Rethinking Clinical Trials with Faster AI-Driven Decision Making - with Shefali Kakar of Novartis

Welcome, everyone, to the AI in Business podcast. I'm Matthew DeMello, Editorial Director here at Emerge AI Research. Today's guest is Shefali Kacker, Global Head of PK Sciences and Oncology at Novartis. Shefali joins us on today's show to discuss how data-driven tools are changing investment decisions in early-stage drug development from the molecular level to full-scale portfolio strategy. Together, we explore how in silico discovery, advanced modeling, and failure data are creating a smarter, more nuanced way to evaluate risk and predict success long before a drug reaches clinical trials. Shefali also explains the organizational changes needed to better harness these systems and how AI is reshaping capital allocation across life sciences. Just a quick note for our audience that the views expressed by Shefali on today's program do not reflect that of Novartis or its leadership. But first, interested in putting your AI product in front of household names in the Fortune 500? Connect directly with enterprise leaders at market-leading companies. Emerge can position your brand where enterprise decision makers turn for insight, research, and guidance. Visit Emerge.com slash sponsor for more information. Again, that's Emerge.com slash S-P-O-N-S-O-R. the leaders who stand out today are those who transform data into better business outcomes NYU Stern's one-year part-time MS in business analytics and AI prepares professionals to do exactly that this segment is sponsored by NYU Stern's MS in business analytics and AI visit their website to start your application today Without further ado, here's our conversation with Shefali. Shefali, welcome back to the program. It's great having you. Thank you, Matthew, for having me here again. Absolutely. Last time we spoke a lot about the clinical trials process, a very instrumental, to make the understatement of the year, part of the drug development space where we're seeing all kinds of AI being deployed. Right now, we're going right to the beginning of the process in terms of how life sciences organizations can make the best investments possible and decrease the most risk in terms of where they're investing in these drugs. And just as I was saying towards the end of the last episode where we had you on, we're seeing efficiencies that create all kinds of trade-offs at the beginning of the process, at the end of the process, different kinds of technology being deployed. I was making the example towards the end of the last program about how we're seeing such step-level changes in protein engineering. We're able to make so many more specific drugs for so many more rarer conditions than we did before, and that needs to be a consideration on the other side of the drug development process and clinical trials. We need to have stronger patient segmentation. You were telling us in response to that, that that's been controversial in the past. One of the many ways we're seeing data change these processes is that's no longer the case anymore. But just in terms of really right from the beginning, where are life sciences folks going to put their chips on the table just in terms of what drugs based on this new deluge of data from all parts of the process? to get a greater clarity on what's really going to make it to market 10 to 15 years later, which is still the numbers that we're citing. Everybody that comes back on the show for all of the promise of these technologies to bring down that cost from the proverbial $1.5 billion to $2.5 billion price tag to develop a new drug in the 10 to 15 years it takes to market. Those costs and that time spent is still very much in place. But how we at least make those bets on what is going to get through these processes is a lot different now. How are smarter, more multifactorial models reshaping early drug development, if we can kind of take that much larger view from the beginning of the process? I think if you remove the word AI and really think of like really what are the different ways we are utilizing the mass data, right? And there's so many different ways to look at it. Really as early as when we start to think about the chemicals themselves, every carbon to hydrogen to nitrogen change can actually be modeled. How is it going to translate into your safety? It's going to translate into your pharmacokinetics. How is it going to change maybe your activity? And these relationships are now no longer being looked at just from one program level but across industry program level There are companies that have emerged across that actually just focus on doing this kind of analysis for you. And small biotechs could utilize this and large pharma companies could utilize it. Some of them also do the analysis and then make the right molecule for you. So I think it's a whole new world that has emerged on before even trying to come up with the molecule, almost doing this in silico drug discovery in finding this. The same is happening on the biologics run where we're really dealing with protein chemistry and saying, you know, certain parts of the protein can actually be, again, assumed that this is going to really help us with certain safety parameters or really making sure that this drug lasts in the body for almost a month because when we are trying to go to the market with something that needs to be injected to you once a month, it has a higher probability of success on the market than something that you have to inject every single day. So we're always trying to look for ways to really ensure what are the different aspects of the molecule that you can design in and then be able to say, I want to, it's almost like creating your own little, you know, your drug. I want it to last longer. I want it to be a little bit stronger on working on this particular protein versus not hitting this three other proteins that cause a problem. You're also a commitment to what's on the market already and then trying to be better than what's already on the market. Right. So we have this much clearer view about what is a safe bet. And we've had so many financial leaders come on the show, especially from the wealth management space, especially from the capital market space. tell us all about how we're seeing a lot of these similar capabilities, painting with a broad brush there. And much of the reason we have you on the show today is to bring some nuance to that. But much of the same capabilities be able to tell us if we put this amount of money here for a capital market based on this market analysis, then we know it might turn into this kind of investment. Obviously, the life sciences space is much, much different. Even where we didn't get a chance to talk about this in the last show, but even for where we're seeing the potential of digital twins, especially in our broader understanding of how human systems work, the potential that there could be that we could really bring down animal testing, human testing of any kind to have some certainty about even where controversial deployments of drugs could affect the body. What does that mean for where we're putting the capital? How can AI help improve capital allocation in portfolio decision-making? I think a lot of these things are already happening where there are some things called even in silico clinical trials. And I've only read about this more so. I haven't really seen if decision-making is really taking place based on this. But I would imagine if I was an investor and I wasn't really in it, I could actually deploy such, at least making an informed decision. You could actually do an in silico clinical trial where you would actually know the probability of success. And this is something, the concept of probability of success is very much part of the equation when you're doing these financial modeling. And the probability of success is based on sort of similar programs, similar drugs. What is actually the market size? What is really the market potential? And really looking at some of the past data and trying to integrate that together. But I would imagine this probability of success for a clinical trial is going to become more and more refined as time goes along. Right now, if you go deeper into it, it still seems, it almost seems like a, well, you can throw a lot of holes in it most of the time. Right, right. I would like to believe that the concept is actually quite powerful. And if you really put in all this, as we went back to your previous segment, when you talked about this cross-functional thought process, I don't think today this probability of success has as much of this cross-functional integration into the questioning of what would make this a high probability of success. But if you were to actually make it a little bit more refined and think about all of the different angles that would make a drug succeed or not succeed and then put that in the modeling. Even if it's a rough guesstimate, it probably will continue to get better and better with time. I would not be surprised if in a decade from now, we ask ourselves, do we really even do this phase three clinical trial and just use the probability of success modeling? And if it is above a certain number, we say go. And if it's not, maybe we take less of a risk. And it would really depend on the appetite for that risk taking in those scenarios. And with more certainty that, as you're saying, it really changes the conversation around risk. And it wouldn be hard to imagine a world where there a sort of attitude of like we have all this data now Why would we even what stomach would we even have for risk It's hard to imagine what that part of the conversation is going to sound like. In your first answer, you had mentioned, hey, if we stop talking about this like AI and talk about it like it's a process, an elevated process with technology, you're just having the same – double the amount of people do it faster. If we more look at it in that way, then these are the results. And I really appreciate drawing that contrast, especially for what you were saying there that's so entrenched in data science. You know, we'll see these systems get better and better every year. That's Moore's law. That's, you know, where we're seeing the pace of technology outpace so many, the ability of so many organizations and industries to really keep pace. And that's not necessarily a terrible thing, especially in regulated spaces like life sciences. But just in terms of where we're seeing structural changes, where we're seeing the technology change, the step level changes that we know are coming down the pike, even if we don't have a full view of what that end result is going to look like once the dust settles and we see widespread adoption. the proverbial one to always note at this point of 2025 is we see agentic coming down the horizon. We know this will mean structural changes for organizations. But even just for the state of play, for the reliable technologies that we're seeing in these spaces, especially at the beginning of drug investment, drug targeting, You know, what structural changes are you seeing necessary for life sciences organizations to really engender through the organization and are needed to best support, you know, data-driven decisions across R&D teams? I think the biggest gap today is the access to data and maybe just the data format itself, right? So they often, whenever we have a question, the first point always is, do we have the data? Okay, we have the data. Now, can we actually access the data? There are all of these creations of data lakes that's going on right now where we're trying to just collect the data. It doesn't matter what the format is. Let's first just collect the data and then we'll start to format the data in a way that it actually is harnessable. To me, that's the part that is probably going to be the next few years. Most companies probably are investing their energy and time in ensuring that that they're able to bring all of their data together in a lake or whichever format they want to put it together in so that they can actually harness the final value of the data. I also think that when there are all of these acquisitions of companies or mergers of companies, I think sometimes that's the one value that is often lost is the data itself. So it may be a drug that didn't make it or it might be information that took place, but it doesn't have any meaning to any particular molecule moving forward. But there is so much data associated with all of that information. And most of you know, you learn the most from your mistakes and you learn the most from the failures. And sometimes I think that's probably the lost data set that we have, unfortunately, not really taken a lot from. So I would like to see a lot more of sort of our mistakes and failures really informing the future of how not to fail again, how not to make the same mistakes again, rather than only taking the success factors. Always, always the one of the philosophical problems with regulated industries, which is this very complicated relationship with failure that you don't see in a little bit more tech adjacent industries or the closer you get to Silicon Valley where it's like, you know, fail fast and break stuff, which you obviously cannot do in lives or you can't really take on that that that philosophy wholesale when lives are on the line, of course. But maybe having a different attitude about the failures that we've already seen and what lessons we can learn from there, of course. I want to go back to what we were saying a moment ago just about risk, that conversation changing in ways that we might not be able to tell. But even in your last answer for knowing that we have this gap in data access and, oh, man, we could do a whole episode just on your last answer alone in terms of user experience, in terms of how best to close that gap in a very incremental way. But let's maybe fast forward into the future and try to see what that world looks like, where that gap is closed. How does that change maybe the discussion around clinical risk? What does that look like going into the next decade as we start to see our Indeed teams have more ready access to data? I think I would actually go back to your point about the fail fast cannot happen in pharma I think if he had all of his data I think fail fast can happen and should happen because you want to expose the least number of patients to something that doesn work You want to expose the least number of patients to something that may cause harm So in essence, you actually, if there is a failure of a drug, you do want it to fail fast. And sometimes having access to this data and understanding what are the things that are clearly going to be a no-go is actually a very important thing. So I wouldn't actually say that it's not the desire for the pharma company. Of course, we don't want patients to have not a good experience or have adverse events, but I would much rather very few than a lot, right? So you don't want something to go all the way and then you find out the drug never works. But if we can actually utilize all of the data to know, well, this drug would never work, I would much rather know that in phase one than in phase three, because in phase one, you probably have 20 patients. In phase two, you have, you know, 50 to 100. But in phase three, you have thousands of patients. And I would rather the 20 than a thousand. Right, right. Absolutely. And we'll see how this this conversations or many of the different conversations we're talking about on today's show evolve into the future, especially because I've only think we've in so many ways, as you've mentioned in your answers over the last two episodes, we've for the most part, we've only seen technology and these capabilities only really scratch the surface of where we can go. And when we start to see it go deeper, we'll have to have you back on. Shefali, thank you so much for being with us these episodes. I think it's been incredibly insightful for the audience. Thank you so much for having me. This has been a pleasure. wrapping up today's episode i think in both episodes we've had shefali on for this year she had a lot to say about the ways that data can inform drug discovery processes from the investment funnel all the way down to clinical trials here are three big takeaways from today's episode we'd like to summarize especially for life sciences leaders in our audience first multi-factor modeling is redefining how early drug decisions are made by simulating chemical interactions and patient outcomes before clinical trials begin. Next, AI is transforming capital allocation in life sciences, helping companies gauge success probabilities and reduce risk exposure across large R&D portfolios. Finally, structural changes are critical, especially in improving access to siloed or unstructured data and learning from failure, not just success. Are you driving AI transformation at your organization, or maybe you're guiding critical decisions on AI investments, strategy, or deployment? If so, the AI in Business podcast wants to hear from you. Each year, Emerge AI Research features hundreds of executive thought leaders, everyone from the CIO of Goldman Sachs to the head of AI at Raytheon and AI pioneers like Yoshua Bengio. With nearly a million annual listeners, AI in business is the go-to destination for enterprise leaders navigating real-world AI adoption. You don't need to be an engineer or a technical expert to be on the show. If you're involved in AI implementation, decision-making, or strategy within your company, this is your opportunity to share your insights with a global audience of your peers. If you believe you can help other leaders move the needle on AI ROI, visit Emerge.com and fill out our Thought Leader submission form. That's Emerge.com and click on Be an Expert. You can also click the link in the description of today's show on your preferred podcast platform. That's Emerge.com slash ExpertOne. Again, that's Emerge.com slash ExpertOne. We look forward to featuring your story. If you enjoyed or benefited from the insights of today's episode, consider leaving us a review on Apple Podcasts and let us know what you learned, found helpful, or just liked most about the show. Also, don't forget to follow us on X, formerly known as Twitter, at Emerge, and that's spelled, again, E-M-E-R-J, as well as our LinkedIn page. I'm your host, at least for today, Matthew DeMello, Editorial Director here at Emerge AI Research. On behalf of Daniel Fagella, our CEO and head of research, as well as the rest of the team here at Emerge, thanks so much for joining us today. And we'll catch you next time on the AI in Business podcast. I'll see you next time.