← The AI in Business Podcast
The AI in Business Podcast
Breaking Bottlenecks in Life Sciences R&D with AI Innovation - with Aziz Nazha of Incyte Pharmaceuticals
00:30:57
Loading summary
Transcript17 lines
- [00:14]
A
Welcome everyone to the Emerge AI and Business podcast. Today's guest is Aziz Nazar, global head of AI Innovations Institute at Insight Pharmaceuticals. Insight Pharmaceuticals is a biopharmaceutical company focused on research and development. Aziz joins emerges Matthew DeMillo on today's episode to discuss why life science science's organizations struggle to turn advanced AI capabilities into meaningful business outcomes. Pointing to cultural readiness, talent gaps, fragmented data environments and unrealistic expectations as the core barriers. He describes how rethinking existing processes can shorten scientific cycles, reduce avoidable delays, and enable organizations to capture measurable gains over time. Today's episode is sponsored by Deloitte. Just a quick note for our audience that the views expressed by Aziz Naza on today's program do not reflect that of Insight Pharmaceuticals or its leadership. For our solutions partners, position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap for the opportunity to showcase your solution to the executives currently funding and scaling global initiatives, Partner with Emerge. Secure your partnership@go.emerge.com partner that's go.emerj.com now the conversation with Aziz.
- [01:40]
B
Aziz, welcome to the program. It's a great pleasure having you.
- [01:43]
C
Thank you, Matt. It's a pleasure to be here.
- [01:45]
B
Absolutely. Life sciences leaders today are telling us that they sit at a difficult crossroads. They're trying to build AI systems that offer unprecedented scientific capability, yet R and D organizations struggle to translate these advances into measurable business outcomes. We had Ben Nino on the show from Deloitte a little bit earlier talk about that. We now have the scaling capabilities to develop more medicines for rare diseases than there are grains of sand on the planet. And that is a fantastic development. Definitely like what I love to repeat at the proverbial high school reunion about all of the good that AI can do. But then there comes to the question of, well, that's a lot of inputs. What are the outputs going to look like? On the business value side, teams are facing cultural resistance. There are unclear blueprints for workflow integration, major gaps between prototype success and enterprise wide adoption across discovery clinical development. We're seeing the same core friction over how to run massive scientific practical gains. In so many words. We're trying to just nail down what those incremental challenges are, especially as we turn those inputs into outputs. What challenges are you seeing as defining the current state of R and D&AI adoption in life sciences for turning that awesome capability into business value?
- [03:00]
C
Yeah, that's an excellent question. And I think it's a. It's in the mind of a lot of leaders today. And one side. You know, when you listen to the news and the hype about AI, it looks like AI is sort of redefining pharma, redesigning drugs. We're living in this utopia of the new drug discovery and the innovations and alphafold, win the Nobel Prize and all of this stuff. On the flip side to that, on the grounds, organizations trying to implement AI systems is still having difficulty capturing return on investment. And I think problem of the issue is really not the technology. The technology is great and it's going to significantly enhance what we do every day. I think the challenges are in four things. Culture, talent, infrastructure and setting up the right expectations. And before we talk about the technology and how we apply the technology, we need to take care of those four things. And if we do, then the technology can have significant impact. So we start with the culture. Most cultures, the question of the culture. Is your culture amenable for AI? Meaning do they know what the capability of or they know the knowledge and have the capabilities of those AI systems? Do they understand them very well? Is there like a fear of this AI replacing them? So what does that mean? Is that we need a lot of education for our workforce about AI, the proper use of AI, what do we mean by AI and how that's going to incorporate with them and also the type of AI systems we're building and those stuff. So bringing up the culture up and that's happened by upskilling your, your organization. And I would argue what we need to do is reskill people. And that's very hard to do. Reskilling. I think we should start with upskilling. Having people use the technology in their job first, understand the shortcomings of the technology as well as the fabulous stuff that it does. Because I think sometimes we only talk about the great things that the technology do. But also they, there are some challenges and ethical challenges and problems, especially when we apply it to industry, which is in healthcare, which is different than other industries. So, so that's number one. The second one is talent. And I would argue that we don't have, at least in pharma and healthcare, the talent that have the deep knowledge of the technology at the same time, deep knowledge of the business. And you need to have this both deep knowledge. You can bring somebody who knows AI very well but don't know drug discovery. They get to have challenges and the same thing, bring somebody who knows drug discovery and clinical development, but they don't know AI. And we need those phenotypes of talent that today is very Hard to find and it's very difficult to find. So we need to build that talent. The third thing is infrastructure. Do we have the right infrastructure? Meaning do we have our data is ready where we can train these AI models? Do we have the cloud and the compute infrastructure? Because some of these models require a lot of computation and in a way it's a smart computation and do we have that set up and what you discover? A lot of organizations sort of have the infrastructure, but it's either disjointed, the data is disorganized, is not put in in the right way and, and that makes it difficult when you try to bring the technology to sort of build the models internally. We know from our own experience that a lot of models out there that is developed in academia don't really apply to novel targets. When you try to apply some of these models in terms of protein optimizations and others to novel novel targets, nobody knows the protein structures and other stuff. They don't apply. So which means that you have to sort of fine tune some of these models or build your own models internally and that's what required to have the right infrastructure. And finally, which one of the most important things is setting up the right expectations upfront. So if the right expectation that AI chatgpt going to wake up tomorrow, have a new novel target and going to design the protein and all of the stuff all AI driven and we push a button and that goes to the FDA approval for ind, that's probably delusional. If we set up the right information. Meaning the expectation is that today our work process or drug discovery. Let's start with drug discovery is lengthy, not cost efficient. It takes about depending on the target between 3, 5 years for, for sort of known target, sometimes 7 to 10 years for very novel targets or first in class targets because it's required a lot of processes. Yes, it's not perfect, we, we don't like it that way. But this is what the day and the failure rate is high. So you could quote any number you want. 5%, 10%, you know, success or whatever, 15% if you want. Some people take offense of 5% but reality is very small percentage of those drugs will touch patients at the end of this lengthy, very expensive process. So if you set up the right expectation that AI going to change that. I'm going to make it two years and we're going to have, you know, the utopia of all of these 100% acceptable drugs. I think that's not the right expectation. If the expectation that we're going to use the technology to improve the processes that we have, shorten the cycle and have a higher chance. So maybe it's not 100%, maybe it's 30% success rate. To me, that's a huge success in the industry and that's what we have been trying to focus on. So you have a small processes. If I'm able to improve those processes and optimize them using AI technology, then after sometimes or overall, I will turn that into slightly higher success than what I have today. To me, that's the success. To me, this is how we need to set up the expectations.
- [08:41]
B
Absolutely. You talked about in upskilling and talent, about the need for employees to kind of get their hands dirty with these technologies to begin with, to really understand their limitations. And this has been very necessary. And I think in the broad strokes, you are, you are correct. But also this ends up a challenge, especially for organizations in regulated spaces like, like life sciences, in that you have to endorse a certain amount of shadow AI, and that itself comes with risks. Tell us a little bit maybe about how you're thinking about or how, you know, life sciences leaders are thinking about a limited capacity for shadow AI. What is acceptable shadow AI that really fields that, that, that allows this scaling to grow rather than inhibit.
- [09:32]
C
Exactly. So, so that's an excellent question because now it's interesting, a lot of new studies have shown that depending on the study that you quote, but about like 60% of people using AI with and approved sort of tools by their organizations in health care. And you know, again, we could debate that, meaning some of the people are using, for example, ChatGPT on their personal account, which, which you should not be using it, putting sort of business data in it unless it's approved or it's enterprise edition and is vetted by your enterprise organization. But that goes back to really the fundamental questions we're trying to ask here is, is it the knowledge that we need to give to people? Right? Because I think when people know or get educated about these are the tools that you're supposed to use, and these are the tools that you're not supposed to use, and this is why you're not supposed to use that. Right. And these are the challenges when you use it outside and we have, you know, this sort of approved tools that you need to use, and then this is how you use them, I think then the conversations become much easier. Of course, this is easier said than done because one of the challenges that we have today in AI education, which to me is the most important step that we need to take now to get the technology full potential that we all would like to have is that in healthcare and pharma we don't have content developed that is specific for the industry. What I mean by that I could for example come and talk about, let's say chatgpt and how do you use it in the company. But if I bring somebody today to talk about ChatGPT, general stuff and then how ChatGPT is used in marketing and you know, automating your emails, that's not going to be really useful for a biologist who will use ChatGPT. Completely different location, that doesn't exist today. Because one of the challenges again for all of those tools today that the blueprint or using the tool in your job is not out there. So for example, so let's say I'm a scientist, right? So my job is to write a protocol, you know, put patient narrative, summarize patient narrative, interpret clinical data, right. So how do I use it in my job? I don't really care about, you know, using ChatGPT in marketing. That's not going to help me. It might be appealing to me for example to use Copilot to automate some of my emails. Right? But it's not that much appealing, right? Okay. It's saving few hours a week, that's fine. But the gain will be how do I use it? Copilot, ChatGPT, whatever tool in my job and if I get to do that then organization going to start seeing the return on investment meaning they going to start seeing improvement productivity and improve off of decreasing the cost. But one of the challenges that we have today, we don't have that content out there. And the problem in healthcare content, it's either too superficial, I can use computer vision to read CT scan, It's like that doesn't help me too how does that work? Right? Or too deep and too technical where people is like this is too much for me. So we need to find and that's what we have been doing at Insights, try to develop those that content internally because we couldn't find it outside. So we have this educational seminars where I put all of these use cases, try to explain to people how to use it, have a grand rounds where we bring outside speakers that enable us to share their experience, how they are using the tools. With our organizations we have office hours where people can show up and say I'm trying to use charge P T or Copilot in this. How do I do that? Right? So I think all of this and we need Definitely more that give the organizations the opportunity to upscale or upskill the team to understand the technology and then you get better engagement and less use of shadow AI.
- [13:47]
B
Right, right. And you talked about upskilling at the beginning of that answer. And I know this is something that life sciences leaders have come on the show and talk a lot about, usually in the case restricting. But it sounds like you need a certain amount of what gets called shadow AI or in so many words, letting your employees use chat GPT to accomplish a task, albeit with some guardrails or at least guidance from the organization. Largely, I think over the last year that I've been hosting this show, I hear a lot more about how shadow AI needs to be restricted. But just from your first answer, it sounds like that really the best course of action is to endorse some portion of shadow AI and give it guidance. Tell us a little bit about how you envision shadow AI being integrated into the upskilling process and being able to drive effective AI adoption, especially in the agentic era.
- [14:44]
C
No, that's an excellent question. Right in the easy. Well, it's not the easy answer, but the potential answer. Let's put it this way. What I found in my experience, when leaders use the technology in their job, then they become all it. And the key is use it in your job. So we have leaders who use it, for example, you know, to ask like generic question, you know, or we have people use ChatGPT to, you know, I found it sometimes to help me help my children their homework. How do I open my garage today because it's closed and I don't know how to troubleshoot that. Right. So that's the use of ChatGPT. That's not what I'm talking about. What I'm talking about is using it on any tool. I keep mentioning ChatGPT but, but you know, any tool. So I'm agnostic of the tool. But any tool of, of generative AI that we have, when leaders start using it in their job to do their job and then they start seeing how it does it, then it click, then becomes easier conversations and less leaders do that. It's always going to be hard conversation because the conversation now, which is interesting, right, is that from leader's perspective, I'm putting an investment, whatever that investment is. Right. I'm not seeing return on investment. And the return on investment is a P and L change. So we're in that period of the challenge today is adoptability first before you see changes on the P and L. But if you from leader Perspective. I already put X amount of investment. I need to see that P L now and that's not going to happen right now. So an easy analogy I always use with leaders is like installing a solar system. You know, you're going to pay the money up front, you're going to put it and then the first three to five years you're not going to make money. But year six, it's all money, right? So, so we're right, we're in that period too. And I'm not saying that we should invest in everything and hope that we will capture that later on. We have to have a specific projects that is tied to roi. And one of the difficulties now is how do you turn productivity gain into, in the pnl. Right. So if I make my workforce more productive from neither's perspective, am I decreasing the number of the workforce or you know, unless I'm decreasing or somehow shrinking something or I'm not going to see anything on the pnl. But then what happened with that technology is if you get to people's like hey, use the technology and you know, we want to shrink the team and those things, then they're not going to use it. So the idea now is no, use it. Improve your productivity and eventually I will be able to do more stuff faster and better. That will enable us to capture the ROI down the road.
- [17:37]
B
Absolutely. And I think that tends to play into a fallacy around zero sum gains in hiring versus human hiring versus AI development. I think there's an idea both really among the leaders who haven't really touched this technology and very understandably nervous subject matter experts, labor leaders all thinking of well if I invest in it now, I'll have 10 AI employees down the line instead of 100 human beings. And it doesn't actually work that way in any zero sum sense. It really depends on your operations. It's not going to be this distinct trade off even as, as much as it's obvious there's going to be some level of job displacement that we see with these technologies largely. And I mean that that narrative has been around for, for a couple of years now. We used to tell people on the show, oh, don't even bother like bringing that up. But there's, there's just so much out there about what, what this job displacement is going to be. And I think a lot of those fears have generally been shown to be overblown once you get into the nuances of it. All that being said, you mentioned education as the largest barrier. I think even when business leaders, they're starting at ground zero, they know they have to jump into the pool. They're trying to get started with a. I, I think it's those organizational and cultural shifts that can be the most daunting, especially because those words tend to have the loosest meaning. You know, going from organization to organization. What do you, what do you see is what's most important in setting that foundation to make education the highest priority? What organizational and cultural shifts do you see are required to integrate agentic AI into R and D workflows with that education layer?
- [19:22]
C
Yeah, that's, that's another excellent question because as we talked at the beginning is really not about the technology. And now it becomes clear, at least in my mind, that two things you have to focus on as a leader. The first one we talked about, which is upskilling your team education. So that's, that's very important. I mean if you don't do that, I can bring the best AI multi agents framework in the world, give it to the people and I can guarantee you they're going to trash it and nobody will use it. And we've seen it in, in, in many organizations. The other important thing we recognized early and we worked hard on it, but I think sometimes organizations underestimate is that the current workflows that we have are not designed for AI. When you try to introduce the technology in a current workflow, sometimes people reject it or sometimes we make the workflow worse which make the AI looks like has no impact on the workflow. So I'll give you an example. So we're building internal model to try to optimize large molecule interaction on a novel targets we've been working on. We tried all the publicly available libraries, they really didn't work. This is very difficult target. We've been having difficulties in this target for a while. So we were able to build, internally trained on our data a pipeline that generated 400,000 AI generated molecules. And then in silico we were able to test the properties of the interaction of the large molecules with the protein and identify about 85 of those 400,000 that we believe in silico that have potential sort of, you know, good, good properties. So we were able to do that work in about three days. And honestly we probably would have shrink it even farther. That works typically in a lab or whatever. Experiments sometimes take between three to six months. So now with AI you could shrink that to three days. That's huge, right? So we wanted to validate this. So we give it to the team to sort of Go to the lab and validate it. It took the team about four months to get the result back to us. Why? Well, the team is busy, right? They have multiple programs, they are short staffed. And now you added one extra thing to their plate. Where is that extra thing going to be? Is it on the top of my list? Is it on the bottom of my list? Right. And then let's say my team capacity is to do for four, let's say projects now I have five or six. What do you want me to do? You want me to drop other two or do you want me to pick this up, give it priority? What do you want me to do? That's what I'm talking about, redesigning the workflow. Because this current workflow made it seem AI actually didn't add anything because we still took us about six months to validate the results. AI didn't have anything, but it wasn't AI, it was just the workflow. So one important message here, that we really need to think about our current workflows and then potentially start redesigning them to be able to capture the technology in them. Otherwise we're going to have very difficult time introducing the technology to current workflows. Which by the way, when you start doing this exercise, you realize your current workflows, forget about AI, they're not good, they need optimization in yourself, right?
- [23:09]
B
They can be bad. They can be bad. And you never had an excuse to really look at them or go through them with a fine tooth comb until AI came around. Which brings about this kind of false blame game you articulate so well in your last example about the workflow itself until we integrated AI into the process. And now there stands to be this situation where we end up blaming AI when really we didn't have a great workflow to begin with. Very interested in talking a little bit more about how do we mitigate that kind of false blame or that, that very problem. As we start to see especially agentic AI become more broadly embedded across life sciences organizations. Also as we start to see that, that large scale industry wide adoption, what happens to the dynamics we talked so much about how we, thanks to this technology, have astronomical scale in terms of inputs. But what would we, what would we see as AI becomes more widespread and adopted across the organization for those efficiencies on the output side? There's been a lot of talk for many years that we're going to see this change in the offer cited, 1.5 to $2 billion, 10 to 15 years that it takes to bring a drug to Market, especially since pandemic, maybe maybe five years is too fast to ask for AI to have a significant impact at least on, on those time and cost spend statistics. But what would we stand to see long term as AI becomes more embedded in the drug discovery process in a way that shifts those barriers trying to
- [24:49]
C
understand the process, right? So, so how do you do drug development? Well, you have target identification. You identify a target that is important. Then you sort of study that target in terms of like do you have the protein structure of the target, do you know the protein structure? Or you have to sort of find that protein structure and then how do you sort of tackle that target? At least in oncology where we work, you know, do you need a small molecule or do you need a large molecule? And then how you design those molecules to get to, from target to hit and then once you get that hit, how do you optimize the molecule to become, you know, study their, their impact and then how do you take that and you know, once you optimize it to make it very sort of best in class if you wish in terms of drug properties and then you drive it to the ind. So, so, so let's say that's the process. So in each bucket you will have a huge opportunity to use the technology. Right? So for example, target identification. Now we have multiple ways where you can. AI is great in integrating a lot of data. So searching the entire PubMed and the Internet and outside to try to search for novel targets where in the past it takes month to do that, now we can do it actually minutes. Right. And then other opportunities is to take large data and then use machine learning to try to identify novel targets in that data and then validate that targets in some of that big data. And there are a lot of work there, right. In terms of molecule design, docking, you know, interactions, large molecule designs. There is a lot of opportunity to design molecules in silico, optimize their properties in silico and then move those optimized drugs to be tested. That makes the process faster and cheaper because today sometimes you in, in a year you synthesize like a hundred thousand molecules maybe to get one or two good molecules at the end. That's a lot of time work, money consuming process, right? If I'm able to do it, give you those two molecules like the example I gave in the past where we worked on the process internally, give that to you in three days and then you quickly, you know, test that in the lab, say no, these are not good or give me feedback and then I take the feedback, tweak the model a little bit and go back probably after two to three times, you're going to get really good molecule. So. So in terms of drug properties optimization, I think there's a huge opportunity there. So this is how you start tackling it in step by step, try to introduce it. But keep in mind again the workflow and how you're going to redesign the workflow and then how do you educate people about using the technology and also what you need, what we found is also important a little bit top to bottom, meaning you're going to have some leader is going to come and say I want you to do this right. Because sometimes becomes harder. It's easy for all of us as a human being to just revert back to what we do every day, like change. So sometimes, unless that change is forced, you might not be able still to achieve the return on investment that you put in some of those tools. So you need a little bit executive sponsorship and push. Not too much by the way, because also that's can fire back, but also a push to say, look, this is a priority for the business, we need to do this. So you get people also to kind of move and see the urgency of that.
- [28:41]
B
Absolutely. And I think, you know, with those carrots and sticks, those incentives, I think especially once you get that formula right based on your organization, you're going to start to see a lot of that large scale adoption across the industry. Aziz, a lot of nuance here, a lot of specificity, which I know our audience appreciates. Thank you so much for being with us this week. This has been incredibly insightful.
- [29:05]
C
Awesome. Thanks a lot Matt for the opportunity. Foreign.
- [29:22]
A
I think there are three key takeaways from our conversation with Aziz. First, leaders must address cultural readiness, talent gaps and fragmented data environments before expecting meaningful impact from advanced AI systems. Second, teams need guided job specific exposure to AI tools so they can understand both their strengths and limitations. Reducing reliability, reliance on unapproved tools and improving adoption. And finally, existing scientific processes often block the benefits of AI and redesigning those processes is essential to shorten cycles, prevent false attribution of delays to the technology and enable measurable gains over time. If you have an AI solution, position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap for the opportunity to showcase your solutions to the executives currently funding and scaling global initiatives. Partner with Emerge, Secure your partnership@go.emerge.com partner that's go.emerj.com partner for further executive level analysis and to join our network of leaders delivering workflow impact with AI. Visit emerge.com on behalf of the team at Emerge. We'll see you on the next episode.
- [30:34]
C
SA.