Good morning, good afternoon, good evening, ladies and gentlemen, depending on whichever part of the world you are situated in. A very warm welcome to the 1st 2022 Unlock the future of artificial intelligence driven smart pharmacovigilance virtual event. Co hosted by Amgen AstraZeneca Moderna TCS. My name is Alejandra Guerchioff and I'm your host for this event. Today's event has seen the participation from more than 30 life science organizations around the world. In addition, our speakers represents the best of pharma and technology experience. During the course of this event, we will be focusing and brainstorming on few of the most discussed industry relevant pharmacovigilance themes and identify approaches to convert this concept to reality using the power of technology. Before we begin. A few housekeeping rules please. I'm not sure if not all of you are familiar with Microsoft Teams interface, so first. All attendees have been default, been placed on mute and thus would not be able to speak into the mic. Two, if you have any questions, please post it through the chat box provide in the tool. Our speakers will take up the questions one discussions conclude, which should be taken roughly 45 minutes. 3. If we are yet unable to answer all queries, responses shall be emailed, post virtual event. Our virtual event will be recorded and the recording shall be available online a few weeks after. After the virtual event you shall receive a feedback e-mail with the link embedded to which we will appreciate the value input, what you like about the event and in your opinion can be made better. This will help us to evolve and enhance our future virtual events and provide you with the best experience. So now without further ado, let me first introduce the four amazing professionals that will share this event with me today. And my first guest is Bharathish Rao. Bharathish leads the initiative in the area of using of automation cognitive technology in safety case processing at AstraZeneca. He works closely with therapeutic area, process and technology leads and oversee product projects and business as usual delivery with technology partners. He's a certified in global leadership from the Kellogg School of Management in Chicago and Indian School of Business. He comes today with 21 years of experience in life sciences and automation and cognitive technology in vision on safety transformation. He also oversee large scale transformation as part of pharmacovigilance implementation. In more than 75 pharmaceutical companies and managing victims across 5 geographies. My next guest and speaker is Michael Murphy. Mike leads the PV operation team within the global patient Safety at Amgen, a position that he has held since April 2019. Responsibilities include overseeing and managing teams, accountable for adverse events, case processing and medical review. Periodic reporting, safety reporting obligation for clinical, commercial and marketing programs and oversee the safety technology. Michael has over 20 years of experience within the Pharmaceutical industry, focus within pharmacovigilance with experiencing change management process excellence and safety system. Let me now introduce Arpita Bhowmick. Arpita is the Director of Global Contact Center Digital Technology for Commercial at Moderna. Her responsibilities include the omnichannel strategy and technology. Arpita has an extensive experience in managing big portfolios for Enterprise Contract Centers, Artificial intelligence and transformations. She has worked with global customers and partners like Google, Microsoft and Mitsubishi. She holds an MBA in Technology Management from IIT Delhi and is by training a mechanical engineer. My next guest is Mangesh Kulkarni. Mangesh is a safety physician by training and head of pharmacovigilance practice at TCS. Mangesh has handled leadership roles at several functional, operational and strategic levels for various pharma global business function units. He has an overall experience of 21 years in medical research, regulatory and drug safety and has handled leadership roles in various setups, included multinational pharmas, CROs technology enabled services and also brings today significant experience from his clinical practice. After this introductions, I don't know the audience but me. I can't wait to hear from our guest. So my first question. Goes to Bharatish and Bharatish. Let me say that we have seen the merging of pharmaceutical and technology industry during the last year and it's really a revolution and of course artificial intelligence and pharmacovigilance is one of such merging and example. So what is your viewpoint of this unification? Can you share your experience when engaging with the technology provider, please? Thank you Doctor Alejandra. Let me start with the quote. I'm all for progress. It's the change I don't like. This mindset, I think needs to go. People would love to talk about technology, would love to talk about AI. But. They worry more about what change it is going to bring. So #1. This unification is going to happen. Is happening already, so there isn't any Plan B for this. Second part of this is. Technology, when used in the right proportion, actually brings a lot of advantages now more closely in the PV world. As you can imagine the case processing wise the case count has increased. Now more so in the last two years when a case of COVID. For the vaccine companies especially, the number of cases just drastically increased. So it was something that nobody had planned for. And this is where the technology when embraced fully, will actually get a big advantage because every time there is an increase of case count, there is a budget pressure of to process these cases. So automatically the budget pressure will then have its own. Knock on impact. So thereby again technology is going to be our vehicle actually to help us offset those cost and go towards savings And equally the time that is saved for us using this technology we can invest elsewhere as part of more strategic decisions and strategic programs. So that is another scenario in our case for us for AstraZeneca. Our case processing partner also happens to be the technology partner as well. So in a way we were having a kind of a head start if I may say that way that we always wanted to make sure that we will work with somebody who will give us that advantage or know how of the case processing. So in our case that was a box that readily ticked from our side. The third part of this dimension is compliance now. While it is all good using technology, PV at its core is all about compliance now making sure that in spite of using technology, the compliance part is not compromised. So that's another dimension that we always have to keep in mind that compliance plays a significant or I would say centerpiece to whatever technology we want to use. The 4th part of this angle is actually knowing when not to use the technology, and I know it's funny enough, but. Technology is very funny thing it's you can be get carried away with it. So you should as a company, as a team, we need to know when not to use it. Not every time you should use the technology and then say I'm going to get a benefit. So always start with an outcome. Have that outcome to say, what does success look like and will AI actually help us in all of these? If it is not going to help you, then there is no point in actually making that happen so. For me to conclude you this unification of technology with pharmaceuticals and is going to happen, the pharma companies are generally seen as maybe late comers in general, but I think things are changing quicker. I think people are recognizing that AI is here to stay and actually we if leveraged properly, I think the benefits are going to be huge, not just the cost aspect that's a given, but. Just the amount of time it saves, the repetitive task and the skill set it actually brings and then it will enhance the team also in terms of the areas that we actually want to engage ourselves in future. So that's my view on that particular question. Excellent, excellent. Thank you. Bharathish and, I think you your answer is composed of lot of elements that we will discuss today, budget efficiencies, when to use and not to use. Timely partnership. So I think it's the great introduction to the rest of our discussion and in fact your answer triggered my next question and this question is for Mike and Mike because here Bharathish Bharatish said OK, when to use and not to use that technology. And The thing is I know you have a lot of experience working with artificial intelligence platform for processing of safety. Places. So what do you think is the biggest benefit when you use technology and to achieve those benefits from automation in PV, mike? Thank you for the question, Doctor Alejandra. I think Bharathish definitely led into some of that question. So for Amgen, we've been a little bit on a journey for the last few years looking at different automations and different possibilities and it's really to scope out our future strategy and that encompasses a couple facets that we want to make sure for the future you know us in the industry can handle, especially when you're talking about. Processing and I know you mentioned case processing which is a high highly transactional type of activity. So we want as a highly regulated industry, we want to make sure obviously compliance is key and front facing for us. So anything that we're looking for to do is to at least maintain and improve our compliance. Now not saying that we need improvements, but you know you're always trying to manage high case volumes, you're always trying to manage certain situations that can come up with a large. Volume of cases, I know with the vaccines that were mentioned there's a high volume that come in. So you have to be able to adjust with that and not just throw extra people at being able to process cases and I'm talking specifically a case processing here. So you know we look for automation and AI technology to ensure that we can maintain and improve our compliance especially when we hit certain areas that you know high influx of cases or things that may be a little bit out of our control. A few other areas that I would like to cover too really focus on efficiency and simplification. Like we said, we get complex cases. We wanna try to standardize as much as we can. With an AI tool. You know the system learns and can strategize around building efficiencies in process so that cases can flow very easy. The system learns and it also helps with some predictive analysis and that's what we're looking to do. In addition, that flows right into the quality because the more consistency that you have with the system producing data in. The cases and the flow of cases, the higher the quality, less corrections, which then roll into compliance. So as you can see, the theme really rolls along altogether. They're not individual benefits, they all are linked together. But you know. As we look to the future and we see that volumes will go up, we know that because as we get more drugs on the market, we have to be able to hand our higher volumes that also at the end of the day impacts our budgets. So that is something that we also have to look forward to in you know next 3, 5 ,10 years to make sure we can plan out what is our budgets going to look like, what are the benefits of doing the cost upfront now to get an AI solution in place so that in the future we can handle this this control of our budget and not have to constantly be increasing budget over year over year and trying to have more stabilized fiscal responsibility. Awesome, Absolutely, Mike. And let me hold some of the concepts that you just mentioned regarding compliance, regarding efficiencies in automation and budget because I will hold those concepts for my next questions. But you mention about volume and complexity and now I like to turn a little my attention into what is the complexity and volume that we are seeing today in terms of data collection and. And see if Arpita can give us some light on that because when we think about the challenge on data collection, the first one that comes to my mind and regarding also what Mike mentioned is the increase on the volume that he mentioned. And we will discuss more from different and diverse channels all around the globe of communication. And we know this has been an explosion of social media and. And multi channels, Omni channels, so how Arpita regarding this? How do you think that artificial intelligence driven platform can enhance the safety intake from those diverse reporting challenges? Thank you, Doctor Alejandra. And as you are correct and as Mike mentioned in Bharathish mentioned that AI powered platforms are very effective in elevating the repetitive and the human and the manual intervention in the entire value chain of the PV intake, right. What's the role of the AI in the current ecosystem is that it powers the smart capture across all the Omni channels through intelligent intent matching, right? They are diverse reporting channels but these AI driven tools are cognitive and they're empowers the self-serve and actually this is very imperative as you correctly mentioned the three ways of data right, the volume, the veracity and the variety of the patient data. That that's continuously being monitored by these AI engine and therefore it becomes very important strategy for bringing an AI engine manning the concierge of all these Omni channel or different diverse intake touch points and bringing in that feedback optimization and the ongoing training of this AI systems that becomes an increasing importance and as you know right like any other AI. System. It's just not we can't replace the human element, but it can complement the process, it can fasten it and it identifies some of the hidden AEs and PVs. So not to discount the power of AI, but there should be always a fall back. Option available for a human agent to handle those very intricate and sophisticated AEs and PVs, but AI gives a whole lot of power in the entire value chain. Excellent. And thank you, Arpita. I another very important concept to keep in mind when we use technology and maybe I will back to that concept of human and machine interaction, how we handle that and I will be back to that concept and it's excellent. So now maybe and regarding the complexity that you talk about the cases, let me bring my next. Question to Mangesh. And Mangesh, we talk about here about the challenges, Bharathish Mike Arpita, they mentioned one common element which is the increase in the safety case volume and the spikes, they don't, they mention increased, but also we know year by year, but then we have spikes and we talk about budget and we talk about people reallocation when we have these cases of a spike. So how an artificial intelligence can handle? Can be on a scalable approach to handle these safety case volumes and we talk about compliance here. So we need compliance and accuracy for ICSRS processing. What is your taking on this Mangesh? Thank you, Doctor Alejandra. Let me put this into two, three aspects. Now we have ICSR process which has got multiple steps, but let me group them into intake, processing, follow up and submissions, submissions to the worldwide authority authorities. Now with this process steps we are also challenged with as you know the analysts have said, very high volumes, high to very high volumes and especially with the pandemic times. We have experienced or we are seeing those volumes. The important other aspect is the unpredictability of those regular volumes. We cannot have a production factory model set up. With that kind of predictability and there is a timelines demand, there are training needs and there is also high scrutiny. So at a certain point of volumes growing, the volumes growing out of production capacity, the sources, you know, the databases, the human resources. It can become unsustainable for required for the required patient safety reporting demands. So artificial intelligence in the form of now again if I you know the 2nd aspect is the natural language processing, the machine learning, the robotic process automations, I would start with maybe if the question is about the scalable approach. I think to start to start with the first if I take 2 steps back. We should have a detailed process map and standardization in place as much as possible so that we can automate and further to standardization. I see that simplification of those steps also should be as much as possible undertaken so that we can automate or put the natural language processing or robotic process automations on top of it. So if I say that how do we? Scale up to all the case volumes or to all the products or all the complexities that ICSR process brings in. It may be not possible in one go, so we'll have to take up a staggered approach. So maybe the one which is, as the other panelists also have said. A highly rule based, highly structured, highly repetitive process. Such steps can be taken up initially with a again a risk based approach. What's the risk in it? Then do the plan, pilot, analyze and again plan and staggering it over a period of period from designing whether an algorithm or whether we are designing a model, NLP model and then deploying it. So it is it goes through that. Cycle. So we take up within that ICSR process the one which is highly rule based, highly structured, highly repetitive and then move on to the next one. The simplest example is a bot or a macro typically if I say in a case triaging or prioritization scenario or maybe you know finding literature articles with adverse event reports. There are many further applications like you have duplicate check which can be automated validity check of a case, automated seriousness autocoding. Auto Narrative is, again, you know, a very good example for doing the RPA. Or labeling part, so each model, algorithm or application tool needs to be tested for accuracy and precision. For continuous improvement, so at further complex steps like maybe know causal analysis or causality. One may have to take up that for large volumes and other process steps with other portfolio of products at a later stage. But I suggest when we scale up this AI for all the case processing in a large volume setup and this should be the in my view standardize the process as much as possible. Process steps and take a small sample size maybe and or one process step may be a key's intake. And then deploy, deploy the design the design, deploy the AI based solution for a less risk category to start with, monitor, analyze and then scale up. That's my take on it. Alejandra, over to you. Excellent. Thank you, Mangesh. And an excellent, yeah, excellent experience in terms of scale level. And I think we hear again a lot of important concepts for from your experience on AI and I see we see benefits, we see there is no doubt a need in the industry to. Start thinking out-of-the-box with the traditional PV process. You mentioned unsustainable. And I think the beauty of this panel and the amazing reach of this panel that I want to bring today is experience. And I'm get back to Michael because we hear from all of you the benefits of artificial intelligence. Mike, you have rich experience in implementation and now you are. In a post launch state of your automation tool. So I we hear benefits but I believe during the implementation and life process you have been seeing few challenges. So can you share some light for the benefits of other our audience today about those challenges? Thanks, Alejandra. Yes, I mean you know it's when you launch AI, it's not just a flip of a switch and everything works perfect. And you know, I think anybody that has worked in systems prior understands what it means to put a new technology forward, whether it be AI, whether it be migrating to a database, you always have your challenges. That being said, you know at Amgen we started a few years ago prior to launching our most recent AI where we did a little bit more simplistic approach and started. Implementing some bots to help with some of the technology, some of the processes. So we had this technology implemented. So it gave us a little bit of a flavor and experience on what to expect and I think that really did benefit us going forward on some of the expectations. But I think it's important that people understand that you know and also senior manager under management understands because you know these systems cost a significant amount of your budget. Where then you outline what your future fiscal impact is going to be and you can see that overall benefit. So what expectations are sometimes is that you flip the switch, everything's working great, finance is contacting you saying hey, you got the savings already. So I think it's really setting expectations not just within the group and I'm referring to my group here in pharmacovigilance, but wherever you know launch the technology to set that expectation. And I think people have to understand with AI, if it's an obvious evolution, the system will learn. So over time and it's gotta get experience with different in our situation, different types of cases and what that actually means for the system to learn. Now we launched a few months ago and you know right away we had fixes to do which is expected. I mean I think again it's all about expectation. You have to set expectation to real realistic expectations because in our area it's not, you know we don't have hundreds of systems out there that are already doing AI. This is still fairly new in the industry and going through some of the especially for case processing and I know there's a lot of other innovatives. Opportunities out there that other companies are pushing, but it is really about looking at how the system works and then being able to adjust to provide any fixes that you need to put in pretty quickly. And I think that just sets your strategy. You have a very specific Hypercare time frame that you work with your IS teams, where you work with the vendor, where you work with the business and with having that then you're able to adjust. And put in any changes that you need to make. We are seeing a lot of benefits, I'll be honest and I think that's going to grow overtime after we get a little bit more stabilized. But you know again it's expectations. You know people have to understand that when you start a new technology, the system is starting fresh regardless of even if the technology has been in place for many years. It's going to be different depending on what your company or what your situation is. And I think it's important that people understand that and that there's one other piece that's extremely important that I have to hit. And for anybody who's doing any implementation in that is it's part of my background is change management and that has to be very strong and it and that sets the expectation, it helps the users adjust to what they need to do and it helps actually your sponsors senior management understand what that's going to be. So change management is a key piece that is typically forgotten about and a lot of implementations, but that's something that really has to be on the forefront any time. That you're doing any changes or any system implementations. Awesome, awesome Michael, thank you. This is a great share of experience. So we see there are pre implementation business decision and leadership decision on invest on this technology. Then you have your change management changes and challenges of course in place. Then the conversations and expectations set up for the project by itself and the good collaboration and communication. With that technology provider, excellent. And now so this is from your experience point of view and from Amgen. So let me switch now to Mangesh and I like to connect. Now Mangesh, you work from the technology provider. So we hear what are the challenges that might face at his organization from a farmer, what about? From a technology provider side, what are the challenges when you work with the pharma industry in implementation of this technology? Thanks, Alejandra. I think, yeah. Again, there are three or four challenges typically can be foreseen and one of them of course Mike mentioned change management is a is not a challenge, but it's a big exercise and it's a very exercise that has to be done meticulously. The other things which are I quickly mentioned in my previous discussion. When we take up any automation or any automating any process, as I said the important one, the standardization or simplification of that process has to be undertaken first. I mean when we do that manually and if we are doing that repetitively, the lean and agile process or fit for purpose process is a good process to automate. Otherwise if we have too many variations in process and too many independent decisions in a process. Then like for example, typically an e-mail with you know, an adverse event reported and you know if the subject is unstructured, the AI, the automated automated machine or automated algorithm which picks it up every time. If it comes in a varied manner then we have a problem there. How much ever innovation we do with the technology, it's still it's it could be an error prone process. So the standardization is 1, which I see that. From the provider point of view or partner point of view then we can reduce multiple handshakes, we can do the leaning of the process, multiple accesses to the systems likewise and so forth. And Mike also mentioned about within the change management you have to set up a. Good governance and metrics of performance measurement. As we do the automation step by step, case type by type, we have to define the metrics of performance, measurement, accuracy, precision, recall and the you know, various statistical, although I'm not a statistician, but the statistical inferences that we want to draw and of course the experts, those who want to draw that which is the level that. We accept that you know from here onwards this will be the whole process will be handled by the machine and rest is the you know continued with manual oversight or manual process. So the balance between the machine doing it and the further support to the manual processing that definition and that governance with the decision making is another one which I see as a as a you know as an exercise. Then there are timelines to implement. Mike said about this, you cannot have in one go all the things in just a switch of a button, all the things machine would do. It's a process which goes through planning, analyzing, doing a pilot and then assessing the results so and so forth. These are typically three or four things in a setup where we have as from the technology provider side, I see that. Working with working with the pharmaceutical companies across the globe. Yeah, Thanks Alejandra. No, thank you, Mangesh and let me now because we hear from Amgen and we hear from you Mangesh. And I'd like to bring now Bharathish because I think in AstraZeneca you also have implemented. So we hear the challenges are good concepts, we hear technology and what is your experience? Can you share? Maybe real experience and key lessons learned in the process Bharathish. Yes, I think in AstraZeneca we have, we have a three phase program if I may say that way. And the way the program was devised was phase one was more of foundation. Now I think my quality touched upon certain key parts of it #1 is. For any AI project, one lesson that we all learned is we need lot of tolerance I think. Unlike other things with AI, it is going to take its time to learn now. Good things or bad things, it will learn. So the question is how well the humans train is, how well the AI will actually be reacting to it. So we can't simply wash away from our hands and then say, well AI is not working well. The question comes back to us it how did we teach that? So there is an element of as respecting this to say that AI is not going to be solving all the problems on day zero. The more case that will it pass through, the more it will learn. The more it learns, the better it will be able to predict. So the second part of this is that make sure that you have a pilot. And again within AstraZeneca we had this concept of fail fast or success fast, whichever way you want to define it. As part of that you test the technology. If it doesn't work absolutely fine, there is absolutely nothing to worry. But then make sure that you make that decision that yes, this is going to be the one which we want to start that. Again, always have the progression over perfection, make sure that you progress further. So in our case, we finished our foundation phase live then we finished last year our expansion. So our foundation was only maybe less than 10 studies that we started with some state. And again every time the AI doesn't work, it shows a mirror to me that's what it is. It shows a mirror to say, well here is how our data is. OK. I don't like the data, but what that's what we're collecting. So we need to go back and some cases it's an upstream impact, some cases it's a downstream impact. But either way this is the beauty of it shows, it keeps on showing as a mirror that OK, I need to do something. The second part is it will also tell us what aspect of the process needs to be tweaked. So there are some part of the process which we nobody knows why we are doing, but you know what, we have been doing it religiously. So this is an opportunity and we should not shy away from that making bold decision that you know what, this is the scenario that we have encountering. All these years it was all in control of human. Now human is still in control. It is just that the machine is taken the front seat, but it's been guided by human. So now this manual QC I think is something very essential because again I always give this analogy within the team. Also that having an AI is like hiring a fresh college graduate. The fresh college graduate comes with an excellent knowledge but no experience. But he or she comes with tremendous ideas and that's what AI brings us. It has got lots of potential. The question is you can't have an overnight experience. You will have to push lots of cases. Some will work fine, some does not work fine. You keep an eye on it. The question is, we should always again, the tolerance level should be slightly higher. And the last point, the lesson. And again, Mike already touched upon. This is stakeholder management. The general belief, and rightly so, is you put AI project in place, your benefits will start to come in immediately. So there is a dollar number coming and then everything. The cost is going completely downwards, which is fine. However, the rate at which it you will get the benefit needs to be weighed in terms of how fast you are actually pushing the initiative. Also there has to be a balance if you push it too fast. Your accuracy goes slightly on a lower side. If the accuracy goes low, then the dependency on human will start to increase. So the pace at which you will have to, you know, take this forward needs to be governed. And in AstraZeneca we finished two successful phases. We are now into the last phase, what I call the benefit, complete benefit realization. So once this happens, I think we can say that we allow the system to start to be more mature. And then the benefit will come there is it's absolutely sure, it's absolutely certain, but we should not push this to the part where this is not like a traditional project where you just start implemented and then you will generate testioms and you're done with. No, this is something it will be attached to learn every day. And I think that's the key thing that we also within AstraZeneca, we notice that OK, we need to be open, we need to be agile and where possible we should not shy away from. Learning and then in some cases unlearning. Also, we might learn something, but it's better to unlearn when the technology or the timing is right. Definitely and thank you Bharathish Rao at and we talk about challenge in this section and this set of questions. So we see a lot of challenges in the technology and the what what's come new and I think you touch the key point here learn and unlearn what we have done in the past to be able to move forward and when I think about challenge. Maybe and I like to get back to Arpita little now to see because when thinking about implementation and challenges we hear that but social media is something that literally is what bringing a lot of concerns today and big challenge to extract information from this million of social posts Arpita do you think that the technology artificial intelligence. And machine learning would be able to accelerate the process and mining of safety information from those social posts. And what would be a challenge here? What is your view on this? Yeah, thank you, Alejandra. Of course, with the incredibly high users of social media, this channel can can't be ignored, right? But given the ease of use and accessibility, both the Internet and as well as social media have high potential for faster a signal detection. What comes in mind for two major advantages of using social media data over the traditional a reporting channels is that this has direct access to the real time. Data, right. You, me are there every time active on the social media channels. If we happen to see any adverse event, we can quickly go and report it. But then and it's a very potential, a richer source of report, but not discounting the direct reports from SCPs and patients, but sometimes these reports are even richer than those that are filtered through SCPs. And in my opinion, accepting those social media datasets for PV intake opens a great opportunity to capture. Potential and hidden AEs in the vast mix of languages and representations because in social media is a mix of different cultural backgrounds or demographic nuances of communications and the machine is intelligent enough to scam, scan and screen and pick up those potential AEs from a lot of noise which is a little bit difficult if you put a manual tools and therefore AI plays a very important. Role in picking up these AEs & PVs from a huge set of datas. But definitely it comes with challenges and obstacles also. And what are those challenges? There's a high volume of data in the social media, right? And you can say sometimes it's called noise, right? It produces so much of raw data that it gets difficult to analyze and therefore we need huge computing powers. And if you carefully analyze, there's a relatively smaller. Or a very significantly lower number of adverse events from a large amount of data. Next, what I want to touch upon is the accuracy of the information. Social media users, they might not be sharing medical data. Data that is relevant to make an adverse event acceptable to the. Agencies or the PV operations team, So many posts that we come across do not describe the symptoms and doesn't have proper medical terminology and does not explicitly confirm what are the certain conditions or diagnosis for that adverse event or you know, what drugs might have been referred there. Because in social media we tend to see that there are many, many creative or idiomatic expressions of the terms that are not. Found in a medical terminologies. Are not existing in the medical lexicons, but do make sense for the end user or the patient, right? And the Third Point which I want to bring up over here is privacy and data protection, right. There's a personal data in the social media channel and the personal writing, the post presents the personal data and therefore we should be very cautious on, you know, whenever we use that data because that increases a lot of data breaches, risk of data breaches and cyber attacks. And the 4th point which I want to put forward is filtering. There has to be a statistical semantic. And many other approaches or methods to accelerate the data processing and filtering the right data is the most challenging task because there are lots of false positives and false negatives in the entire proportion of the results. So yes, it comes with challenges, but yeah, the benefit overrules the challenges and as currently mentioned by all the panelists that it learns over the time, right. And then if you have to unlearn and again there is a kind of, you know. For all walk run know approach in implementing any AI solution, but that's the future right? Absolutely, absolutely. And thank you what I listen the four of you. Before the this event, I thought there is no guidelines, there is no manual of operation on what to do and what not to do when we talk about implementing. Artificial intelligence platform. But after today and after the experiences and the knowledge you shared, I think there is definitely a value on what to do, what not to do on this. It's it was an amazing fascinating discussion from all of you and but now I'd like to open the floor from the audience and I see there are a lot of good and interesting questions today and let me. Let me go first with. Mangesh, there is a question for you and say well it it's clear that Artificial intelligence enabled process can handle safety cases but what about ICSR complexity and what can bring the best efficiency for complex cases in AI? Sure. Thank you, Alejandra. If I have to say ICSR complexity is there are complexities of types, you have clinical trial literature and the variables coming in there, then we have process steps. You know one process step as a case intake versus causal analysis or submission distribution. So you have case types, you have process steps and you have complexities which arise out of either the systems databases that we need to access or process in. Or the last complexity is the variations or you know differences within health authorities the way we want to report you know additional field and needs to be created at times or additional tick box in a you know case processing needs to be created or processed in a different way for particular country or health authority these variables or these complexities bring in and then we have to then see what is again as I said what is. Highly repetitive and highly structured. So you also have complexities of formats. You have complex a different formats coming in, some unstructured, some structured like in clinical trials. You'll have SCE forms coming in from investigators filled in with a definite structure there, but some of the other formats they would not bring that structure. Literature articles would have you a lot of format structuring but thing is there are lots of. Reports would come unstructured and that again poses a problem for AI to bring that kind of efficiency and compliance into the system. So again, my view is we take up pilot with a certain product case type, process step and a particular sample size and automate that to the fullest extent. Possible. And then take up the next more advanced task in the case processing and then go on with good documentation practices. Good machine learning practices also would evolve over a period of time. I'm just this is just a word. At this point of time there are no such good machine learning practice. But this may evolve that what could be those good machine learning or good AI practices that would be futuristic at this point of time? Does that, I hope that answers the question. No, absolutely, absolutely. And you also touch a very important point here on how regulations will evolve in artificial intelligence. As of now, there are not so many. Guidelines that we can see and reactions from regulatory authorities, but for sure the that is something that is coming on the horizon and of course discussions already are in place, but we will see more and more in the future regarding compliance and regulations. And I see another, I love this question and I'm going to ask Bharatish this question because the next one is very specific for Amgen. So Bharatish is a short but I. I have to select this question because it's amazing. Is what would be your recommendation to potential new adopters and? It I maybe we need another one hour event here for you to answer, but at least as I think this is amazing and I like to bring your pick your brain here. Not my first recommendation is do it. There is no, there is no need to wait. And I think everybody's experience is going to be different. So there will not be any magic bullet or you cannot say there is a template out there because as Mike also referred, every company's data is different, every company's process is different. So you cannot simply say that OK, I'm going to just wait for five companies to go live and I'll be the 6th. While the technology is going to change every day, there isn't any right time to wait and then start. I would say start sooner, start with the pilot And again the 3rd aspect is make sure and again Mike already touched upon this in terms of in the pursuit of the cost savings. Be realistic with your projection so make sure that this AI. It'll beyond the buzzword. You have to get inside and then see what is realistic for my company. What is realistic for the type of data that I have. Not every data will be clean. Not every data will be ready. So you cannot have an AI. If I can't read my own handwriting, I can't say that well, AI will do this. No, I have to improve my handwriting. It's the same way. If my data is not clean or cleaner, then I need to make sure I look after that. Then make sure the AI is actually. Going to help. So again there are aspects to this I would classify. Let's say first look at the intake, see how ready it is. Once that aspect is there, then you look at the main data entry part of it and again there will be process elements you will have to tweak. My other thing is be open, be open and be flexible. There is there should not be stringent or anything. There will be scenarios where we will make mistake and we might have done something incorrect. But let's acknowledge it and then be open and then be flexible to change tracks because with AI one good thing is if it learns an incorrect thing is also good because it is remembering it. If it's learning, a good thing is also correct because it is making sure you're going to predict the right way. So there isn't any right or wrong in that way. I would say just jump onto this bandwagon. It is good actually, but come with lots and lots of shock absorbers and tolerance. Yeah. I'm here, I'm here. I don't know what happened with talking about technology. I was like, my God, I'm here. I'm here, I'm here. They disconnected.Bharathish I think I hear everything that you say and thank you so much for that and. Again, this is like the manual of operation, the event of today because it's really enlighted from all of us. And thank you. And then the next question that I mentioned is direct to Mike because it's an urgent direct question. And Mike also I think this question talks a lot depends on your answer on where you are, what is your experience because is would you continue to grow? Your automation portfolio or Amgen identifying additional opportunities and possibilities with artificial intelligence. Interesting, know? That's a great question because it's like when do you stop or when, how far do you want to go? And the simple answer here is, I mean from high level management down, we have the message to continue to innovate and it's, it really is deciding on where you want to go. And like Bharathish said, you have to pick and choose some different areas that you want to focus on. Sometimes you go smaller, sometimes you go larger right now, I mean like I mentioned before we started. Some early simpler bots that actually were very beneficial. Then we took a bigger leap and did case intake and case processing with a touch on medical review not too much. We will further enhancing that with our next release and also we have an additional release coming this year and that's for inflow end to end processing. But there are other opportunities out there. I mean we've seen translations, we've seen things with social media, we've seen there's a lot out there, there's a lot of companies that. Promote a lot. It's good to. I would recommend a lot of companies want to give you demos, look at the demos. There's no, you're not buying into anything. It's great to see what's out there. It's great to compare. But for Amgen, yes, we will continue to innovate. We'll look at different opportunities and we'll look at the cost benefit ratio too to see what we can implement because obviously these do cost significant amount of dollars at the end of the day. So you want to make sure you're getting a long term benefit from that and also other benefits. It's not just about budgets and I know. People talk about it because it's always on the forefront, but there's a lot to do with just the data for predictive analysis. There's a lot of opportunity out there for in the pharma industry and I know other industries too, but in the pharma industry that can really benefit us holistically. And I think this forms great because I think we can learn from each other and then move forward. But yes, we will continue to innovate at Amgen. Awesome. Thank you. Thank you, Mike. And there are tons of excellent questions. I'm sorry to the audience that I cannot take and Arpita, I can't tell you how many of them are related with social media and Omni channels. And I I'm just trying to you know we select one that I think is going to be like summarizing your view, but the challenge is in implementing the novel AI solution for multi channel model what? What is your view and your taking point from there? Arpita lot of questions on social media multi channel. Absolutely. And thanks Alexandra for giving the opportunity to conclude. Yeah, I mean Omni channel is the way ahead, right. And putting the AI engines on the considerate of these Omni channels in the form of white spots, chat bots, self-serve tools, integrate across different platforms, automate and bring the direct web to case. These are, these are all talk of towns but. But we should be very cautious enough on what are the resources that we are putting in to bring those AI tools, How do I monitor those AI tools? How do I keep talking, optimize those models? And a kind of approach is that from low code to no code, right? If I spend a lot of time just to bring on or massage my AI tools, probably I'm not doing the right path and how do I do integrations, both downstreams and upstreams? And what's the long term impact? And does it really benefits us over over the long run, Is it something that that's good to have, it's just that everybody is doing and therefore we are in the band bandwagon. Is it like that or do we have to stick around on it and be a realistic and function around the guardrails of compliance and continuously be the digital 1st and offer the self-serve serve channels to the customers? That you want to report in a why don't you try your these self-serve platforms and then hand over the most integrate and critical adverse events to live agents so that I don't miss out the human aspect of the interaction while doing the Omni channel integration. So yes, AI is great, it's there to stay as all the panelists touched upon, but we have to critically evaluate it and this is a continuous learning. Approach and continuous journey which we have to be cautious but at the same time aggressive enough. Sure. Thank you. Thank you so much. And I'm afraid we will need to end this panel here today because we are spilling over the time agree I there are tons of questions. I'm sure we you know we will think about a future and follow up event after this one because we are not able to answer all of them and their interest. Please remember to send your comments and feedback on the survey and if you have. Any queries or questions? My e-mail is there on the screen. You can reach me out and I will. Try to forward your questions and answers immediately to our panel and also visit our website, the link alongside to understand more about the TCS ADD safety cognitive automation platform to transform the PV landscape. So with that said, thank you everybody for showing us today and look forward to having you in all our subsequent events. Take care and have a good day. Thank you.