Please note transcript has been edited for brevity and clarity.
FLG: Hi everybody. Today, I'm here with Stef James, Head of Operational Data Strategy at AstraZeneca. Stef will be speaking at The Festival of Genomics, Biodata and AI in Boston this summer, where she'll be leading a roundtable discussion on AI-ready teams. Stef, thank you so much for joining me today. How are you doing?
Stef: Yeah, really good. I'm really excited to get started. I'm interested in AI and everything that it brings in the conversation. So, yeah, I’m excited!
FLG: I think the best place to start is to tell our audience a little bit about yourself.
Stef: I’m Stef James, I head up a team called Operational Data Strategy, a relatively new team in AstraZeneca. It was created in June 2024. Organizationally, we sit within research and development, within clinical operations. So really, the engine of how clinical trials operate. The types of capabilities within my team are business analysis, managing data products from source systems, and being able to leverage those data products through the use of data science visualizations, that type of thing. I'm interested in KPIs and metrics, from the clinical study protocol to the clinical study report. That's the remit of my team, and we're always looking to advance and integrate AI into everything that we do, so that it becomes not a bolt on, but actually embedded into the way that we work. So that's just a little bit about the team. I've been in AstraZeneca for six years in varying different roles; I actually started out in more of a privacy/legal role, but now I’m definitely more in the space of data analysis, data utility, and maximizing the value from data.
FLG: AI is really transforming the way we do things, it seems in every sector. But what feels genuinely different about this compared to previous tech advances, especially in the pharma space?
Stef: It's a really good comment. I think in pharmaceuticals, we have a vast amount of data, both internal data and external data, that we acquire. I’m thinking about clinical data, but also literature, journal articles and real-world evidence as well. And unlike previous technology advancements, like shifting to electronic data capture records and things like that, we’ve been able to synthesize that vast amount of data in a structured format very quickly using AI.
The other piece, and it's something my team are heavily involved in, is instead of looking at clinical trial performance in the past, and having that rear view mirror approach, it's more around the predictive modelling. So, actually predicting what the dropout risk of patients could be, or what site performance could be before the trial even starts. I think that's really the differentiator within the pharmaceutical industry.
Ultimately, what we're focused on in pharma is to ensure that we keep the patient at the center. So, the direct line to the patient, making sure that it's not just like an IT upgrade, but it's actually going to accelerate therapies for patients in a way that can really optimize how they live their lives. I think that's why it's so different. It's just such a critical sector to be involved in, and I think AI is accelerating everything that we do in that patient-orientated space.
FLG: And from your perspective, what's the biggest gap between AI ambitions in pharma and the actual reality of how it's being used today?
Stef: This is something we talk about all the time. It's using AI for the sake of it. It's in every single headline at the minute. It's the piece of information that everybody wants to talk about. But it's not just deploying it to say that you've used AI; I think we need a clear operational use case to ensure that it actually makes a difference, and you don't place more burden on the process or the people that are interacting with AI.
I lead a data team. And the reality of this is that data is often siloed, and really messy and complex in an organization the size of AstraZeneca. Actually bridging the gap and ensuring that data is truly AI ready is one of our biggest hurdles. Then being able to seamlessly integrate any AI advancements within our processes and ways of working, and into the daily workflows of clinical scientists and site coordinators without adding that burden. So, [the gap] is the AI for the sake of it, data readiness, and how we actually embed this into a new way of working without adding additional burden.
FLG: I mentioned you're going to be leading this roundtable at The Festival, and it's focused on what AI-ready data teams look like. What does being AI-ready mean to you?
Stef: It's not going to be any surprise that I'm going to say the foundational data strategy. It needs to be robust, it needs to be interoperable, and we need to understand where we can access data that's clean, accessible, and most importantly governed, because we can't build AI on a shaky foundation. That's the crux of AI readiness. On top of that, it's then being nimble and agile enough, but with the right guardrails, so that we've got the patients’ trust in everything that we do. We need to ensure that we use AI ethically, that we adhere to regulatory and privacy frameworks from day one. It’s just absolutely not negotiable.
But the other piece of AI readiness that I think is often missing is the cultural readiness. AI has advanced so much, and I think everybody is actually a little bit behind the curve in understanding the true impact that AI can have in our processes. A lot of that comes from data literacy and being open to using AI insights, whilst also keeping the patient at the center of the process. So, I think there's a cultural uplift that needs to happen as well to truly be AI-ready.
FLG: What does a well-functioning, cross-functional AI team look like in pharma?
Stef: Really surprising profiles! We need massively diverse expertise; blending data teams into the business and, in my space, putting clinical operations experts, regulatory specialists and IT data scientists together. Data scientists can't work in a vacuum, and the context is truly key for that to really work and to understand what a specific data point means for patient participating in a clinical trial. So if we all share that very singular goal, around this patient centric view of the world, then I think that team needs to be truly cross-functional, truly diverse, but aimed at that mission of ensuring that the AI initiatives actually translate to better, faster and safer clinical trials for patients. So, going back to not using AI for the sake of it, but where you can actually see the outcome of where it's going to help the patient.
FLG: In your experience, what factors most influence trust or even contribute to scepticism among scientists when AI tools are being introduced into workflows?
Stef: I call this the black box of AI - not being able to understand the why. Having that explainability and transparency is really key for scientists to trust AI and understand that it's been trained on the right dataset set and that you've got validated evidence that the training has been done appropriately.
There is a piece here around replacement versus augmentation as well, and that drives scepticism. I think having AI positioned as something that works with you, and that takes the administrative burden away from the tasks so you can actually focus better, in my space, on clinical judgement, rather than on administrative tasks that can be really time consuming as well. Ultimately, though, it's reliability. So, if an AI tool hallucinates, or if you get inaccurate insights earlier on in the process, the trust is instantly lost. So, it's being able to rigorously validate that on real life scenarios, and embed scientists into how that operates and works, so they see the advancements early on as well.
FLG: And from what you've seen, what are the most effective ways that organizations are encouraging adoption rather than resistance?
Stef: It's actually a little bit about what I've just touched on. It's co-creation, involving end users - the clinical scientists, the study managers - in those design and testing phases. Because if you build AI algorithms and agentic AI and agents, you build it with their feedback. If they say it's okay, it's valuable, then they'll start championing it and become user champions. And it’s about ensuring that we are actually solving a problem, the real pain points, and being able to ensure that we've done effective training. So whenever AI has been released, it's not being released into the ether, just hoping that people adopt it. We really need to think about that uptick. What does a new training programme look like when you're implementing AIl, and how does that learning loop happen over time? I think those are probably some of the key pieces around avoiding resistance, it’s trying to get ahead of it in the first place.
FLG: And I do have a follow up to that. When you talk about co-creation, I imagine that on both sides - be it from a scientist perspective or a data scientist perspective - there must be a lot of learning that needs to be done. Do you find that people are quite willing to get stuck in and do that learning?
Stef: Yeah, I think especially in this AI space, I think people are so eager to get involved. And what I worry about is that they get disheartened too quickly, because it’s rapidly evolving all the time. It does take time and feedback to ensure that the models advance in the right way. But we have a sprint way of working. We have two week sprints. We do really iterative demos, and I think involvement at that early stage really helps. It really, really does help combating any resistance in the first place.
FLG: Do you think that this sort of cultural shift is perhaps bigger than the technical shift that's needed?
Stef: 100%, absolutely. The technology is advancing way faster than anyone is able to absorb it, even if you're working in the AI space. New versions of models come out all the time. So the technical shift isn't just in how we engineer ourselves and our processes, but the cultural shift and human behavior, it's so deeply ingrained in how we work and operate. So there's a shift in mindset as to how to even approach this, and that kind of agile experimentation as well, which can be really challenging in pharma! Because in pharma, we've got such strict - for the right reasons - safety and regulatory standards. Mixing that with agile experimentation, it's definitely a new way of working, and it's going to take time to ensure that we get that balance right.
And then I think there's this leap of faith. Do you trust a human 100% to deliver their deliverables in a workflow? And is that the same bar of entry for an AI agent? Should the bar of entry be 100%? I don't know. It's understanding that as well, and it's going to evolve so much more. But it's interesting, and it's definitely about the people more than the technology.
FLG: That's a really interesting point about trust in humans. Because I think in this conversation, that point is definitely being overlooked. With every new technology shift or anything that's developed, that conversation must have been had. So ,it's definitely interesting to look at it from that perspective.
Looking ahead five years, 10 years, what skills or roles do you think will be essential in AI-enabled pharma settings?
Stef: This is the most difficult question. I think it's that bilingual compatibility, people who can speak both in the clinical setting and the data science setting, a translator between the two that can bridge what is actually operationally realistic versus what the technology is and the feasibility of that. Usually you have an IT organization and the clinical operations, and in that kind of space, some of those roles may begin to merge a little bit more. I think in AI ethics and governance, I can see how it's already a space that is starting to become really, really well structured. I think as we get more regulations issued, that's going to become key over the next five years. Then it's more around the strategy, so AI strategists or similar in a more operational setting, understanding how you can best deploy predictive analytics, large language models, to actually optimize the trial design, the site selection, patient recruitment. Being able to identify those opportunities and shape a strategy around it, that'll be really key as well.
FLG: At your roundtable discussion at The Festival, what sort of challenges are you hoping that those discussions can solve, and what insights do you hope that people will leave with?
Stef: I think everybody already knows that AI is a game changer. If you're not using it in your professional life, you're probably definitely using it in your personal life. I want to leave a message that it needs to be patient centric, and we can never forget the patient at the end of that data point and have it purely a tech and data conversation. I think the technology needs to serve the patient, not the other way around. So that's one piece.
The second piece is around fundamentals. So somewhat moving past hype, but also focusing on some of the more unglamorous pieces of the work, which is around the data strategy, data quality, how do you integrate it? I often think the actual building of the AI, everybody's really enthused by that. Whenever you start talking about the foundations in the data, though, people get a bit bored of that.
And then there's a call to action to the pharmaceutical industry. I think it's collaboration over competition in this space. Solving the AI puzzle in pharma requires industry wide collaboration on standards, on guardrails that we need to adopt, and the best practices to truly accelerate R&D for everyone. I think we'll start to see that in industry as well. We're already seeing it with industry sharing groups, etc., but I think that's something I'd love people to take away from the round table.
FLG: Well I think it sounds like it's going to be a really exciting discussion. Thank you so much for sitting down to talk to me today and to everybody listening, if you're interested in having a conversation about what AI-ready teams look like, come along to The Festival in June and join Stef’s roundtable. Thank you very much.
Stef: Awesome. Lyndsey, thanks for having me.
)
)
)