already, or take a digital protocol, and move towards automated study builds? Take that protocol and have AI automatically produce the blueprint for your trial and configure it for you. We hear a lot about inefficient workflows. So many siloed workflows across the organization, between teams, across teams, between organizations, across organizations and sites and CROs and so on. So there's a lot of disconnection that we see, whereas with AI, there's a lot of opportunity to really orchestrate these workflows in a whole new way: connecting workflows, surfacing insights at the right place and the right time within a workflow, keeping that human in the loop, and just making the individuals who are performing these tasks much more efficient and arming them with more information to make better decisions. So there's lots that AI can do there, which we'll touch on. The last two are kind of the coolest things. First, growing data volumes. The amount of data that's being collected during clinical trials is growing every day. What was collected 25 years ago in CRFs and these EDC platforms is much smaller than what's collected today, especially with the introduction of sensors and other patient-facing technologies. We're just bringing on a lot more data.

So how can AI be used to standardize and integrate this data in a whole new way, but also clean the data and get it organized in the right way so that insights can be produced and surfaced at the right place and the right time, as I mentioned before with the workflow one? And then the final one, and I briefly touched on this, is data cleanup. It's finding a needle in a haystack, which is tough for humans. So how can AI be used to not replace them, but just make humans better? Detecting things, finding discrepancies, reconciling data across different solutions, platforms, data sources, and so on. So those are some of the challenges that we're hearing from the industry and what we're really focused on, to ensure that we build AI solutions to meet these challenges.

So with that being said, AI is still in its infancy. As we saw in the poll question earlier, all of you attending this webinar today are at different stages of exploring things and so on. But we hear a lot from organizations around pilots. Everybody wants to test drive AI in some way. How do you actually get started? Seventy-five percent of organizations haven't really moved beyond pilots yet. They have taken AI and tried to solve a unique problem with it, and are kind of stuck in that pilot mode. We can talk some more about why we think that happens and things that we can do to help move that along. Insights are trapped in silos. We hear a lot about that, too, from organizations. We talk to other sponsors, CROs, analysts, and everybody in between, and surfacing an insight is one thing. AI is really good at that surfacing. Surfacing the insight within a workflow that I'm already working in is the real nugget, the real win. So a lot of these insights are kind of trapped in these silos. And then the final thing, which is what that ultimately leads to: the vast majority, almost three quarters of you, aren't seeing the value of AI. What you need to spend, in terms of money as well as just your time and energy and change management processes and all that, isn't yielding the results that you're expecting from your investment. So we'll talk some more about that as well. That's really the big thing. AI needs to be there not just to be a buzzword, not just to be some fancy thing that you're implementing, but it really needs to be there for impact. You want AI for impact. You want AI to deliver the impact and the value to your organization that you're looking for. Otherwise, what's the point of implementing it in the first place?

So let's talk about how we transform it into measurable AI. First off, speaking of AI for impact, we want to tie it to high-value outcomes. Where can we add the most value? Where can we benefit the clinical trial process in the most meaningful way? Those are the things that we're very focused on. Another concept that you see across what we're trying to do with AI is embedding it within workflows. We're a very experience-focused company. It's less about the individual solution, less about AI as a product, and more about taking the insight, the workflow, the orchestration, whatever it is, and embedding it within the tools that you're already using today. Having to leave and go somewhere to use, quote unquote, AI and come back and continue on in your process and workflow is kind of a thing of the past. We want to embed it within the workflow. Next, I would say, design for scale. There's a lot of one-off AI solutions. We are building everything that we do on our platform. First off, we have the data to support building AI. AI without good data is nothing. An algorithm doesn't exist unless the data is there to support building it. So we have the scale in terms of data to support building and designing AI at scale. And then secondly, we have the scale in terms of the number of users that are using our execution tools on our platform. So we can take a great AI idea and plug it into our execution tools and experiences on our platform and really bring those to life in a whole new way and at a whole different level of scale.

The last point here in terms of transforming AI is that we're really building it to support the human. We're not here to replace individuals. We're here to support them and make them faster, make them more efficient, and help them focus on the right details. So that whole human-in-the-loop concept is something that we believe in wholeheartedly. And listen, AI is going to do some things, if you want it to, that move something from point A to point B without a human in the loop. But at some point we always want humans involved and verifying. Maybe it's not every transaction, maybe it's some subset of that. It's up to you to decide how you want to tweak and configure that. But we want to make the individuals who are working today to run these clinical trials faster, better, stronger, and everything in between. So we're building AI to actually support their workflows. I think all this kind of comes together to really show how we're thinking about it, to ultimately deliver what I talked about earlier: 75% of you aren't seeing the ROI on that. So we want to build and design AI in a whole new way. Don't just take my word for it. I think the industry is starting to see more and more about what we're doing and how we're doing it. We're getting recognized by a lot of organizations, analysts, and the industry as a whole as being a leader in what we're doing. And people are starting to recognize why that is.

I mentioned before, we have the trials to support building the algorithms, and we have the usage on our platform to really support how AI gets plugged in at the right place at the right time. And I remember years back hearing from different people in the industry, "Hey, Dan, you guys have great AI, so show me how you're plugging it in. Show me how you're making it part of a sponsor's, a CRO's, a site's, a patient's, whoever's workflow, how you're making it part of the tools that they use every single day." So I'm just really excited to see that we're getting that kind of recognition, and not from a pat-yourself-on-the-back perspective, but more from the perspective that the industry is adopting what we're doing. We're doing it in a way that's really allowing people to adopt and run with the technology in the right way. I mentioned experiences earlier, so I know today we're going to really focus on the study experience. And in just a moment, I'll hand things over to my colleague Ken Hamill, who kind of leads those efforts within our R&D organization. But we're a very experience-driven organization. Everything we do is divided into three experiences on our platform. The study experience, which we'll talk about today, of which Ken and I are both a part. That's really a world where we're bringing the execution of a trial, think of it as the CTMS and the eTMF of the past, together with how you plan and forecast and design trials. Again, we believe wholeheartedly, and we hear loud and clear from the industry, that you would never run a trial without simulating it first. Why would you do that? We have the capabilities so that before you even begin your trial, you can learn from all the trials of the past. You can learn from different protocol design mistakes that were made in the past, and so on.

Our study experience brings protocol optimization, study feasibility, and a lot of our AI solutions together with the CTMS and the eTMF of the world. Then we have our patient experience: all of our technologies, obviously, that touch a patient, that get in their hands. eCOA, one of the flagship capabilities of that, our registries capabilities, our eConsent technologies. Then finally, the data experience. Think of EDC as the flagship capability within the data experience, and everything about data collection. We're doing a lot with connecting to electronic medical records to pull data in. Obviously, we do a lot with sensors and wearables and other things, just ensuring that we can get data into our platform in the most meaningful way, get it structured, and get it ready to power all the things that we're going to talk about for the rest of this discussion. So, again, experiences. I want to leave you all with an understanding that we're very focused on the experiences. With that being said, I'm going to turn things over to Ken Hamill to really drill down into the study experience and where we're seeing AI being used in the most meaningful and powerful ways. Something we're really excited about. So with that being said, Ken, I'll turn things over to you.

Yeah, thanks, Dan. And hi, everyone. Good morning and good afternoon. I want to take a quick step back for just a second and talk about the why: what's happening in the industry that's driving this shift towards AI adoption that Dan just teed up for all of us. Clinical operations teams in particular are facing really intensifying business headwinds. Those headwinds are raising the bar and requiring new levels of efficiency throughout the trial, and certainly a far more sophisticated approach to trial complexity. And that complexity makes it much more difficult to operate in a technology landscape that's governed by multiple spreadsheets and point solutions that are minimally integrated at best. At Medidata, that's really the ethos behind, as Dan said, moving towards experiences and addressing these challenges by bringing together all of our applications into the three experiences that Dan mentioned.

And that brings us directly into what is likely the reality for many of you on the webinar today: running a trial under these new and more complex conditions. That's really the focus of the study experience, which Dan and I are involved in day to day in delivering these types of solutions. Historically, the different phases of the trial life cycle, from design to planning, startup, and then on into execution and closeout, have operated largely in silos. The protocol decisions might happen in one vacuum with one functional group. Country feasibility and site selection may happen in another. The overall budgeting and financial modeling might happen somewhere else entirely, in another set of tools or spreadsheets. And then beyond that, the teams that actually have to execute on those plans, the startup leads, the monitoring teams, the processing teams, inherit those plans and are expected to hit those targets and milestones based on assumptions that were created long before they took part in the trial and that they may not have even had a part in shaping.

That disconnect is where we feel a large amount of the operational friction happens in the lifecycle. Even the most advanced CTMS or the most capable clinical operations team is really going to struggle to fix a study that was built on unrealistic baselines at the start. What that means is we have to be able to surface those predictive insights that Dan talked about, really at the critical decision points. That means not just bringing AI and predictive analytics in as a siloed product, but bringing it in at the point where the teams are making those decisions, to help them simulate outcomes and establish defensible plans up front, and then, really importantly, carry those operational plans forward into execution rather than forcing teams to recreate them in downstream systems. Right? This is not really about AI being deployed as a standalone feature, but more about how intelligence can become part of how clinical trials actually operate every day. Let's look at that in more detail. When we think about making the shift from AI as an experiment to AI as an operational foundation, the industry has to really rethink where intelligence lives and how it interacts across what is, like I said, a broad and increasingly complex spectrum of both data and technology systems.

We've all seen, and Dan showed some statistics on, huge investments in point-solution AI over the last several years. But if we're really being honest, it hasn't fundamentally moved the needle on trial lifecycle timelines yet. And why is that? I think largely because those isolated tools often require separate logins and manual data prep, and they create disconnects. If we want to talk about true operational ROI from AI, it can't be an add-on. It has to be woven into the very fabric of how clinical operations teams execute their work every day. And the core of this approach is really embedded AI. When intelligence can live natively in the systems that all the teams already use every day, it removes the friction exactly where all of the heavy lifting and day-to-day activity occurs. That starts by employing AI where the trial is first incepted, at the scientific design stage, by creating the virtual twin of the protocol, instantly turning a static document into a machine-readable format. The protocol is then really a living digital asset. Capabilities like feasibility models can then simulate viable country and site footprints based on that protocol, while simultaneously powering financial forecasting tools to generate budget scenarios that are tied directly to the most operationally realistic plan.

And it doesn't stop there. We can bring that intelligence directly to our study teams within study startup. For example, we rank and recommend which sites should be contacted based on not just whether they've enrolled patients in the past, but on a much broader set of factors, really to optimize that critical path to site activation. And then, once those sites are active, we calibrate those timelines based on what happens on the ground, how EDC enrollment is actually going. So that's another place where we can leverage embedded AI to give teams up-to-the-moment insights on the future performance and prediction of that trial based on those actuals. And then behind the scenes, while this is all happening, of course, is all of the compliance and inspection readiness, with things like document processing, where we can classify and file documents automatically at the point of capture, with active intelligence to elevate all of the decisions at each and every step. So that's really the core.

But for embedded intelligence to reach its full potential, we have to think about interoperability across what is, as I said, a highly complex ecosystem. This is where open standards and strategic partnerships become so critical. One example of this is a partnership that Medidata has developed with Anthropic, where we're leveraging open standards like Model Context Protocol, or MCP, specifically to accelerate site feasibility, with sites based on predicted enrollment goals, historical cycle times, and also active trial congestion. So MCP really serves as that connective tissue. It delivers those feasibility insights seamlessly across the systems without requiring custom-built integrations. And we can take it further and reduce friction by deploying specialized agentic workflows that operate under human-in-the-lead oversight.

One example of this, which we'll talk about in more detail, is document processing in the TMF, where agents can automate and handle document preparation, quality checks, and indexing in the background. By orchestrating all of these routine workflows, the result is that the clinical leads and the study teams spend far less time navigating across the software and more time actually advancing their studies. So if we look a little bit more closely at the root cause of a lot of the operational friction in today's trials, it very frequently traces back to how the industry handles the protocol. Historically, the protocol has been treated as a static text document. When you start with a static document, every step that follows, whether that's feasibility, budgeting, or building your EDC forms, requires a human to manually read, interpret, and translate that text into a downstream system, and then do that over and over and try to reconcile across those.

To truly accelerate clinical decision making, we have to bring intelligence to the very beginning of the life cycle, starting at early study design. We have to stop treating the protocol as a document and start treating it as a living model, what we call the virtual twin of the protocol. That means that whether you're ingesting an unstructured draft document or importing a native USDM file, AI can instantly digitize and structure that study design and those elements of the protocol. Simply digitizing the text, of course, is just the baseline. The real power comes from the insights that we can layer on top of that extracted information. Because the protocol is now in a structured and standardized digital model, you can start to benchmark things like the schedule of activities against industry data, evaluate patient burden, and simulate how specific drivers of cost and site burden are going to impact not just your enrollment timelines, but other operational metrics like dropout rates. This is what allows clinical teams to iteratively and collaboratively optimize their design before the protocol is ever locked.

The goal is really to engineer the most operationally realistic protocol while reducing the chance of costly and avoidable amendments downstream. And then once we have that optimized virtual twin of the protocol, it can act as the digital engine for the rest of the trial in multiple ways. We've put a few here on the slide. For example, it can unlock automated budget builds. Because the system already understands the schedule of activities, it can automatically map those clinical procedures directly into a cost framework, instantly generating a financial model based on the actual operational design of the study, rather than relying on disconnected estimates or having to re-key that information, which can lead to manual data entry and the chance for data errors. Second, that digital thread of the virtual twin of the protocol can drive your study build configurations. Instead of spending weeks manually building out databases, imagine translating that virtual twin of the protocol into an execution-ready system with a single click, where intelligence pulls from validated global libraries to auto-generate all of your EDC and eCOA builds, creating the forms, schedules, edit checks, and synthetic data, all automatically based entirely, like I said, on that digitized protocol.

And this is what it starts to mean, as Dan mentioned up front, to really connect planning into execution: designing an operationally realistic protocol once, and then that intelligence automatically constructing both the financial and the technical foundation and configuration of the trial. So we have this very powerful virtual twin of the protocol, which operates, like I said, as the digital engine of the trial. But turning a protocol into a structured, dynamic model generates a huge amount of complex data. There's an enormous amount of historical data, in addition to that, that we can glean from to generate the insights that we need to make the right decisions during the course of the trial. It would be a lot to expect that every clinical operations lead and every member of the finance team suddenly becomes a data scientist just to extract value from that data. We need a way to translate all of that underlying intelligence into everyday operational actions. And at Medidata, that's why we introduced the concept of DOT to our platform. DOT is really the bridge between that digital engine and all of that data and all of the humans who are doing the work on a day-to-day basis. Because running a trial involves so many different types of work, from simply interrogating data to building systems to filing documents, and the list goes on, we can't have a one-size-fits-all approach to how we expose AI to the different users who are going to make use of it. The AI has to understand the specific context of what the human is trying to achieve in that moment.

You can think of DOT as a multimodal AI layer that helps users understand, decide, and act. It dynamically adapts to the user's specific context by taking on different personas, whether that's being an orchestrator, assembling the right data and tools to configure the trial; operating as an agent, working mostly autonomously with human oversight; operating as an analyzer, where DOT translates natural language into complex database queries almost instantly; operating as a companion, providing contextual guidance on demand; and finally acting as an insights generator when a study team is trying to figure out why a specific protocol design might be a bottleneck to enrollment, or which countries will yield the best performance. Really, it's about adapting to the specific workflow rather than forcing the user to adapt to the software. This is the heart behind, like I said, that multi-persona approach to remove the friction between a study team's intent and how they're actually executing. So let me take a look at a few examples across the study lifecycle of how AI is being used today, across design, planning, and startup, and into execution use cases.

One place where this really comes to life is the way that we approach early trial planning, specifically in areas like feasibility and country and site selection. When we look at that stage of the trial lifecycle, one of the core questions that feasibility teams try to answer is: what is the right country and site mix that will give me the highest probability of hitting enrollment targets within a specific timeline? Answering that question requires synthesizing huge volumes of data: historical data, tribal knowledge from within the organization, and historical data from other trials that might live in disparate places. That often means multiple passes at the feasibility plan, looking at long and disparate spreadsheets, building scenarios manually, and then iterating between assumptions and the operational experience of the teams involved, whether that's at the sponsor or at the CRO. This is an area where AI can make that process a lot more systematic. Instead of manually constructing scenarios, teams can use AI to evaluate different country and site combinations based on user-defined constraints like enrollment targets, timeline expectations, geographic requirements, and operational constraints, and the underlying models can then use historical site performance, the velocity of enrollment, and the characteristics of that specific study and indication to generate feasible footprint options within seconds. The system can find and recommend the optimal selection of countries and sites that reach the target enrollment in the specified timelines.

But what makes this practical, and really fit for purpose and industrial for organizations operating at scale, is how we can balance those predictive models with operational reality. This really allows planning teams to become much more iterative and flexible in how they work at this conceptual stage. What I mean by that is, as planning progresses, teams can refine the site-level enrollment forecasting to stress test their assumptions before startup begins. The system calculates timelines hierarchically, cascading different parameters from the study level down to the country level and down to the site level. If the model predicts a certain activation rate, but your teams know that there is a current regulatory backlog in a specific region, the capability allows you to simply override that baseline. The engine then recalculates that delay and how it impacts your net timeline. So the intent isn't to fully automate the decision. It's to reduce the time required to arrive at the best possible scenario, so that teams can spend more time evaluating trade-offs rather than manually assembling the inputs. And then we can move on.

Now that we have a scenario and a country footprint, we can look at how that impacts the next operational hurdle in my flow, which is translating that macro, high-level plan into an executable site list. Traditionally, site selection relies on past participation or on static spreadsheets that quickly go out of date. That's a primary reason why we see that over 40% of activated sites still fail to meet their enrollment targets. And this is where we can really apply predictive intelligence directly to site selection. Rather than filtering sites based purely on historical averages, we can use advanced predictive ranking to evaluate sites across dozens of operational variables, trained on vast and standardized clinical data sets, looking at things like facility experience, investigator trial history, cycle times, and actively competing trials. The models can then take all of those into consideration to surface the top-performing locations along with the rationale, so that study teams and startup teams have clear visibility into why a site is categorized, for example, as a top performer. And then machine learning can simulate enrollment for every specific site, factoring in their operational history, FPI timing, and, again, competing trial congestion. What that allows us to do is create realistic confidence intervals for what our enrollment forecast is going to look like.

And this brings us to what I'd say is a critical operational handoff. How do we take those optimized sites and that feasibility plan and make it real? How do we bring that into the study startup process without having to recreate the wheel, without having the study startup teams operate from scratch as they begin the process of site activation? The finalized site list and all of its associated timelines and data can automatically populate the CTMS. Apologies for the technical glitch, folks. What I was saying is that the key here is that the startup teams and the downstream CTMS workflows can now be informed and influenced by all of the data and the insights that were already gleaned and learned during the feasibility process. All of the target activation dates, the country-level regulatory dependencies, and all of the site-level milestone requirements don't need to be re-entered into the CTMS. Those already carry into operational execution. That means that startup teams have visibility into true site readiness. They can monitor progress against predicted baselines because they already know how the feasibility plan was established and why it was established. What that means is that feasibility is no longer an isolated, point-in-time exercise. It becomes a living operational thread that can power startup execution from day one.

But we can take it even further and move those AI insights directly into execution. Once the trial has started to execute and sites are enrolling, the key is how we extend that same predictive intelligence into a live study environment. The moment first patient, first visit occurs, the plan starts to meet operational reality, and the challenge shifts from initial baselines to managing performance as it unfolds. Historically, when actual enrollment starts to deviate from the plan, study teams are forced into manual, reactive reforecasting exercises. We've all seen the static rolling monthly averages that don't truly account for what's happening on the ground.

To maintain control over those study timelines, enrollment forecasting has to become dynamic and continuous. So instead of relying on simplistic linear projections, we can apply advanced reforecasting engines that ingest live operational data directly from EDC as patients are being screened and enrolled. Those models can combine actual ongoing study trends with historical site-level performance metrics and actively competing trial congestion. We're able to blend real-time trial progress with underlying site performance indicators at all levels to give study teams immediate visibility into whether a study is tracking as the original plan suggested, and whether it's ahead of or behind the baseline expectations. And because those forecasts are tied directly into the operational execution layer, because they live within the CTMS, study leads can evaluate all of those intervention strategies in real time. It might mean adjusting activation timelines, it might mean reallocating regional resources, or it might mean modifying country footprint contributions. All of that can happen in the place where those interventions can take place. So reforecasting stops being a periodic, reactive fire drill and becomes a predictive sphere that keeps execution aligned with the timelines throughout the execution of the trial.

And then we can go further. As the trial moves into live execution and enrollment starts to accelerate, a lot of the administrative burden and volume of documentation starts to scale exponentially. That makes document processing within the trial master file one of the most critical and natural places for AI to take hold within the eTMF. There are few areas in clinical operations that generate as much repetitive, unstructured data as the eTMF, and manual filing burdens can directly introduce compliance risk and inspection delays. That comes down to a few fundamental friction points. First is the huge amount of time that teams spend logging in and navigating complex folder hierarchies. Second is the compliance risk that's generally created when unfiled documents sit for too long on personal drives, on desktops, on laptops. And third is the reality that quality issues within the documentation often aren't discovered until weeks or even months later, during a periodic QC. We can eliminate all of these barriers with clinical platforms that embed agentic workflows directly into the native operating environment, where we can bring intelligence right to where the work occurs while still keeping humans in the lead. Instead of treating filing as a manual administrative chore, specialized eTMF agents can operate natively inside day-to-day tools. Whether a user is working in a system like Word or within SharePoint, they can simply execute a single action, like right-clicking, and initiate a full file-to-eTMF workflow. Behind the scenes, an agent executes a sequence of automated skills, all under human-in-the-loop oversight: authenticating role-based security, running the structural ALCOA and quality checks, performing duplicate document detection, and then predicting the expected artifact type. And because the agent understands the specific artifact, it can dynamically trigger the quality checks for that specific type of documentation.

All of this allows the user to shift from manual system navigation to fully agentic orchestration in clinical operations. Teams can then eliminate the compliance gap that often plagues inspections and interim analyses, and reclaim thousands of hours of manual administrative time, because trial compliance is built in at the moment of execution. So the true test of innovation is how that intelligence can transform execution on the front lines day in and day out. A compelling and recent case for this operational shift comes from Worldwide Clinical Trials. Worldwide Clinical Trials recently became the first global CRO to fully commit to the full-lifecycle, experience-driven AI foundation that Medidata has been championing and pushing for several years now. Rather than layering more point solutions onto legacy infrastructure, Worldwide fundamentally rewired their operating model. They consolidated their global trial operations onto one single integrated platform, enabling real-time data flow across protocol design, feasibility, startup, and execution. That transformation is already creating a sharp competitive advantage for them. It's reducing cycle times, eliminating unfiled-document compliance risk, and helping them deliver the predictable, audit-ready trial execution that their sponsor partners are demanding. And the methodology that we talked through, those use cases, are not conceptual pilots or experimental exercises. Those are deployed, production-grade capabilities, driving ROI for teams who are shifting towards AI-centric delivery.

Looking at just one recent example from a sponsor who is navigating complex study design, the study team there faced a challenge that might sound familiar to folks on this call, where protocol design decisions were driven largely by subjective opinions and internal preferences. It made it difficult to justify which assessments to remove and which were the most costly aspects of the trial design. They were able to shift that conversation to a more data-driven framework, benchmarking their protocol's SOA against comparable historical as well as active industry trials within the exact same indication. Ultimately, this metric-driven framework allowed the sponsor to achieve the leanest possible protocol design, reducing site burden and overall cost while fully adhering to their scientific objectives and the regulatory requirements. So that's just one example of how data-driven insights are moving study design away from subjective consensus building and turning it into structured, scalable input for operational execution.

Yeah, great, Ken, thanks for all of those insights. Those are awesome, and I hope everyone saw that these are real things that are adding real value to our customers and to the industry as a whole. All of that we stepped through is available today, and it's not AI for the sake of AI, but AI truly to add value: AI for impact across the organization. Great examples. I know we're just zooming in on the study experience. We at Medidata have examples of AI for impact across all experiences, things that we're doing within the data experience and things that we're doing within the patient experience as well. I want to touch on a few things before we close out with some Q&A at the end. We want to leave enough time for you to ask questions. But first and foremost, coming soon is something that we call Plan and Forecast. We have historically had simulation capabilities across our platform for many years, in a lot of these areas, and now, being part of the Dassault Systèmes organization, simulation is even more prominent in what we're trying to do. We've had a lot of capabilities, but they haven't been brought together in the right experience.

So coming soon, you'll see an area on the platform, within the experience, called Plan and Forecast, where you can go through and use all of our capabilities, whether that's protocol optimization, our budget forecasting, our site feasibility, and everything in between. You'll see the trade-offs and the impact that one decision you make somewhere has somewhere else, all within one experience. So it's kind of breaking down the silos of each simulation capability. Something we're really, really excited about. It also brings everything together in a way where you can collaborate and share and talk amongst your teams and your organization about the different parts of Plan and Forecast and your overall trial design. Again, it all comes back to that spirit of simulate first, execute second. Why would you not want to simulate your trial first using some of these tools before you move on? So, again, stay tuned for more on that. It's something we're actively pushing out there, and it will continue to evolve as we move forward.

I wanted to leave you with five predictions for where we see AI evolving and advancing in the industry. The first one is that early adopters are going to get ahead of the game. It's the sit-back-and-wait, "let's see what other people are doing" mentality. We especially see this in the CRO market, in terms of the services that CROs are going to provide to the customers they serve. What we're seeing in the industry, and what we highly recommend, is that if you're not doing anything now, get started with a pilot. Get started with something and start to explore. We can help you with that. But also just explore amongst yourselves what use cases are going to add the most value to your organization. Start having those discussions. Start looking and thinking, "Man, I really wish AI could help with this," or "It would be great if it could do that." Start to do that now. You don't want to be behind the game on this, and it's definitely something where the early majority are going to win in this particular area. Secondly, the data to support AI is imperative, as the slide said. So invest now in the types of data that you're collecting, how it's assessed and how it's cleaned up, and how you're ultimately moving towards a higher state of quality, and how that data is integrated with the rest of your ecosystem.

Just start to think about data, and that could be data that's already living on the Medidata platform or data sources that you have. Interoperability is going to continue to be a huge push for us, in terms of not just the data that we have that's going to power the decisions that you need to make, but maybe data from other sources as well. So start to think about that. What data do I have on the Medidata platform? What data do I have here and there? And what other data would you need to help make the decisions that you want to make? Let's have those conversations. Ken touched on this earlier. The third one is that protocol optimization really is the foundation. We believe everything starts with that protocol and with producing that virtual twin of your protocol. And that can come in a lot of different ways. It can come from us importing a USDM protocol that you wrote or developed in some other solution, or importing an unstructured PDF version of your protocol, which we structure to help move you forward. Or, soon, what you'll start to see from us is protocol authoring, getting into authoring protocols from scratch on our platform and, throughout that authoring experience, infusing our protocol intelligence. But at the end of the day, it's about getting you that virtual twin of your protocol and then all the interesting things that you can do downstream from that, whether it's, hey, I'm just ready to build EDC, or I'm ready to configure eCOA and these other solutions. Great, we have that available today, where we can take the digital protocol and hand it off to these downstream tools. But maybe you want to go through some protocol optimization first because you uploaded an unstructured protocol that didn't originate on our platform. Maybe you want to get into budget analysis, whatever it is. It starts with that digital protocol and the things that we can do with that virtual twin to help you optimize your protocol.

Next, agentic AI moving to operational reality. Listen, there are so many different buzzwords and so many different types of AI and ways that it's being used. But we see AI as an orchestrator. As Ken touched on earlier, that orchestration layer of AI is becoming more and more beneficial to organizations like yours as we move forward. It's up to you to decide where that fits in and how it fits, how much of it is autonomous versus fully human in the loop, how it gets deployed safely, and how visibility and transparency are provided to all the users who are using it. But definitely keep your eye on AI orchestration and agentic AI, and, as the slide says, on moving towards operational reality. And finally, governance, the topic that nobody likes to talk about, excuse me, but we need to, you know what I mean? We've got to treat AI governance as a first-class citizen. It's something that we really need to invest in, and it will enable your AI's value and give you the best opportunity, as I touched on earlier, from an ROI perspective, to realize the value of the things that you're implementing. AI without the right governance has a much lower chance of delivering that ultimate ROI to your organization. So we highly suggest keeping a keen eye on the governance aspect. Again, I know it's not as fun or as exciting as exploring the use cases or implementing them or getting them out in the field, but it can bite you in the end if it's not properly assessed from the beginning.

And then finally, before we move to Q&A, just some early results. I know AI is in its infancy, you know what I mean? But what are we starting to see from this? Again, we're just scratching the surface and just getting started, but we're seeing faster startup times and fewer underperforming sites. It's an interesting thing when we talk to our users about site feasibility and some of our AI and simulation tools: it's not always about producing or selecting the best sites. But one thing that we can help you do is not select underperforming sites. We don't always tell you which sites are going to be the A-pluses, but we can tell you which ones are underperforming or not likely to lead to the recruitment results that you want. We've seen a fairly significant reduction in dropout. And then, as I was just touching on earlier, and again, I think as an industry, and for Medidata, we're just scratching the surface here as well, there's the time saved by connecting workflows with AI as an orchestration layer. On the whole agentic side of it, we're doing some really interesting things with agents inside of eTMF that you'll see coming out soon, where we're automating almost all of the document capture and acquisition. So there's lots of exciting stuff coming there as well. So anyway, I just wanted to close out with a few predictions. Stay tuned for more on Plan and Forecast, which is a really interesting bringing together of a lot of different simulation silos, and then finally some early results of what we're starting to see in the industry. So with that being said, I'm going to close things out here and turn it back over to the host to bring on Q&A and hear any questions you have.

All right. Well, thank you. Yeah, thank you. And speakers, thank you. To the audience, thank you for submitting so many great questions. Obviously we're getting a little bit low on time, but feel free to continue adding questions as you think of them. I think we have time for maybe this one question from a compliance perspective. How does validation come into play when embedding AI in your processes? For example, how do you control AI drift?

Ken, do you want to take that one?

Yeah, I can take that one. Thanks for the question. As you know, AI is not deterministic. So for us, validation isn't a one-time deployment event. We're continuously monitoring model performance against baseline metrics to catch drift. So that's one area of it. It's really continuous validation against those baselines. The other is things like mandatory human-in-the-loop oversight. If a model breaches, for example, a confidence threshold or performance metrics, that triggers, within our GxP validation process, a formal change control review and retraining before those outputs can impact trial data. So it's really part of our overall validation framework, and as it relates to AI, it's something that we're doing continuously with those models.

All right, thank you for that answer. Like I said, we're kind of tight on time right now, so once again, if we did not get to any of your specific questions, we'll do our best to follow up after the webinar. Thank you again to our speakers. Thank you to Medidata. Thank you to our audience for your active engagement. As a reminder, our recording of this session will be available on demand. Thank you for joining us today, and we do look forward to seeing you next time.