In this episode of Eureka!, McKinsey’s podcast on innovation in life sciences research and development, host Navraj Nagra and senior partner Dan Tinkoff talk with Greg Meyers, executive vice president and chief digital and technology officer at Bristol Myers Squibb (BMS) about the evolving role of technology in life sciences, the evolution of generative and agentic AI, and how BMS is rewiring itself with digital, data, and AI at the center. An edited version of their conversation follows.
From early fascination to leading digital transformation
Navraj Nagra: Let’s start with your journey. What brought you to your current role at BMS?
Greg Meyers: It goes back to when I was 12 years old. In 1982, I was the first kid on my block to get a personal computer. It was one of the TRS-80 Color Computers. I fell in love instantly, but it would be years before I really knew what I wanted to do. I eventually got a business degree and went back for an MBA; I had a double major in computer science, but I never thought I’d be in IT. It’s something that just pulled me in. I had thought of IT groups as kind of being a back-office function, and for the better part of the 1990s and 2000s, they were.
I saw IT as a noble profession, but it was focused on running the operations, and I had always aspired to help a company do something more meaningful. So I was lucky that the technology profession evolved such that it is no longer back-office, and I was really blessed to join BMS three and a half years ago. I lead a wonderful team and it’s been a lot of fun.
Navraj Nagra: When you stepped into the role, what was your 100-day plan? How were you thinking about your priorities and how have they evolved, especially given the significant shifts over the past couple of years?
Greg Meyers: All technology groups in Fortune 500 companies have gone through this metamorphosis. Technology has transformed into something that can help differentiate your products, your customer service, and your value proposition.
As we went through the outsourcing craze of the ’90s and early 2000s, many internal technology functions atrophied their technology muscles, because all you needed to do was to buy SAP, Workday, and Veeva, and you were good to go. With what’s happening with technology, you have to become builders again.
So the first thing we needed to do was put the ‘T’ back in IT. We didn’t want to be service brokers. We wanted to be creators. We knew that we needed to reduce our dependency on third-party suppliers and get more control around our technology footprint, and that required us to be more technical. Since then, we’ve more than doubled headcount, and that has really improved our ability to create new things.
Dan Tinkoff: Greg, you bring a real breadth of experience to this role. You’ve worked in agriculture, technology, and consulting on different continents. How does that diversity of experience shape your focus and the way you lead at BMS?
Greg Meyers: Different industries are not as different as they might seem. The nuances, the language, the vocabulary, and the acronyms are different, but I think every company is trying to find ways to get an edge with technology.
Increasingly, technology is imbued inside the core product or service. For every company that’s doing that, the issues are really the same. The key now is to put technology on the same footing as R&D.
In pharmaceuticals, computer science is really now the third leg of a science stool that also includes biology and chemistry. The three are increasingly converging to achieve new things. Every industry kind of has its own flavor of that, but the issues are pretty similar.
Organizing for speed, experimentation, and foresight
Navraj Nagra: How are you ensuring that you’re deploying some of that increased IT workforce to ensure that BMS remains at the forefront of emerging technologies?
Greg Meyers: In my 30 years of working, I don’t think I’ve ever seen things move and change faster than right now. Staying current on what’s really happening and being able to filter out the noise from the signal is probably one of the most difficult parts of the job. It’s not good enough to read the white papers and go to conferences. You have to experiment; you have to go and see for yourself. So we’ve tried to encourage the wider BMS organization, as well as the function that I lead, to not worry about whether an idea has merit based on a white paper, but to actually go and see if it does.
If we can experiment quickly and relatively cheaply, it is worth so much more than the time spent debating what we should be doing. I like to say that a demo is worth a thousand slides, and getting into action provides you with practical hands-on experience. You’re always looking for the gap between what you know to be possible and what you’re actually doing in the company. You need to have a really good footing on what’s actually going on in the outside world, and spending time thinking about that through experimentation is key.
Navraj Nagra: What is some of the impact you’ve seen through experimentation, particularly within your portfolio, either through in-house experiments or through partnerships?
Greg Meyers: Let’s talk about generative AI, since that’s the hot topic. ChatGPT came out at the end of November 2022. By January 2023, we had our first version of a retrieval-augmented generation chatbot up and running, trying to answer questions based on a private corpus of data. So we were really early in trying to figure out where there would be an edge. Fast-forward three years and now we’re working on solutions that, for example, help us automate large portions of clinical study reports. LLMs [large language models] are really great at reading and writing and editing and summarizing words on a page. So that early experimentation has paid a lot of dividends because we were able to get a head start on figuring out how to put the tool chain together, how to build tools that could create edges for us. Then, as models have evolved, we were able to plug them in and work from there. And I think that has probably affected everything—most parts of research and many parts of development as well.
The AI and digital health landscape: hype, realism, and the evolving game in biotech
Dan Tinkoff: You mentioned this incredible collision of biology, chemistry, and computer science. What inning of the game are we in in terms of using those to revolutionize R&D? How do you see this playing out, and when do we get to discernible impact versus hype or experimentation?
Greg Meyers: I view the current landscape as a lot of small tools that are all co-evolving and haven’t really come together.
Here’s why I’m encouraged. There are 40 trillion cells in the human body. Every one of those cells has about a trillion molecules in it. That’s about seven octillion atoms. Any one of those atoms out of place can cause, prevent, or cure a disease. And a lot of the challenges in drug development are largely computational problems. For example, how do you drug an undruggable target? Or how do you get the drug to the target when it’s hard to find? So there are different parts of research, such as protein-structure prediction or shifting from high-throughput screening to high-content screening, where these tools give you the ability to massively increase your productivity. But being able to run experiments at two orders of magnitude faster than you used to does not necessarily yield a working drug.
So this is all really so early. For example, the first generation of molecules co-developed by AI is just making its way into phase 1B development. So it hasn’t had a chance to play out yet. I think there’s too much hype on the idea that an agent running in the background can solve big problems in disease. There’s so much we don’t understand about biology. We know that AI is not good at extrapolating outside of data that has already been sampled, and there’s an awful lot about biology that hasn’t been sampled.
Individual tools like AlphaFold have huge promise. There’s a ton of stuff going on in protein-structured prediction. There’s also lots happening in transcriptomics, but these and other emerging disciplines haven’t coalesced into big drug breakthroughs yet. I’m pretty optimistic that as these mature, we’ll see breakthroughs.
Dan Tinkoff: We’re in the year of agents and agentic AI and the promise is great. As a biomedical knowledge worker, I can have an army of brilliant intern bots that can write my documents and run my experiments, albeit with occasional amnesia or poor judgment. How are you thinking about deploying and scaling agents in R&D, and how do you balance the opportunity with the risks and the challenges?
Greg Meyers: I’ll bring this back to something I’m close to: Code development is undoubtedly the farthest along the path of being agentic. There is a lot of agent-washing going on in the market. Right now, you have the ability to get an AI to perform a subsection of what you already do. We need to get it working end-to-end. Even the world’s best reasoning models are not great at taking that leap two or three levels beyond the core of what your research question is.
In pharmaceuticals, the exciting part of AI for me is hypothesis-free experimentation. That’s really the promise of agents, but I think we’re still quite far from it. Right now, our focus is on improving things we already do, such as trying to get the number of new molecular targets and shots on goal to increase, to extend our time on clinical trials. We’ve already cut the time it takes from deciding to optimize a lead to doing toxicology work by about 20 percent. This comes from thinking differently about data, AI, and workflow. In drug development, we’re on track to take two to three years off the average clinical trial for many of the same reasons. So whether the technology is agentic or not, we see a lot of opportunity and we expect that to continue.
Dan Tinkoff: As the chief digital and technology officer of BMS, what’s keeping you up at night?
Greg Meyers: There’s not a lot that keeps me up at night. One question that’s ever-present in a big company is, “Are you ahead of your competitors, or where are you relative to your competition?” I always think that in technology, particularly at the current pace of change, no lead is safe.
When you achieve major leaps inside the company, it’s easy to want to bask in that glory for a while, so it’s important to keep raising the bar higher and higher so that no one ever gets comfortable, regardless of how much progress we make.
I also think a lot about stuff that’s not technical in nature. A lot of times there’s a combinatorial effect between changing process and changing technology or using data, and there’s still a bias that AI is a magic bullet. You have to keep reminding people that applying technology to a broken business process just makes the broken process move faster. It’s important to remember the basics, such as ensuring your roles and responsibilities and incentive systems people are right, and then also making sure that you’re bringing technology to bear. Those things are all equally important.
Navraj Nagra: As you think about implementing AI into your workflows, how are you thinking about the responsible AI component?
Greg Meyers: When ChatGPT first came out, I saw articles reporting that other companies were blocking it, which showed a genuine nervousness about it. I’ll emphasize what I said earlier: There is no substitute for understanding risk better than real-world experimentation. Very early on, we required every employee to read and understand our AI policy. That kept us clear about expectations: keeping humans in the loop, verifying outputs for accuracy, and recognizing that people, not the system, remain accountable for decisions. We wanted everyone to understand that standard from the start. Since then, our focus has shifted to learning in practice: identifying the edge cases, seeing where these systems break down, and recognizing what happens if you give them too much control. For us, combining those real-world lessons with a baseline framework for responsible AI has been essential.
Dan Tinkoff: Let’s take a step back from AI for a second. The digital health area has also seen a lot of experimentation. How do you define digital health? Where do you see it bringing value to patients and what is its relative importance for a company like BMS?
Greg Meyers: We define digital health as trying to achieve a step change in the way patients are diagnosed, treated, and monitored. We see huge potential synergies in being able to unlock things like patient identification. The average lung-cancer patient, for example, will have less than five years to live upon diagnosis. Less than 75 percent of them will survive beyond that, and those who do have often gone through three or four failed lines of therapy before they find the right one. We still don’t know all the biomarkers that determine why one patient would respond to one treatment and not another. Our assertion is that there’s a ton of dark data sitting inside the body that can be measured, that can indicate how someone may respond or if they will develop an adverse event.
For example, through our partnerships, we’ve been able to build a tool in cardiology that can detect an overgrowth of muscle in the heart chamber. It’s called hypertrophic cardiomyopathy. The disease doesn’t affect a lot of people; it’s often the cause when a high school athlete falls over and dies during a football game. Usually by the time patients develop symptoms, things have become really bad for them.
It’s also quite tricky to diagnose, even for experienced cardiologists. It requires specialists that have seen this disease before to look at the echocardiogram. Well, we’ve worked to build an AI detector that can use a simple 12-lead ECG to find the signatures of this disease. People get 12-lead ECGs all the time, even for routine physicals. So your ability to identify the disease early and get treatment sooner is something we think can have a serious potential benefit for patients.
Similarly, the autoimmune response is one of the main barriers to using cell therapy to treat certain types of cancer. If you can provide an early warning system, you de-risk that therapy.
So that’s why we’re interested. We’ve got more than two dozen projects in development, which we bake into our clinical trials. You have to generate the real-world evidence. You need FDA approval and you need to publish in peer-reviewed publications. Doctors and patients alike can benefit from these tools as an adjunct to the core therapy itself.
Dan Tinkoff: It sounds like you feel some sense of responsibility not just for delivering the drug into the market, but also for impacting the surrounding treatment pathway so that patients get the full benefit.
Greg Meyers: I do. And I have a personal connection. My 11-year-old daughter had a lump that doctors had said was probably a cyst and no big deal. It was really only due to one doctor who had a hunch to test it, and it actually turned out to be cancerous. Now she’s fine, but that excision happened only because we relied on the experience, knowledge, and intuition of one doctor. And when you think about the heterogeneity of doctors’ practices, the ability to influence the tide that raises the boat of diagnostic accuracy for all doctors is a win-win for everybody.
The next S-curve, and learning how to learn
Navraj Nagra: What projects are you working on now that might deliver the next S-curve of impact?
Greg Meyers: One of our facilities in New Jersey is working on cryo-EM, which makes detailed 3D images of cells, viruses, and other structures by flash-freezing proteins using cryogenic electron microscopy. Normally, through X-ray crystallography, you’d have to destroy proteins by removing them from their natural lipid environments, but cryo-EM avoids that. Each image generates around 16 terabytes of data.
Another example is high-content screening, which allows us to run a thousand times the number of experiments we could do five years ago. Then there’s the ability to leverage the AI Center of Excellence we started in partnership with NVIDIA. These all live as separate things and are done by different departments in the organization. I think the S-curve is the ability to bring all these data-intensive initiatives together and leverage hypothesis-free neural network designs. I think that is going to be incredible. But right now, things are co-evolving as separate tools in the tool chain across science labs everywhere. Ultimately, I think they’ll come together, and we’ll see a big rise in overall improvement.
Dan Tinkoff: Technology’s moving at lightning speed. We hear you saying you need to be able to run at it, experiment, and adapt, but is a big corporation the best vehicle to do that? How do you teach a big corporation how to learn?
Greg Meyers: A big corporation is just people, just like a start-up is people. You have to create the structure that allows the right incentive systems to flourish. One of the things we’ve started to see is prototype purgatory. It’s easy to experiment with AI because it’s kind of harmless, but infusing it back into core processes is terrifying for a lot of companies.
We created an AI accelerator that takes groups of six to eight people, which is much smaller than you’d normally see at a big company, and gives them a dedicated focus, expertise, and seed money, as well as a ton of latitude to explore the space.
So many projects, particularly technology projects at big companies, follow a waterfall effect of spending considerable time scoping. They then do a statement of work and a request for proposals with third parties, and then they are locked into constraints of how long the project can be, how much it can cost, and who they can use as their partner. Instead, we have said that even failure is a success. If you think you can do something with AI that can make a 10X improvement somewhere, go and see as fast as possible.
By doing that, you make it safe for people to experiment. Once a big project is approved, it can’t fail because someone’s reputation is on the line. There’s shame and blame and embarrassment, and this is how you get these big technology projects at big companies where they become like Frankenstein’s monster, neither alive nor dead.
By doing it this way, in six two-week sprints, you get to a point where the thesis to use AI is either valid or not. And if it’s valid, then you can go into the scaling phase. We had to create a separate structure than what we were used to doing. This is new, but it has proven pretty effective, and I think now we’re moving at start-up speed.
Navraj Nagra: It takes longer than ever to get a drug onto the market. We’re spending billions more than we ever used to to launch medicines, and our failure rates are higher than they’ve ever been. What is your vision for the future, for the broader industry and for BMS?
Greg Meyers: When talking about the infamous Eroom curve—that is, a reversal of Moore’s law, where drug development costs are increasing even as technology costs decrease—you have to consider the fact that a lot of the low-hanging fruit in drug development has been picked. Look at cancer; in the past five years alone, we’ve had cell therapy and we’re now heading toward cell therapy for solid tumors. There are antibody drug conjugates, bispecifics, trispecifics, and radiopharmaceuticals. So we are starting to get into much more difficult territory, and there’s probably never been a better time to look for some edge. Three major steps of the scientific method—observing, questioning, and forming a hypothesis—are essentially following your gut. But that’s starting to have diminishing returns, which is bearing out in the data. BMS focuses on oncology, hematology, immunology, cardiology, and neurology, and we expect we’re going to be one of the fastest growing companies at the end of the decade. But it is going to mean getting involved in all these new modalities, and knowing what to use and what will ultimately make a difference.
In oncology, for example, knowing whether this tumor has this antibody or antigen on it or not, getting a richer picture of the functionality of the tumor and realizing that cancer is not one thing—it could be 14 or 15 things—opens up the ability to be more selective about how to address the targets in development and how to treat therapeutically in the clinic. Technology and AI could not be arriving at a better time. We need new methods to tackle some of these trickier modalities.
