#412 Aashima Gupta from Google Cloud: Can AI fix healthcare? Unpacking tech change at scale

0:05I always say healthcare is data rich information poor. There's too much data and that means there's no dearth of data. There's a dearth of how do you, as you're pointing it up, how do you collate it? How do you interpret it? How do you make sense of that? The moment that we are in regenerative AI is the moment that healthcare will benefit the most. We are seeing the traction because healthcare does move at the speed of trust. AI has been in healthcare for more than a decade. This moment feels tough for us, James, because technology is far advanced to help with that meaning-making. Because of that, underneath frontier models being so advanced.

0:50I think that's a very, very exciting time. Next three to five years will be fundamental transformation. Hey, everybody. This week, I'm delighted to be joined by Ashima Gupta, who is Global Director for Healthcare at Google Cloud. She's on the board of directors for loads of different companies, previously Kaiser Permanente, JP Morgan, loads of cool stuff. Ashima, welcome to the Health Tech Podcast. How are you doing? Glad to be here, James. Excellent. Yes, because I had to delay because you're a little bit unwell which is going around particularly in the uk so much illness um so much illness yeah people need spending cloud credits on solving the common cold i think but yes we can probably get into that to see if there's anyone doing that uh you're probably on the board of a company doing it but i'm sure we'll get into it but yeah it'd be it'd be great for our listeners ashmo if you could tell your story because obviously there's so much there's so much in there in health care and in technology and yeah i mean to get to a point i think an enviable position, being a director for Google Cloud in healthcare.

1:57I think most people listening would certainly like to speak to you if not be in your position. So yeah, really interested in knowing how does one get to that position? Tell us about your career. Thank you, James. It made me reflect. I feel very grateful to be where I am today. And like many careers, it's never been a straight line. So today I spearhead healthcare industry strategy for product and solutions for Google Cloud. And that means setting the direction for our product team, our solutions team, and working with healthcare leaders across the world to help them transform their business strategies to define either new models of care, revenue generation, and improved patient experiences.

2:40I have 26 years worth of experience in growing, differentiating and improving businesses through technology transformation. But technology is something that I've been passionate about for as long as I can remember. My passion in the computer science dates back to the days growing up in India when my dad first exposed me to a local computer science course. And it influenced me to get into computer science. So I did my bachelor's in computer science, master's in computer science. But the spark started much before. I bought my first computer at that time. Computers, buying a full-built computer was very expensive.

3:22And so I spent money in assembling the computer together. So from that point onwards, there's no looking back. I saw how this box, an intelligent box, could do things, create things. And that was my first interaction in getting deep into technology. And then having studied bachelor's and master's in computer science, my first job was a programmer. And I kind of grew through the rank. My first job was back in India and then moved here in the U.S. and got banking as my first project with Fidelity Investments. I learned a lot. Banking is a fascinating field. We all deal with money. And this was about defined benefits, 401k and building solutions there.

4:18and in 2008, personal reasons, my dad got pretty sick and I lost him in a matter of a week. I had to go back to India and that was my first foray into dealing as a caregiver or as a daughter, understanding, you know, I couldn't get his reports. So that made me much more interested in this world of healthcare And I came back with a renewed sense of, we all want to solve problems. Like FinTech, I was solving problems. What type of problem do I want to solve going forward? So that, to me, was a very recent moment in my own life, that I want to understand healthcare. I want to go deep in technology.

5:05And we all are problem solvers. critical thinking is important but that's how we got into a big shift from a fintech to health care well in fact let me just jump in there a second because um i i can i can relate to that my um i told this story a couple of times on here but my my my brother passed away he was a professor of cardiology uh and he he died of cancer during the covid time not of covid um although certainly accelerated it um but yeah i i looked at my own career at that point as well and actually it's it's it's it's trying to find you know a positive and a negative situation of actually just being kicked into a new life almost or having there's that phrase isn't that only after you've lost everything are you free to do anything and not that you've lost everything but that there's a there's a sense of that you know losing a family member there's a sense of loss that i think does give you this kind of freedom almost because you're you for a temporary period you're sort of not afraid of the consequences because you've got that calling of whatever that higher purpose is it's kind of outshining all of the consequences for that kind of short period of time so like listening to that and tuning into that can it certainly did for me anyway it it definitely just kicked me into a into into a slightly different i mean it's off the back of that that i started some x the company that i've got now like it's it it and i'm not saying it completely is attributed to that but i think there is there's a there's a there's a strength you can find in loss isn't there yeah i think that's that was a trigger and james you hit that right i wanted to commit to a larger purpose can you work in a small way make a difference to help people lead healthier lives and what i can tell you about my journey in health care when i move from a bank to Kaiser Permanente, you see a very different set of pace.

6:58What we very quickly realize is people in healthcare are very, very mission-driven. And it's how they spend their time, both inside and outside of work, how they do it in energy. And I find it pretty compelling. Like, not that I didn't like my time in fintech. That work, this is very meaningful work. So you deal with the bureaucracy that healthcare is. You deal with the slow pace because you do think and you do believe that a meaningful change will be very impactful no matter how long does it take. So that drives you. And to your point, I think that was a turning point for me to go from a very fast-paced, you could argue one way or another, but health care is considered to be a laggard industry.

7:49but that mission and the purpose is i've never found in any other industry and i for me there's no looking back no matter what comes along this is the path i've chosen and i'm grateful for it before i talk about kp because i i there's i mean your role at kp by the way looking at your linkedin sounds right down my alley digital technology incubation and solution sounds ideal i want to talk to you about that second before i do though i actually want to take you back to uh I want to take you back to India and I want to take you back to that that buying the first computer moment from your father I think that's that's a really interesting one because I mean we were talking about it before before we started recording that I wish someone had looked at like what what I was actually interested in and like thrust some of that upon me at a very young age to just see what I would do and perhaps influence my career choices um but it seemed like you had that moment um were you one of the people that sort of broke it apart and looked at the innards or were you one of the people that like really stressed tested its capabilities from like a software perspective like what we what were you like with that first computer like what was like sparking you what were you what were you actually like passionate about and interested in with that first computer so it's a funny story to that as well right when my dad when i was growing up he was always wanting he was very forward looking for his time so he wanted me to have a good career and yet instilled in me, you know, figure out this field, this is going to be the future.

9:22So it was nothing naturally I drove to technology being the career. So there was that inspiration from him. And also in Delhi, it gets pretty hot. And the computer labs used to have the only AC there is in the school or the college. college so naturally I would spend a lot more time in the lab to escape the Delhi heat and so that kind of grew on me the more time I spent in the lab the better I learned and then it came we were from a very humble family background so buying a branded computer was expensive it was three or four times the cost so the motivation was can we get the computer built without paying the top dollar meaning and not in the indian rupees in that case so it was more means to an end to really get the same functionality and so that was getting the computer at you know a different price point but also it made me more comfortable with different parts different pieces how different pieces kind of come together and my first program was a basic program where this was back in the lab to print electricity bill or things very smaller things i said wow it could create it could do these calculations you know technology if done right is like magic connect to that moment like oh my god this is uh this that's that was kind of the foray yeah it's incredible when you run the code and it works the thing happens you're right it's indiscernible from magic right at a certain point in time which that's one of those funny laws or whatever isn't it that technology should be or brand new technology should be indiscernible from magic wonderful so let's go back to Kaiser so you're at JP Morgan your life circumstances change you desire a career towards health care you get that with Kaiser Permanente.

11:29So how did that come about? What was the role and what did you go in there to do? So I had come back from India after my dad's passing away, came back, spent two months freaking out. And Kaiser Permanente at the time had hired a lot of fintech executives. They wanted to run healthcare life with the same level of urgency as they thought that you'd get the discipline from financial because that's where it was moving perceived to be moving in a much faster pace so I knew some of my banking exec had joined and I reached out to him and he didn't have anything in his department but he made some connections for me and technology is such an equalizer James the project I was working on the products I was using were the same products Kaiser was using.

12:24It was a middleware I was responsible for at that time, the middleware server. And so the skills matched. But instead of building web services for a bank or a credit card division, which I was leading in the bank, it was for, and at that time, I was naive enough, It was to build it over an EHR called Epic. Like, oh, I did that. And got into the job and very quickly realized it's nothing like banking at all. Technology being the same, but still the stakeholders are very different. I had not worked with physicians' leadership at all in the bank. And I didn't realize in that you build a technology, it solves a problem, you roll it out.

13:14The whole aspect of change management, it's there in the banking as well, but it's a completely different level in health care. So I had to quickly learn that it's not about creating a technology solutions. It is about how do they solve the problem for two big stakeholders? One is the patients, other is physicians, which was a very different kind of mindset. And I made my share of mistakes in Kaiser. I still remember when I was leading the digital, we were building this pediatric app for childhood obesity. It was a cause that Kaiser was, Kaiser is a very mission-focused organization. And in that way, my technology hat, we built an app and to gamify that experience, that how much exercise you did, things of that nature.

14:06And I remember very categorically talking to a pediatrician and they were like, Ashram, you're solving childhood obesity with an app? Meaning that the kids will spend more time on the app, sitting on the computer. It completely defeats the purpose. So then you kind of think back. You really, really need to think a very different empathetic approach being told in the shoes of the patient and the shoes of the physician think of the solution do you think so it's quite interesting that you say that actually you know the comparison with banking that it's less about just solving the problem and it's actually solving the problem but now having to consider these these stakeholders do you think that the in some way the goal should be to simplify how easy how easy it is to make decisions and solve problems in health care or is are they i mean are those extra complexity how necessary are those extra complexity complexities do we are we weighting this correctly do you think like based on the work it did at kaiser like okay you come in and of course the big stakeholders the physician the patient of course i get it there has to be some waiting what's your what's your sense of how like the the is is there an imbalance of power here of actually getting things done like like what's your sense of that i think beyond the um the those stakeholders there's a lot of bureaucracy in health care in general right there's i think so there's something to be said about can we improvised the processes.

15:55So as an example, we had to, for digital, which was a new discipline at the time, we had to launch process let. It's a, you don't need, you know, 18 months to launch a new feature, a new solution. That was not the problem. And I think part of that is the way healthcare technology works. There's a lot of customer off the shelf like you're building on top of char or a billing system so i believe there's um so you move at their pace and they have so many clients so you are kind of in a queue so that was earlier on very clear to me and and part of the reason why i left kaiser to join a startup was after doing that for seven years you wonder it's not moving fast enough interesting because you can do the same thing and to me that software engineering discipline that exists in a tech company or in a startup in a healthcare environment that's very dependent on third parties and then that maneuvering that creating a roadmap that aligns with that vendor's roadmap and your roadmap that takes a lot of time so the startup that you joined tell me about that so in kaiser one of the things i was launching for and again story there um my my elder daughter is allergic to peanuts and any kinds of nuts and i have to be very careful with her and as she was um maybe in the middle school and we had gone to lake tahoe for skiing and we're coming back and my husband was driving and he said well i'm feeling sleepy let me find out near a starbucks so he pulled up his phone we go to starbucks he gets his coffee he gets two cookies and we we're off uh onto the highway again we didn't know the cookie had nuts and i think i started getting reaction and now i was able to find a starbucks near me using the app i started to look for er and hospitals near me just pretty complex at the time to find out the hospital that was open.

18:17We were in the middle of nowhere from the Sierra. So that led me to, and that, you know, all the products are sometimes based on your own personal experiences. So I wanted to open Kaiser's location API, like what facilities are open, where, like, why is it so easy to find a Starbucks coffee versus a life-threatening situation to find which hospitals open at what time and with the services so what is the answer to that by the way why why why is that i think because most of this information in the case of kaiser a was on the website it was not easily available as an api so yeah okay so um you need to know and that was uh my first at the time was opening that api is like really taking the functional Now, it's already public, by the way, but can you not make it in the PDF?

19:12Can you make it electronically accessible to this concept like API? And that was my foray into, oh, I could do that. And then I was building in Kaiser. Another moment was the diabetes management, which was a huge disease burden from Kaiser Permanente in our patient population. and if you looked into the workflow that nurses called people they had a panel of thousand patients or a defined set of patients and for them to call them every two weeks it's a 20 minutes call hey how's your diabetes how's your you know glucose meter reading things of that nature so very manual so my first project in that disease management sector was this remote monitoring of diabetes where we were giving the glucose meter with the connection, built it on the cloud and watched the fluctuation in the readings so that nurses, when they're calling, they have the data beforehand.

20:16Now, this is, I'm talking 2012. Now, there's a whole bunch of startups doing that work. But even that where there's a lesson in empathy I learned was day of the launch, it was a control pilot, 200 people we were launching. I was very proud of it as a technologist. Again, on the call, we had sent the playbooks of how do you use such a solution? And it was a beautiful looking book, bigger font. I was very happy. And first call, a gentleman picks up. He was from Colorado. And my first statement of that was open internet explorer. because I wanted him to use our website. And the question was, what?

21:00What Explorer? So it was... Interesting. It's humbling, isn't it? And this was a 70-plus-year-old gentleman dealing with diabetes, dealing with part depression. And now I passed off that burden to him, meaning that was, again, then we did the ethnographic studies, who are the people who need it, what is the best way to deliver the solution. I believe that's why technology sometimes fails in having that empathy for who's the deceiver. Are we making it easy or hard for them to embrace it? So what comes next? So then I joined this Apigee, which was the API build company. So I had gotten good at working with EMR, figuring out the API that could be exposed to build experiences like I was building on the diabetes management.

21:59That was a two-year journey. In that, I got exposed to interoperability because now I became much more on the product side that would be used not just by one hospital, but many. and got more exposure into this whole interoperability wave, the fire trackers and APIs, that it's not enough to open APIs. You actually need semantic interoperability. That's what the fire standard did. So as an example, if you're talking about building an app ecosystem, like we all know there's an app for everything, right, today in the app ecosystem if you go. But how those apps connect to your patient record, if it is a bespoke connection each time, that's where the cost adds up.

22:50That's where there's no standardized way to connect to that system of records. There's a lot of body of work with ONC here to standardize those APIs. So for clinical records, patient is called patient API. you know, your demographics, conditions. So they had figured out a few clinical concepts and the way to call it the same. That way, no matter your hospital A or hospital B or C, you're calling patients the patients. So those were the APIs. So I launched the interoperability fire accelerator in Apogee. And within, I think, 18 months, we got acquired by Google. And that was my entry into Google.

23:41And Google, as you know, is an amazing company where it's very consumer-focused. And Google Cloud was a division which was focused on the enterprise. That was my world. How do you take the amazing technology that is built by Google and make it meet the enterprise? So I call myself as Chief Translation Officer. we have incredible tech incredible engineering resources and if the world can get even the fraction of that we would change profoundly change the trajectory of how this technology can be embraced built at that scale amazing so there's something that i want to ask you about here semantic interoperability so you said it's not enough to just have apis you need semantic interoperability i.e everyone needs to speak the same languages all the systems need to speak the same languages if you clerk a patient in all of that stuff needs to be interoperable with everybody else that's also recording that stuff and then you can get interoperability across sites you can get interoperability between hospitals you can get interoperability across the network so that is a that is a wonderful concept and you're absolutely right fire standards absolutely have clearly and obviously helped with that i think one one criticism of health care is just how messy the data is and actually how much data we just don't actually capture the amount of beeps the amount of words that disappear into the air there's so much data that that that we that we really struggle with i'm interested like because i i understand that problem through the lens of a clinician my mind always goes to intensive care and i can just sort of like see all of this data just disappearing but actually loads of it being connected being collected in intensive care because actually you've got an arterial trace you've got a co2 trace you've got an o2 trace you've got you've got you've got every we've got everything there you've you've got venous pressure you've got you've got like anything you possibly got on that yet still there's people talking to each other in the corridor and there's like a blood gas machine over there that's not that for some reason that's connected or like whatever like there's there's lots there's there's lots that i see even even where we collect the most amount of data as a technologist that is in health care where do you where do you think we are and where do you think we're going with this like are we entering a world where we're starting to collect a lot of this data in which case how many semantic languages do we have or do we need like what's the size of the mountain here because i see quite a lot of startups coming up particularly ones because my eyes on it with oh we can take a load of intensive care data and we can start to predict decline and you know some have done okay a lot haven't so it seems to be a problem that isn't easily solved but clearly with enough data the problem can be solved it's just that i'm not sure how close we are to collecting it let alone in a standardized format and let alone sharing it across multiple sites hospitals organizations companies groups etc so i'm interested in your view i guess from a technologist's point of view how you see that problem and where we are and how it might potentially be solved you painted a pretty accurate picture i often say i thought about a fair amount if you hadn't noticed yeah i could totally tell um you are you've given it a thought and i always say healthcare is data rich information poor there's too much data right and that means there's no dearth of data there's a dearth of how do you as you're pointing it out how do you collate it how do you interpret it how do you make sense of that meaning making from that data and to me we talked about fire earlier but fire is just one subset that's just clinical record there's data coming from devices, as you mentioned.

27:51This data coming from our wearables, our glucose meters, these are called patient-generated data. And the data in claims, in your financials, where, you know, what treatments were given to you, all that is just one subset. Then you talk about imaging, like CT scans, X-rays, MRIs, completely different data modality. So to me, healthcare has been on this journey for electronic. We moved from paper to electronic. Now we're moving from electronic to digital. And I believe the moment that we are in with generative AI is the moment that healthcare will benefit the most. this is the industry this is the complexity of data standards for different data formats so we talk about multimodal in gemini multimodal means that inherently in this case block for gemini but it's important to understand that it is a very tall order to ask everyone to now comply and make and harmonize to one thing that people will continue to build.

29:13What multimodality in the context of large models means that they inherently understand different signals. And because Gemini was trained on images, on text, it's able to find the nuances and correlations. So we have technologies like semantic search now, where the burden is search is figuring out how these concepts relate together. So to me, you know, AI has been in healthcare for more than a decade. This moment feels different, James, because technology is far advanced to help with that meaning making. because of that underneath frontier models being so advanced um and i think that's um very very exciting time next three to five years will be fundamental transformation we'll see in healthcare three to five interesting i'll come back to that um so we made it what 28 29 minutes before generative hours mentioned so now we can actually start the podcast so yeah you say it's an exciting time and actually you said three to five years you mentioned gemini mentioned multimodal we know that there's you know med palm too and and there's there's obviously lots going on at google why don't you tell us what your role is at google what you do now as global director for healthcare at Google Cloud.

30:49Tell us what you do. Tell us what that means. And then tell us your relationship with models like Gemini and MedPalm and all that sort of stuff. And we can go into that. Yeah. So my role, as I mentioned, is a chief translation officer. I work with, so there is Google is an incredible company with amazing talent, researchers, engineers all across the company. So when they hire someone like me, who's come from the industry, my job is to figure out which technologies will resonate the most to solve a given business problem. So I work in that case. So for example, let's take Gemini. And when we heard that the frontier model is able to, for example, it's a very good example that it was given a silent movie change, which was a 45 minute silent movie, no audio or movie.

31:44a Gemini was able to ascertain the plot, the characters, the storyline, and no word spoken. Like it's able to understand. You know, you take that powerful technology, now you think in terms of healthcare, what is the implication? And to me, healthcare is inherently multimodal, right? You don't just see a CT scan or an X-ray. You see a human, you see an individual, right? When you go to a doctor, they see you, they see how you are, your physical, mental status. How you've walked in the room, what the pitch of your voice is compared to normal. And this, by the way, that multimodality is exactly why it's going to always struggle to build trust in this opening phase with clinicians, especially clinicians that have been around for a long time.

32:42because you know that that perception piece that like we all as human beings will just assume we're incredibly good at it better than any computer can do even i'm struggling to be like could this thing spot something before the person sits down like i could you know i mean the answer is absolutely yes and if not now it certainly will in the next five minutes when it doubles in power so like yeah yeah but no you're absolutely right i think that to me i i don't i think what health care is much more than multi-modality meaning multi-modality is understanding the context of the patient completely but that healing that judgment that empathy i don't think machines can do that so we believe very strongly that this will be your very powerful assistant that has collected the data for a physician.

33:38So take another example. We talk about population health, reaching the population, and take breast cancer as an example. Let's say as a hospital population health executive, I want to reach out to all the female who are from the age 45 to 55, have family history of breast cancer in the clinical notes that they have shared that information with us. They have a health plan. They're eligible for screening, but they haven't gotten it done. You want to run a query. You're touching six, seven different systems. So to me, it's not about diagnosing whether I have mammograms saying cancer or not. Of course, the models are getting advanced in aiding that diagnostics as well.

34:28But in healthcare, we're starting from very, very humble beginning. Can I collate the data and then figure out the trends for a patient, then level up for a population, level up to different communities? So I used to, in Google, we say, inpatient is a failure. outpatient. Outpatient is a failure of health at home and health at home is a failure of community health. So the more we, when we believe in that prevention mindset that means we need to collate this information much, much more earlier than at the end when you're now in the hospital or in the inpatient. I think that's where we have failed.

35:25So not all diseases fall into the same category, but if you believe in that paradigm, that of prediction, of prevention, that means collating that different modes of data and making sense of it, finding the trends, figuring out. And AI now is now getting into that realm. These models are very well advanced. And here's the difference, and I call that big AI versus small AI, meaning we had built AI systems that we've been doing in world healthcare for a long time. And in traditional predictive AI, you collected all this data, you labeled it, you then built the model, you trained it, and it gave over.

36:14now the general purpose models are so advanced that they've been pre-trained but what they lack is the evaluation i think that where the burden is that's where the effort should be in what are the evaluation framework that the output that these models are creating is ground meaning it's grounded on the truth it's citable it you know it's not black box you know where the information is coming from and how have we built this evaluation framework. And that framework would differ on how this technology is being used. If you're writing a discharge summary, yeah, have nurses, clinicians, what does a good summary look like?

37:01So getting that human feedback loop, we call it reinforcement learning by expert feedback, not just the human feedback. So getting those experts defining how would they evaluate the output of that model, as an example. So to me, the frontier models have significantly lowered the barrier of building this technology and building into healthcare, different parts of value chain, be it revenue cycle, financial, all the way to diagnostics. where we are seeing the traction because healthcare does move at the speed of trust change. So you need to do that.

37:48Healthcare, so you need to bring these stakeholders along and not every project is a Gen AI project. Some projects are good for predictive AI. So I think that's where I think a lot of discipline and governance needs to happen. and having that very thoughtful approach in where you will apply AI, where you will apply generative AI, where you will apply now. We talk about agents, which is the next step. And I think that distinction is critical. And to me, my other concern is, I think this is a new level of engineering. this is not things are moving much much faster like this is the every three six months is new advancements so now how do you it's not going to be just one model or the second model or the third model they'll keep coming at a much rapid pace how do you then create an abstraction layer where you can plug in what makes sense to you but what doesn't change is that kind of plaques on that how how i will embrace or adopt or absorb these different technologies in my work yes so many excellent questions in there and so many excellent points so the first thing i want to talk about is um ai versus gen ai gen ai versus agents all of this is getting conflated in the term ai and this is causing us a big problem and it's interesting what you just said actually about what is the platform onto which these models come and are audited for their ability and for what they can do and for their evidence and for their capability and all the rest of it it's almost like now there needs to be per site or organization a group of people or technologists and people that understand the clinical as well to work together almost to to be this to be this platform to build this platform to act as this platform almost but to certainly be a filter because it seems that i mean does that fall into the remit of a cio i guess but it's almost like this is a super specialist role it's almost like this is a mixture of the cio the ccio and a group of computer scientists and a group of clinicians that are early adopters of the ai gen ai and agents and they might be doing it in their in their alternative lives outside the hospitals and actually all these people need to come together and have a sensible conversation about what are we going to implement how we're going to implement it but to your point about a platform like what is our baseline philosophy here how much evidence do these things need to come in and do a job how much does the need need to be before we adapt that evidence requirement what if the the opportunity of bringing this in is so great that actually it's unethical not to do it and actually we really should be and that that actually is is i i think probably one of the most important things you've said about this this platform this filter this philosophy this group of people like whatever you want to call it that the pace that we're moving is so great that it's frightening almost to me it's so exciting but it's also so frightening because i i don't know where this falls with like medical device like certifications and and and like obviously in the us fda and here hm hra and like are we going to get are we going to see these models regulated as class three devices like is that ever going to be a thing that obviously makes people comfortable if not then okay we can split clinical and non-clinical and we can think about let's just deploy as much non-clinical as as humanly possible to speed everything else up but we don't leave that clinical behind because to your point with more and more and more data this is an absolute sitting duck for a decent model to come in and start diagnosing rare disease telling us the best investigations telling us the most likely diseases by differential percentage and all this sort of stuff or you know directing care and and exactly to your point freeing up clinicians to actually deliver the the two and a half types of empathy that you know ai can only deliver half of quite well you know half of one quite well so i i definitely agree i just wonder like i guess my question i like what do you think we're ready for now even when you think back to like kp and where you imagine kp to be now and not them specifically other hospital groups are available etc but like what what do you think we're ready for in the short term where do you think like these you know ai generative and actually i'm really interested where you think agents might be deployed like now if not already what what do you think's going to be first over these hurdles and over these lines that's a great question james and we we spend a considerable amount of time talking to the healthcare leaders like we are just seeing the opportunity and and one thing is clear that there's many parts of the value chain where the technology can be applied, but not all projects are generative AI projects.

43:24So unrealistic expectations can lead to wasted investment. It can lead to misallocated resources. And too many pilots, too many AI models will just further complicate it. So we need to focus on proven high-impact use cases where there's an ROI. So I'll give you some examples. In October, last October, we did a Harris Poll survey about growing burden of admin tasks in healthcare. Now, it's not the most glamorous place to start, but I can tell you it's very impactful. So for example, we talk about burnout. What is a burnout? When you talk, it's reading, writing, documentation. And healthcare industry is facing this crisis of paperwork.

44:14When you talk about doctors, nurses, administrative staff, they're drowning in this paperwork. And just to give you some stats, in that survey, clinicians said they spent around 28 hours a week on admin tasks. 28 hours in a week. and perhaps not surprising, but medical office staff spends around 34 hours per week and insurance claim staff spend about 36 hours a week. And there's a direct link between too much admin work, burnout and staffing shortages. And we all know like WHO gave a report out there's a shortage of 10 million healthcare workers by 2030. So, you know, you walk back from that reality that if we are seeing the trend line that there will be a shortage of 10 million healthcare workers, you need to plan for it now.

45:13And you need to create the capacities. Not that you're going to all of a sudden create more school. We should do that for sure. But you need to know where in my value chain am I giving? Are they working at the top of their license? This is a term we use a lot. When you double down, you know, what we found is paperwork takes, when it's where you're spending the most time, patient care often can take a backseat. But what if technology like As Power as Powerful, Alternative AI could shift the balance back in favor of human connection? In the survey, when we asked healthcare workers how they feel about AI helping with the admin work, more than 90 % of healthcare workers across the industry feel positive about this.

46:09And to me, the areas where G &AI can reduce errors in healthcare by automating tasks prone to human errors, such as data entry, clinical documentation, billing, Doing it at scale where a human, you know, look at all my notes for thousands of patients and tell me if I made a mistake. This is scale AI is good at. And to me, that's where we are seeing our customers innovate. A good example, we work with MetaTech. They are an e-assure company. They are using AI semantic search for their expanse EHR at Mile Bluff Medical Center. And what they found was in the pilot setting, clinicians are able to save around seven and a half minutes per visit because they are able to search the entire patient record and get a summary.

47:06To me, that's much more humble use cases of AI where each time you visit a doctor and many times you come to hospital from a different clinic, there's a 600 to 800 word notes. If you're dealing with a chronic condition, maybe you're making those visits more. So you have each visit, 800 words, 10 visits in a year, go on five, 10 years. Now you have a plethora of like 40, 50 pages document about a patient. Now, it's unrealistic to ask a human clinician or nurse to make sense of it. And to me, that's where GNI is a smart assistant to you to figure out, okay, these are the nuggets. And by the way, these are the citable references.

47:55We worry a lot about hallucinations. So that trust, safety, grounding is critical for that. So we are seeing use cases like that, like seven and a half minutes per patient, so huge saving for that. And it's not, again, in diagnostics, it's much more in the back office. Similarly, we are seeing with Highmark Health, they are using prior authorization. And as we were talking with the executive there, one of the stats they shared stuck with me. It was 30 % decrease in missed lunches. Ha! that's very real isn't it that's very real right instead of writing memo prior authorization requests they were able to have more you know peaceful known as a lunch to do this work 18 percent increase in discharging before 11 a.m like these are they have a hospital arm so time you're discharged could mean you know when you talk about bed management that's they saw that and uh so to me same same with hca we're working with them on nurse handoff like that handoff today takes 75 minutes can we use it and at scale these small incremental changes are transforming because they have a profound impact on where people are spending time i think that's really interesting there's a there's a few there's a few things i want to mention here so handover that you just mentioned like oh my goodness the amount of time like my entire career in healthcare and digital health and health tech whatever it's been called along the way over the last 15 years people have been trying to solve handover it's it's it's the messiest problem to solve and actually there's a there's a guy called elliot from a company called infinity um that that i can remember him talking about this that you try and solve handover and the problem is you start pulling the red thread of handover which is basically okay what is handover handover is a group of tasks that have accumulated during your shift that you haven't managed to complete that you need to tell the next person about that are all connected to a patient right fine so let's make this quicker so we need to understand the task more cool where did the task emerge okay why didn't task get done okay interesting and then you go a bit further like okay well if that's the reason the task didn't get done then can we speed that up you end up getting further and further and further back and eventually you've built an entire task management solution that basically is the only way to actually solve handover the reason i explained that is because i think you mentioned a humble use of ai i think that is a very that's a that's a cool phrase I think because I think we could all think about more humble uses of AI like handover rather than jumping to drug discovery the whole time and you know these things that are going to change the world genuinely that are already happening by the way but there are real problems right now and again using the word use at scale there are these problems happening at scale these margins at scale make a huge difference that clearly there's a there's a need for AI for the other thing i just want to call out with my comms hat on uh and my marketing hat on is that that headline 30 decrease in missed lunches is a stroke of marketing genius because we talk about all the time all the time about speaking to the audience that matters in their language my goodness there is nothing that speaks to a clinician more than losing a pen or missing their lunch so like or someone stealing your pen shall i add because you never frankly lose it you put it down and someone else picks it up especially if it's a nice pen so yeah i think 30 decrease in missed lunches is a wonderful wonderful phrase to use but again humble uses of ai i think is a is a very very very interesting thread i think generative ai you're right has a place and has a certain certain specific places clinically but non-clinically we really need to be thinking about this a lot and i think a lot of where people's minds are going to at the moment is clinical you mentioned earlier we all have personal experiences of healthcare with ourselves with our families and we all want to all of us that are sort of amateur generative ai users Gemini, ChatGBT, Perplexity, like we're all sort of amateur AI users really.

52:38You're a technologist that really understands this stuff and it's interesting where you're sort of weighting your excitement towards these kind of marginal gains of the humble use of AI particularly in the non-clinical world and all that sort of stuff. So yeah it's fascinating to me and I think especially with like i mean yeah you've mentioned this before but we're generating so much data in the community as well and there's so much emphasis on prevention that it's only natural i think for lots of people to try and skin a generative ai model for something out in the community to help with prevention sort of bill it as a lifestyle thing so it avoids regulation and get it out and all of a sudden it's telling you what type of headache you've got and like is it is it telling you to take prazitomol is it not is it saying that's generally the done so there's lots of obviously risk and stuff and loads and loads and loads and loads happening but i think with so much happening we're going to start to cut the wood from the trees um i'm interested in agents i'm interested in what you think about the agent model because this is another area where i think the ceiling is well you you can you get to an exponential very very very quickly when you think about the agent model because as soon as you've got agents talking to agents and generative AI talking to an agent which talks to a generative AI which talks to another agent all of a sudden could you algorithmically solve health care that is sort of my that is sort of my moonshot question Like, could the agent model at scale eventually solve healthcare?