Chief Executive Officer & Chairman, Bristol-Myers Squibb Co

Fireside Chat with Chris Boerner and Euan Ashley | Stanford Medicine's AI in Life Sciences Symposium

🎥 Jun 10, 2026 📺 Stanford Medicine ⏱ 35m 👁 37 views

How is AI reshaping the pharmaceutical pipeline today—and what has that experience revealed about what works, what doesn’t, and why? Chris Boerner, CEO and Board Chair, Bristol Myers Squibb Moderator: Euan Ashley, Chair of Medicine, Stanford School of Medicine The Stanford Medicine 2026 AI in Life Sciences Symposium convened leading voices from pharma, biotechnology, and academia convene to explore the potential of AI to accelerate the pharmaceutical lifecycle and deliver tomorrow's medicines. Learn more: https://med.stanford.edu/raisehealth/... Stanford Medicine: Stanford Medicine: https:...

Watch on YouTube

About Christopher Boerner

Christopher Boerner, CEO and board chair of Bristol Myers Squibb, has discussed the company’s use of artificial intelligence, its financial performance, and its approach to drug pricing. At a Stanford Medicine symposium in June 2026, Boerner said AI is “going to change and probably transform every piece of the value chain” of the pharmaceutical business, from discovery to commercialization. He noted that the company saw immediate impact from AI on regulatory writing, reducing the time needed for documents from months to weeks or days. He also cautioned against overhyping AI, stating that claims about curing cancer should be met with skepticism because “cancer is not one thing” and “when you treat it, it evolves.” On earnings calls from late 2025 through mid-2026, Boerner reported that the company’s growth portfolio delivered strong sales increases and that the company was raising its revenue guidance. He said the company was “taking deliberate actions to rightsize our cost structure” and remained focused on making a projected revenue trough “as shallow and as short as possible.” On pricing, Boerner stated that the company agrees with the need for U.S. prices to come down and for prices outside the U.S. to rise, while preserving “the ecosystem for innovation.” He also described the company’s Eliquis direct-to-patient offering as a way to reduce out-of-pocket costs for patients and “cutting out traditional middlemen.”

Source: AI-verified profile updated from Christopher Boerner's recent appearances. Browse all interviews →

Transcript (43 segments)

✨ AI-enhanced transcript with speaker attribution

Euan Ashley0:03

Welcome back to Stanford and to our AI and life sciences symposium. I'm Euan Ashley, I'm the chair of the Department of Medicine here at Stanford and a computer nerd by birth, cardiologist by training, and I have been fortunate to be here at Stanford for the last almost 25 years now. First things first, a huge thank you to the events team and the strategy team from the Dean's office for putting on today's event and in particular to Lloyd Minor, our Dean, for his vision in bringing together this group of people because you're very much part of what we're hoping to achieve here not just today, but this week. It's a pretty dizzying time for AI and life sciences. You know, you could pick whichever benchmark you want. The SWE benchmark is a good one for software engineering that some of you might be familiar with out of Princeton, real-world software engineering problems. When this was first launched in 2023-2024, the models were scoring, you know, 20%. And over the course of the last 2 years, the most recent Med Atlas preview from Cloud is scoring 95% on that model. What about USMLEs, medical students? You know, when those models started taking those tests, they were scoring 30%. Soon it was 90%, better than most medical students. And now on real-world problems, those models are scoring really highly. But the benchmarks aren't perfect. If you look more carefully at that SWE benchmark, there's a lot of leakage. So, some of the models have learned the answers. It turns out they're not just superhuman at intelligence, they're superhuman at cheating as well. And in my lab, we recently looked at the chest x-ray benchmark. One of my students, Muhammad Asadi, accidentally forgot to share the images in a new model that we were looking at, a visual model to score its ability to read chest x-rays. He forgot to share the images. How well did it do on the benchmark? It did really well on the benchmark. He built another model that went to the top of the ranking that did not access one x-ray. Right? We don't understand how these models work entirely and yet we're putting them in front of our patients. We're using them to develop drugs. So, we're hoping to dive into some of those things in a little bit more detail over the next few hours and we have a great program for you. And we're going to start right now with a fireside chat. We have some lightning talks coming up. Cancer will be a focus a little bit later. And we're going to finish with another fireside with Uratipravakar, former director of the Office of Science and Technology at the White House under Joe Biden. So, it's an amazing pleasure for me to welcome you here, but an even greater pleasure for me to welcome my guest. Chris Brunner is the CEO and chair of Bristol Myers Squibb. And so, our topic is going to be drug development. Now, Chris, welcome to Stanford. Thanks for being here.

Christopher Boerner2:59

It's great to be here.

Euan Ashley3:01

You know, another guest we had at Stanford last week is Demis Hassabis, who you have heard of. I don't know if you've met him, but he's the CEO of Google DeepMind. Also the CEO, just as a side job, of Isomorphic Labs, which is an important and now $2 billion funded organization focused on drug development. He mentioned while he was here, he said this publicly before, that he really felt, based on their experience with AlphaFold, that he might be able to move drug development from something that took years to something that took months to something that could take weeks. In another later session here at Stanford, he at one point recapitulated that point and said, 'Months, weeks, and maybe even seconds.' Which I think he was referring to the idea that originally protein folding was something that took years or months. And now it can be done in seconds by a computer. In your annual report for BMS last year, you made the point right up front that you thought AI could help shorten the timelines for clinical development by 30%. You're the one actually developing drugs at this point. How do you put those two things together? The excitement of the potential that AI can bring to drug development with the practical reality of all the work you got to do to put it actually in the hands of patients.

Christopher Boerner4:16

Well, look, I've been very public in saying that I think AI is going to change and probably transform every piece of the value chain of what we do. Everything from basic discovery all the way to how we commercialize our medicines and hopefully ultimately how we get our medicines into hands of patients, many of whom frankly may not get access to the medicine today. And I think that all along that continuum we're investing significantly to find ways to take cost out of the system. Remember, this is still a business where when you have a drug candidate, the likelihood that drug candidate ever becomes a medicine is about 5 to 7%. It's going to take 10 to 15 years and it's going to cost somewhere between a billion and two billion dollars to do. Anything we can do to take cost out, to reduce the amount of time, to increase the probability of success, and I do think ultimately find a medicine that we couldn't have found otherwise. I think all of that's going to happen and will we be successful? Absolutely. We're investing in Nvidia's super pod that gives us the ability to accelerate by 20 to 30% the time it takes from when you have a drug candidate to when you get it into humans. We've already said that at least 30% of the time in drug development, the longest and most expensive time, we'll be able to take out. So, I think all of that is certainly on the horizon. I do think there's a risk that you can overhype this and I think that's something we should probably discuss because you don't want to set this technology up for failure and when some of these claims get made, I think we're bordering on setting it up for failure.

Euan Ashley5:56

Yeah. I think it's really interesting you think about the value chain because clearly I think and we should unpack that for a few minutes here like at the different parts of that chain there are different impacts and I think that the point Demis was making is that you can make inference now very very fast and our models are getting better if we think about where AlphaFold started, where it is today, where we think about binder models that are just at a different level than where they were even 6 months ago. It does seem that there's almost sort of tech scale change happening in some areas. But I think we would say that doesn't necessarily translate to that 5 to 7 year timeline getting down to a few weeks. Tell us a bit more about how you're thinking about maybe start with the early stage and excitement around small molecule development or other modality development.

Christopher Boerner6:44

Well, I mean it may be helpful just to tell you the journey that we've been on around this because in some ways it's investments we're making and in other cases it's ideas that the teams are generating themselves. When the first chat GPT models were developed in call it November of 2022. We were about to do what every major pharma company historically does with new technology which is you surround it with lawyers and process and strangle it until you have no choice but to engage with it. And gracefully we had a head of digital transformation on the team who wasn't from pharma and said, 'Let's not do that.' And instead let's put this on the desktop of at least a thousand or so executives and get them sort of working with it and 3 months later we had done that. Today when you log in at BMS 34,000 employees have a full suite of AI tools and the idea was, 'Let's let a thousand flowers bloom.' Let's just give folks this technology and we've continued to add to it and let them embed it in their workflow. See what ideas they could come up with and then across discovery, early development, that's certainly been true. And then in other areas we've said, you know what, let's make significant vertical investments in these areas to try to make transformational changes. So early discovery, I mentioned what we're doing with Nvidia. We now have 100% of our small molecules and about 75% of our large molecules go through a suite of AI testing to assess the probability of success of that asset moving to the next stage. And then where we find there are opportunities to improve, it gives us suggestions as to how to modify the asset in order to increase the probability of success. Once we get to the clinic, we're able to identify specific patient populations where we might better be able to do some of the early testing. And that sort of thinking transcends every piece of the value chain as you go through R&D.

Euan Ashley8:44

And is that software that you've built yourself or are you partnering?

Christopher Boerner8:48

Yeah, we partner, but you know, we've hired a ton of engineers and that's increasingly hard to hire these engineers and we can't afford them anymore, but you know, you're hiring engineers to try to develop these tools internally. One of the challenges we have right now is we're building these tools, we're partnering to bring these tools in, but because we've let a thousand flowers bloom, we also have people doing it on the side. So we're trying to rein some of it in to ensure we're being thoughtful about how we're investing, but you know, we've got significant investments in AI across every aspect of R&D.

Euan Ashley9:23

Yeah. One of the I maybe hesitate to call it a critique, but one of the let's call it a critique, why not? Like of the AlphaFold advance was that it was built on multiple decades, of course, of training data that an entire community had put together. Now, I think not to underestimate the engineering work that was done to build that and to make it freely available. I would never underestimate that. But, the question as to whether that can be repeated across other lakes and small molecules, for example, where is the equivalent training data? Most people would probably guess, if it exists anywhere, exists in a pharma company. So, how do you think about that in terms of building your own models that might be bespoke for the questions you're trying to answer, particularly in this early stage drug development process?

Christopher Boerner10:11

Well, undoubtedly, the challenge that we've had is number one, whether you're building yourself or you bring in technology from outside, it really these models are only as good as the data that you put them on. And so, you know, we partner with outside organizations using our proprietary data. We bring data into the company, but the biggest exercise we've had to do early on is make sure this data is curated, it's all in the same place. I mean, believe it or not, this company is 170 years old. A lot of our data exists in notebooks and Excel spreadsheets. I mean, this is a monumental effort just to collect this data and put it in one place and curate it so that you can get at very specific scientific questions.

Euan Ashley10:54

And is that something you've been doing for multiple years that you sort of seen this coming or are you struggling like most people to catch up?

Christopher Boerner11:01

We're catching up. I mean, and we're doing this everywhere. I mean, this is obviously a scientific discussion, but you know, just managing our financials. I mean, this is all on Excel spreadsheets. And when you want to run any of these tools, the first exercise is a whole team of people have to go in and try to find these data, curate them, pull them together. But, when you do that, what you can do with this is just extraordinary. And I'll give you one example from this morning. I had a call with the We were moving a product from phase one to phase two. And when we did, we made a slight tweak in our manufacturing process. Now, this is a particular technology we've been a leader in and somewhere along the way, 2 years ago, maybe a year and a half ago, somebody working on the team said, you know, we've got all of this data. Let's bring it together. They then hired a team to build some tools, AI tools around it. And what we were able to do is when we started to see a spike in a safety signal, these guys went in one week's time from the identification of the signal to finding that there had been this small tweak in manufacturing that was likely the culprit. We've run to the ground and we're now going to probably be able to accelerate getting the study back on track. That would have taken months to do. Had not somebody just taken the time and energy to pull that data together, build the right tools. And that was literally from my 8:00 to 9:00 this morning. Was talking about how where we were with this program and what we had done. That's happening everywhere across the company.

Euan Ashley12:45

One of the other interesting areas I think is where you would see that even in your deepest coffers with 170 years or looking at all the notebooks, you don't have the data yet that you need to train the models you would want or to get the candidates you would like. And I was intrigued by a post from Stanford's own Daphne Koller who has, you know, the CEO of Insitro. And I think you have a relationship with them. And she was talking about some a new candidate or some work you were doing collaboratively on ALS. And in particular, she had alluded to, I think this was in a LinkedIn post, to 200 cell lines and starting to do sort of multiple levels and maybe even multi-omic analyses including high you know, large-scale photography with perturbation of cellular models to come up with new targets and working together with BMS on that. And that now you were moving those together towards the clinic. Can you talk a little bit about maybe specifically that program if you like, but the general principle of when there isn't data to train the models that you need and you have to go build it.

Christopher Boerner13:50

Yeah. Well, I mean, that's actually more often than not the case, especially when you're moving into a new area. We just invested, right after I became CEO, we had had a small sort of early stage set of programs in neuroscience, in neurodegeneration, and I wanted to begin to accelerate the company's re-entry into that space. And so we went and we bought a company, we launched the first new medicine in schizophrenia in decades in neuro site, and said, 'Okay, let's put these two together, let's build a franchise.' But because we had been out of that space for so long, we didn't have a lot of the data, and we certainly hadn't curated it. And so we've begun actively doing partnerships with companies in that space, including other pharma companies, frankly, to try to say, 'Okay, can we start getting data together to understand better the nature of the biology behind diseases like Parkinson's, like ALS, like Alzheimer's.' And frankly, anytime you're going into a space where you don't have deep history, you're going to end up having to do that. And quite frankly, you're probably going to end up having to do more of these industry-wide collaborations to have the right data sets curated in the right way to train these models to really get the most impact from it.

Euan Ashley15:06

Yeah, and that's great that you mentioned that because I was wanting to ask you whether you know, how you view the sort of pre-competitive space with otherwise companies that you would consider less partners and more competitors. How do you think about that in this context though of sharing data in a way that would lift all the boats?

Christopher Boerner15:26

Well, I think, you know, we're still emerging here and every company has a different approach to it, but certainly as we think about whether it's institutions like this, companies like competitors in our space, finding areas where you know you're going to be working on some basic scientific questions, way easier to align that you want to sort of share data. And then obviously as you get more into a bit later in the process where you've got potentially competitive projects, then you've got to step back and say, 'Okay, where do you want to collaborate versus compete?' But at the end of the day, this sounds like a marketing tagline, but for most companies in the sector, and certainly we've thought about it this way, the competitor at the end of the day is the disease that you're trying to go after. And, you know, I was just at an oncology conference and I was asked before a CNBC thing, 'Well, what do you think about this competitor or that competitor?' And I said, 'Look, we're fighting cancer. Like, whether And by the way, if you know, we're all going to have our moment in the sun. And so, at the end of the day, do everything you can to compete. That makes us faster, it makes us smarter, that's all a good thing, but you know, we got to stay fixated on what the real culprit is here, and that's the disease we're trying to go after.

Euan Ashley16:44

No, I think that's really well said in terms of the sentiment. Moving a little further along the value chain, to clinical trials, because that's where the time is going to be very hard to compress. And so, I'm really interested in your thoughts and how you're thinking around AI when you think about both patient identification, recruitment, moving on through the trial, even clinical trial endpoints. Certainly something we at Stanford, Jane is in the audience, I think, has written elegantly about this. How do you think about it?

Christopher Boerner17:16

It's the longest, most expensive part of the whole process. And it is the place where we think actually AI can be probably the most immediately impactful. And so when we first started making investments in the space, we immediately saw impact on regulatory writing, for example. You know, these are teams of hundreds of people who spend months writing documents that are hundreds of thousands of pages long. And we took that down from months to weeks to days. And so that's an example of where you see that kind of transformation. That was immediate. We built an entire new unit in India where all we did was put people on drug development using AI technology to try to accelerate components of this process and do it faster. So, patient selection, site selection. What sites have the right patients for a given trial? What sites are best in terms of enrolling those patients? To now we're beginning to think about when you run these very long studies as you know, especially in your space. When a site begins to have trouble we can now get ahead of identifying that this site is slowing down, help them figure out why they're slowing down, synthetic endpoints, trial designs. It every component of this screams we can use AI to help. One of the big challenges is you got to get regulators to come along with the program. So, we've spent a lot of time in Washington trying to work with the FDA to say, 'Okay, let's get smarter with this technology and where we can use synthetic endpoints and the like.' But once we get that, this is open for us squeezing the time, being more targeted patient selection, and ultimately hopefully getting better outcomes.

Euan Ashley19:17

Yeah. And is that data that not this time the data for training necessarily, but the data that you want to run the algorithm on. Is that readily available? Is that data you have at the company or do you have to go out and collect it?

Christopher Boerner19:28

It's a little bit of both. We have a lot of data in the space. And so we can begin to train the models, but then obviously every project is unique and specific to whatever it is you're working on, what part of the drug development process you're in. And so there you may have to go out and collaborate with others. It's also an area quite frankly where industry is prone to collaborate because we all face this problem. And so us being able to accelerate the process generally is something we're all interested in. We let ourselves compete on the products. And clearly getting to market faster is better, but if we can shorten this process generally, make it more efficient, it's a tide that will raise all boats.

Euan Ashley20:12

Yeah. I ask partly just because as you mentioned in my area as a cardiologist in particular, long studies, very expensive, take forever. Right, cardiomyopathy, finding patients can be a thing. And we have some work again being run here by Sneha where it's funded by the American Heart Association to try to find patients with hypertrophic cardiomyopathy. As it turns out you have a great drug for that. But this is somewhere where literally everything's aligned. Patients are aligned, our academic health centers are aligned, community cardiologists are aligned, drug companies are aligned. If we have a good drug that we can get to patients who don't know they need it and can benefit, then that literally is the definition of a reason.

Christopher Boerner20:55

Well, and that's where we've actually used technology to try to identify those patients and use AI to you know, I think you know this project well. We were able to say can we take the exam results and say you've got using AI, you've got potentially a underlying cardiomyopathy that we need to now refer you to somebody who can give you a definitive diagnosis and ultimately give you a medicine to treat. When we first got into that space, I don't know what the rates are now, we assumed that about 25% or so of patients were actually being diagnosed. And obviously this is one of those that sneaks up on you and can lead to catastrophic outcomes if you don't get somebody treated. So finding those patients sooner rather than later is important.

Euan Ashley21:43

Yeah, I think a classic situation where the symptoms patients present with can be very non-specific. But the diagnosis is very specific and then the therapy is by definition precision and it works well for them and some people can be thinking that they have very non-specific symptoms, but when they get the right therapy, they suddenly realize what they've lost and it's a great example of that. I'm going to mention we can take some questions from the audience. So have a think and in a moment or two we have roving microphones that can come around if you want to follow up on any of the topics we've mentioned here or any others. You've mentioned the regulator a couple of times and that's always a challenge and we used to thinking especially place like this part of the world where we think about technology a lot. Technology moves very fast. By definition regulators, you know, they're not designed to move fast and we don't really want them to move too fast. How do you think about that as a big pharma CEO?

Christopher Boerner22:39

Well, we spend a lot of time trying to engage with regulators on this particular topic. Now, there's sort of a good news bad news situation. The good news is that if you think about the FDA, this administration has at least been open to having more conversations about how do you use technology, how do you try to reform to enable us to do for example synthetic control arms, for us to find ways to accelerate clinical trial enrollment and the like. Of course, the challenge is there's been a lot of movement at FDA, and so finding enough stability in order to get some of these things done is a secondary challenge, but look, it's just a matter of trying to keep regulators up to speed on where the technology is going. So, we have a lot of conversations on that. And then it's project by project. And the more that you can engage and try to get sort of novel ways of thinking about how you can incorporate this technology into these clinical trials, it's one after another, and you just have to keep at it.

Euan Ashley23:43

Yeah. Okay, so if anyone has a question, stick your hand up, we'll send you a microphone. I have one more question for sure I want to ask. Someone will bring your mic. With something And there's one question here, someone Another one here. Someone will bring the mics. Let me just ask you this so before we get to that question. With the technology that is so ubiquitous, like we've talked about the entire value chain, we've talked about everything from designing literally at the molecular level in silico design all the way through to clinical trials and regulation. How do you organize that in a complicated company, right? Do you have an AI division that is over here that now has a remit or whatever? Do you try and embed it in all your different therapeutic areas? How do you think about that from a complex organization?

Christopher Boerner24:28

Well, it's an evolving discussion. So, we have an AI organization that we begun to model on just essentially all we want them to do is be focused externally and to set standards internally for how work should be done. And the reason for that is if you're a clinical scientist at BMS, you have a day job. You may not actually spend a lot of time thinking about where's the technology going. And so, you want to have somebody you can go to internally as a consultant say here's a problem I've got how do I solve it do I have to build it myself can I contract with somebody has somebody already built it and thus I can just tap into it and so making sure that we have the right people who can help the teams internally get the work done for the bigger projects we have to have a whole group of folks that we can have go and actually execute this we have to have now a whole finance team trying to figure out how much money we're spending on this what sort of return on investment we're getting I became CEO 3 years ago and I remember talking to investors who said please don't tell me you're going to solve the world's problems with AI because I'm so tired of hearing about AI. The next year they're like what are you doing in AI? And now they're saying what kind of return on investment are you getting and how much money are you spending on this and so this has been a really short period of time when you went from complete skepticism to what are you doing and how much are you spending and what can I get for it? And as that conversation has evolved we've had to evolve how we're structuring around this and that's whatever we're doing today it's going to be different next year.

Euan Ashley26:09

Yeah. No, I think the curve from what's a cloud token to you know can I afford all the cloud tokens my team need you know. Just that you know we were adding up how much we were spending on it nobody was factoring in the cost of the tokens and you're like well that's a miss so we got to get back in and to gain for someone else. That's right. All right, let's take a question over here first of all.

Vinod Kumar26:33

Vinod Kumar, professor mechanical engineering Texas A&M. The data questions you're working with company to company, you know a lot of knowledge is lied in the startup domain in the academia have you thought of opening the data to a bigger audience maybe create a mechanism somehow to do that?

Christopher Boerner26:53

Yes. The big caveat to that is the following. You know, healthcare is an ecosystem where data we have more data than pretty much any other sector. If you look I think something somebody said like a third of all of the data in the world is healthcare data and it's growing faster than anywhere else. I think it's easy to say let's open up all the data and all we can we can start solving more problems. And to a certain extent that's true. The trick though is often times it's not availability of data. It's sort of what we were just discussing. It's how is that data curated? How is it being brought together in a way that you can then train models specific to what it is you're trying to do. Putting more data into generalizable models probably not as helpful as us having very specific questions, the models to address those questions, and then curating the right sets of data in order to solve those. And in those instances, I think certainly we and others in the industry are open to bringing in proprietary data and combining that with academics, smaller companies, larger companies, etc.

Euan Ashley28:02

Great question there. We have another one, please.

Sua Cho28:04

Hi everyone. My name is Sua Cho, partner at SignalFire leading the AI and health and pharma tech practice. First of all, thank you for being here today. It's been really inspiring to hear about your leadership and what you're doing at your company. Dario and Anthropic projects that over the next 5 to 10 years we will see similar to 50 to 100 years of progress be made in this realm. I'm curious what you think will be possible perhaps in the next 5 years and goals that you all have set and metrics you guys are tracking to measure the progress that you are making towards that.

Christopher Boerner28:41

Look, I think anyone trying to predict where this technology is going to go and what it will do is going out on a limb or a plank depending on how you want to talk about it because I just think nobody knows. What we have said publicly on our last earnings call, you referenced it in the annual report, but look, we're going to cut down the time to do clinical trials by at least 30%. We're going to start discovering medicines that you wouldn't have been able to discover otherwise. Just the ability of this technology think about the most basic level. This technology enables us to interact with the science on a dimension that humans weren't evolved to ever do. So, where does that open up possibilities? It opens up possibilities on every piece of what we do. Everything from what diseases what even what the diseases are. So, think about cancer for example. Lung cancer when I started in the industry was two things. Non-squame, squame. Well, actually it goes small cell, non-small cell. Squame, non-squame. Now, you've got hundreds of subtypes. So, this technology enables us to understand what we thought was one thing is actually something else. And then start to target specific medicines to that. We now have the ability to go and say, we've got stuff that we discontinued that we can apply this technology to and now find that it actually works in one of those subtypes. I think the only caution I would say is when you hear somebody say we're going to cure cancer in our lifetime, you should be skeptical because cancer is not one thing as I just said. It's hundreds of things and it has the unfortunate quality that when you treat it, it evolves. And so, what we don't want to do is set this technology up for failure. And that I think we need to be cautious about.

Euan Ashley30:45

Thanks so much.

Ryan Harris30:46

Ryan Harris, I'm a sober healthcare private equity investor. So, that sort of betrays my question, which is first two parts. One is so, what has been the ROI on the tokens that you have spent? That's one. And two, if you're

Christopher Boerner31:00

I guarantee you I will not be telling you the answer to that.

Ryan Harris31:04

It may be directionally how I think about it.

Christopher Boerner31:06

Yeah, yeah. That's actually probably more important anyway.

Ryan Harris31:09

And second, you gave an example in India how you decreased the time and span for regulatory writing. I'm curious in that project and that work, did that lead to downstream consequences? How did you check for errors? Because when you go down from such a large period of time to a small period of time, how do you make sure you didn't have a downstream consequence?

Christopher Boerner31:26

Yeah. So both questions are important. So, the first question is look, the challenge of establishing an ROI in the space is profound. Because to get an ROI, return on investment, start with what am I investing? So, just that question is super hard to get a handle on. And that's why I referenced that I have a whole team of people now trying to track that question down. And we'll probably not even try to get at the stuff that people are doing that's below a certain threshold because it's just impossible to find it in an organization the size of mine. The second is how do you determine the return? So, I just referenced that we might have saved a program with the use of this technology. Well, what's that? How do you calculate that? So, I think just understanding how you track and capture an ROI is hard. But we've put a whole team of people on it in part because I have a whole bunch of investors who are asking me how I'm going to be calculating this, but I don't anybody has a great answer for it. And it's a super complex problem because the technology is moving so fast, we're investing so much and so quickly. You got to kind of rein it in in order to be able to show you what the investment is. And remember, we're still at the invest and hope stage in many respects. And so that I think is also we need to just give it a little bit more time. So that's a totally unsatisfactory answer, I know. On the second piece though, we did spend a lot of time. The regulatory documents that we send to the FDA are a crown jewel. You've spent billions of dollars to get to that point for a given asset. You don't want to screw it up in the last phase. What we did though is we've done it now enough to know that the models that we're applying work and work really well. There is embedded in this a key point though. This idea that this technology will replace all the humans is just I don't think fundamentally right for some of these bigger, more important steps for sure. You will always have people double and triple checking the quality of the work that comes out because it's just too important to not do it.

Euan Ashley33:52

All right, we have time for just one last question. You have the mic, so over there, sorry.

Audience Member33:59

Hi. I'm the founder of a healthcare startup called Dr. Ruby AI. And I want to know in your opinion as a company like us trying to collect the patient level of data outside of clinic, what would make the data more credible enough to make the pharmaceutical use or clinical use? Thank you.

Christopher Boerner34:24

Oh, wow. That's a tough one. Look, I think in some ways it's hard to answer that in a general way because it depends upon what specific data you're collecting and how you would plan on using it and how we would plan on using it. What I would say though is that the advice I would give you is partner early so that you can sort of help build it together because it's the type of data, it's the quality of data, and then it comes back to how you're going to curate it in order to get real benefit out of it. The best thing I can say is I don't have a general answer to that. It's more just be willing and able to partner early to co-develop. That's always a better way to go.

Euan Ashley35:11

All right, I'm afraid that is all we have time for. Please join me in thanking Chris Burnap for joining.