From promise to impact with gen AI in healthcare & life sciences

Today, we're going to talk about the promise and the impact of gen AI for healthcare and life sciences.

I think all of us know the trend and the hype right now with AI, but a lot of our organizations are talking about what do we do in terms of using and maximizing the value from it. So we're going to talk a bit more about that.

I think when you look at AI, I'm assuming that many of you have worked with chat GPT and asked questions and looked at summarizing content. We look at it from three ways:

1. A lot of work right now in gen AI is around productivity gains, right? So how can I make my daily life better? How can I use copilots to make sure that I can automate part of my life? So one example would be when we look at pharmaceutical industries, we can help incentive compensation organizations reduce the time it takes to create their ICE reports using gen AI, right? So it's all about productivity. A lot of focus in the organizations right now is on increasing productivity.

2. But I think we believe that gen AI's real value will be when you start moving from productivity to deeper insights and finally better decisions, right? So when we talk about deeper insights it's not just automation, but how do you bring different data together to answer questions in a better way? Right. And I think I will reemphasize many times in the presentation that gen AI will not be the moat or the differentiation for the companies. It will be actually the data that will be the new moat for the organization. So how do you bring this data together?

3. So for example, will be, can I generate insights from 40,000 patient, you know, voice calls and give a better insight around that? Can I synthesize these insights into a multimodal kind of dimension and ask better questions or answer better questions? Yeah, I think the next stage which will be transformational, but it's still a way to go is how do you drive better transformation the processes itself and the way we work human interfacing with the AI is going to change the way we dramatically work in future enterprises, right? So that's where we won't be just doing summarization. We'll be asking questions like what physician reactions can be simulate based on the messages, right? Can we have generate new messages for physicians or patients using gene AI which was not done before and make that embed into the workflow itself, right? So I think that's where we see the real value we start getting with AI.

So when you look at that, the immediate focus, if you see on the left, it's all about driving productivity. But if you see on the top, it's more about can we shape these early experiences to shape the future capabilities, the talent, you need the organization readiness and how you solidify your foundations to maximize value from AI right.

So a lot of focus on the left will be about, can I make my productivity? Can I be better at what I do? But let's be clear that will not be the differentiation for any organization. Most of you will start using gen tools to be productive but you will not be differentiated. I think they are going to move in the future will be a fundamental shift in how humans do what they do right now and how they do it in terms of leading more competitive advantage for the organization.

And I think there are three factors there, one will be how do you make gen AI adopt faster in the organization? Second would be about how do you have better, more deeper generative AI capabilities? And third thing is how do you actually combine classical AI with gen AI? Right? Many times you just focus on generating new stuff. But once you have to understand that most of the complex problems are still will be solved using classical AI methods. So when you combine them together, that's where the real value will come. For most of the organizations including health care are life sciences.

So let's just go a bit more in terms of where the value will be generated. So I think this is an example of what the what the health care and life science organization will achieve or can do using AI on the top are the various use cases.

So what kind of problems I can solve in the research and development side? Can I make my drugs faster? Right. Um can I sell my drugs better to the right people, the right patients and the right physicians? Can I optimize my supply chain as well as in the enterprise level? Can I be more productive? So there will be several use cases. I'll go through some of those examples in the next slide but I think AI will also add value in two other dimensions for an enterprise.

One will be around productivity like we talked about, right? And that could include many things, day to day assistance. I'm sure you'll use many copilots office 360 to just make your life easy. Can I write my email faster and all that stuff?

Second thing will be content creation and summarization. Can I create new content and summarize it faster? There will be a lot of productivity gains across many, many parts of the organization, process automation will be another area. Can I use gen AI to automate my processes and make it more effective?

I think you must have seen somebody here, most of us, you are maybe technologists here. How can I use gen AI in my software development life cycle? Can I use the process of code development faster? There will be another productivity area for gen AI.

And finally, I think a lot of focus will be on knowledge management. How do I make the access knowledge in my organization across the enterprise be more efficient and more productive? So there will be a productivity layer which will be valuable, not just for health care, it will be productively for all the enterprises across the across the domains, the other area.

I think again, important for maybe folks around here as well. is there a lot of investments even ZS is making investment on. Can I use gen to automate my data management and analytics itself? A lot of money is spent by the organizations to get insights from data and can I use that to you know, use AI to automate that?

So the examples will be and I'll share some examples for use cases. Can I generate synthetic data using gen AI? Uh can I use data management tools to automate the process itself? A lot of investments, I'm sure even AWS is investing a lot in that area, data quality next generation reporting instead of getting reports can ask questions and get my answers using analytics. And finally, can I make advanced analytic applications using gen AI?

So there's a lot of areas where gen AI has a potential if I go a bit more deeper and very quickly, some use cases across the life cycle which we have started to work on. So if you look at the research side and the clinical side, many examples will be, can I use gen AI for example, to identify my drug molecules faster? It takes a lot of effort right now to identify which molecules I focus on. And that will be one area where gen has a lot of promise.

The other area is clinical trials. Can I use gen AI to make my clinical trial process more efficient? So there's just an example, I take a lot of time to write the protocol which determines how the trials are run. Can I use genAI to speed that process up?

Then of course many areas in supply chain, can I optimize the whole end to end supply chain process in the pharma space? For example, we use a lot of time to generate documents which are submitted in the FDA. Can I make that process faster then?

Commercial? Of course, commercial and medical affairs. Can I use gen to do better market research? Can I spend less money on market research? Can I use better capabilities to do medical literature review etc to target the physicians and all that.

And finally, enterprise functions, legal compliance risk, all the unstructured data they have. Can I mine that data and take better decisions? So many use cases across life cycle. I'll take some examples of them of what we are doing in ZS to solve some of these problems.

So for example, for R&D, we have some agents called Alter Ego. It's a ZS tool which we have built what it does is it can actually take all your clinical documents. And the Alter Ego tool can actually summarize and visualize all the SGTM data for clinical protocols, right? It can take all the protocol documents and summarize it for you and it can be used for protocol authoring.

Second is an example which is actually used not just in clinical but in supply chain as well. Can I actually use clinical GANs to generate synthetic data which can then be used for predicting patient drop offs? We have seen that we have led to 10 to 20% better efficiency in identifying patient drop offs within the clinical trial cycle time itself. Again, a huge use case which can be further scale.

And finally, in the, in the discovery side, can I precisely extract and link between T cell phenotypes and associated genetic allocations, right? Alteration. Sorry. So can I use GPT to do a better job in doing that? We actually did a proof of concept where we identified that we were able to increase that by 20% in terms of model accuracy, which is quite good for the early stage and actually led to 99% reduction from 260 to 2 days for screening 25,000 relevant abstracts. Because it was a very manual process to look at all documents, it actually reduced the time by 99%. So again, very good examples of actual usage.

I think if you look at supply chain again, in terms of time, I'll keep going forward many examples of productivity. If you see on the right side, large language models and machine learning is being used to automate the migration and tech transfer process within your process development organization. So we've been able to take manual processes and make them 50% faster in terms of new product and process change optimizations.

Another example is on the quality side in terms of your supply diagnosis, we've been able to cluster non performance issues using AI and actually that process.

And finally, another example of synthetic data, how can we use synthetic data to create data which can be used for what you call digital twin scenario analysis in supply chain, right? Another some examples which are showing some promise in the space.

And I think finally, a couple of examples on commercial as well. I think market research is an interesting space. We started actually creating a tool which takes all your market research data and put it in one place and the insights are available. So you are not doing duplicative studies across the board. Straight example of using gen AI for solving problems, which takes a lot of effort and time for the organizations.

Similarly many other examples of using genAI to mine unstructured data in medical review and in patient identification, content generation, et cetera.

I think then there are some examples in the enterprise itself, right? One example on the right, we've been actually able to use gen AI and of course, usual analytics to create a single protocol, single portal to have all the KPIs available across the enterprise. So this is called a real time sensing platform which allows again AI and gen AI and future capabilities to make the process of making these insights available faster. Again, some examples here.

Now you have many use cases but we are in a technology forum. So I just want to cover technology first. What will it take to scale this for any organization? So if you look at this example, I think there are three aspects from a technology standpoint.

1. One is your baseline cloud platform strategy. How do you decide on new capabilities to solve your gen problem? So at the bottom, you have your data management vector databases ops and governance, which is required to be the base to enable your gen foundations.

2. Then the other thing is what foundation models you use. And I think there are two main brackets. One is, can I just use what's available? So it could be, you can use a closed loop models like chat GPT or open AI and use them to create your use cases. But many of the organizations want to create their own foundation models and fine tune them. So can I use an open source model and optimize it for my own organization?

3. And so all of these are required to actually depend on that. You can decide which applications to run whether it's a basic application. It's like a simple chat GPD like use case for your enterprise or you want to make a complex application depending on that you decide the architecture you'll use.

Now, I know the next one is a slight eye candy, but this is how it looks like in terms of what will it take for you to develop a long term capability around gen? So if you look at examples like you need your cloud infrastructure, but at the same time, you need your MLOps to make sure that you're not just building algorithms but you're building algorithms which can scale. So how do you operate these LMs in long term?

Then the element is, how do you set the right foundation models? OpenAI Anthropic, Palm. Can you bring all them together? AWS Bedrock, etcetera. What's your prompt engineering capability? Ultimately, it's about the English prompts you use. Can I manage those prompts better? How do I govern. We talked about hallucinations in, in uh in gen AI. How do you manage that? How do you ensure that you are secure and risk free in terms of using genAI validation? Putting the ground rules human in the loop UI engines. Sorry, I'm going a bit fast, but just to say many capabilities required to bring together to make an architecture work. It's not just about one model, it's many things to which comes together to make sure that the use cases are working.

Now, I think again, the next slide, what does it look like? One example would be an architecture like this. This is an example of macAI which on Thursday, we will do a demo as well. This architecture shows how you can bring it to life on an AWS stack.

So on the left are all the models, you can choose any models you want, you should have the capability to bring to use these models and then combine them with the data, you have your PDF your images, whatever store you have. Once you have this data, the whole process of cleaning takes place. How do I convert this unstructured data into a data which I can use? So that's where concepts like vector databases starts taking shape. This is where you have clean data available to you after that.

How do you fine tune my large language models? So once you have got to the process of my data exists. You need to use the LMs to be fine tuned but not just fine tune. How do I orchestrate them to be purposeful? So then come the job of what we call the tool chain of large language models. How do you ensure it has memory capability? Which means when I write a prompt, it remembers for long term. What prompts am I writing? How do I make sure it doesn't hallucinate? So I have components so that I can make sure that it doesn't hallucinate. How do I make sure that I have various other prompt capabilities in one place? And finally, on the right side, how do you ensure that these elements are being enabled by the right front end? So your chatbots, your query engines, this is where you put the application layer on top. So that's an example of what it takes to actually bring this to life.

Yeah, but I think if you just look from technology, I think technology is not enough from a ZS standpoint. Few things have to come together for any organization to scale the value from gen.

So I think basically you need the right vision and strategy otherwise you will be ending up with 1000 experiments and no value to show. So what's your vision? Why you want to drive and what you want to drive with? AI a second thing is how do you identify your use cases? How do you prioritize them? Everybody wants to do AI but which one do you choose?

Yeah, once you've chosen them, how do you measure what's your value framework? And once you have deployed them, how do you make sure that you continuously prove the value of that to the organization? So that is one part apart from that, you also need your right operating environment.

Do you have the right change management capability in place? Right. Do you have the right capability in terms of your skill sets? Do your people know how to use gen in a good way? Do they know the right data capabilities, et cetera? And finally, you have to have the right enabling infrastructure, technology, data. And also how do you make sure it's performing?

Yeah, so all of this has to come together and this is where we have a framework we have built for our clients to take them through this journey, especially our health care licenses. Clients make sure that they use, find the right use cases, focus on value, realization, build the foundations and ensure that strategically driving long term outcomes, right?

So I think I'll skip this one. So I think just come, come to the close of the talk. So I think what does it take to make AI a reality? Um so I think there's a whole process of taking your vision and bringing it to life. And I think it's not just for health care, it's for across the organizations.

So I think even if the vision, the the pain point, it's death by use case everybody in the interface is talking about what use cases can I solve? But I think the main focus and the focus is on productivity. I think what shift needs to happen is to identify those really valuable differentiated use cases which are driving competitive advantage. Here is the remaining transcript formatted for better readability:

So I think you have to be a bit more purposeful about what you choose. And also how do you make sure that you've had the right partners who can take you through the journey of the vision path, you need to have the right access to infrastructure data and APIs.

The pain point is that everybody wants to do gen but how do you do this in a safe way? Right. Making sure that you're not sharing confidential data, especially patient information out in the cloud and all of that could have a crippling effect if you don't have a right. So make sure you have a right structure process put in the hands of people with the right guidelines, balance between innovation and risk and finally build a long term data strategy itself.

The third thing is ability to prototype and deploy fairly quickly, right? It's one thing to have gen use cases. The other thing is to deploy them fast. So your infrastructure and capabilities make sure that you have a center of excellence which can take these use cases and deploy them at scale and finally have clear KPIs to track success because once you start building your gen and deploying them, do you have a right framework in terms of KPIs which are measuring and proving to the, to the executive audience?

I think if I just my last one minute want to summarize in some takeaways, three things, gen AI is here to stay. If we all agree, I think it's a, it's a completely disruptive trend and it's going to create a lot of impact, but it's a massive technology and technology is just not, not, technology is not enough. You have to do beyond technology to get value.

It will not completely replace classical AI. So you need to have a strategy of bringing the two together. And I think finally, and I'll reemphasize proprietary data is the moat. That's what's going to differentiate you. Pharmaceuticals and life sciences, healthcare have enormous data. How do they bring this data to be used by? AI will be the trick which will differentiate leaders from the laggers.

There are many pressing questions. I don't think we can answer here now, but I'm sure I can help you in answering them long term. What do you buy? What do you build? What do you partner? Where do you focus? What do you invest on making sure you're focusing on the right things?

And finally, I think get going now you should not stop. But remember it's a marathon, not a sprint. It's still evolving. So there's a lot of hype but you can make some no regret moves to be successful. Thank you.