Lesson 08

Novartis

Build in increments that are individually valuable & collectively transformative.
background placeholder

AI’s benefits don’t come much bigger than the potential to deliver new medicines that let people live longer, healthier lives. But bringing a drug to market involves so much more than just finding the right compound. Novartis are applying AI along the whole continuum of their work, to improve decision-making and get medicines to patients faster. Each step makes a difference. Collectively, they’re transformative.

󠀠

󠀠
Look at the pill in the palm of your hand.

󠀠

It doesn’t seem like much. In colour, texture and (possibly) taste, it’s like a tiny nub of chalk - and it looks about as complex. Compared to the smartphone in your pocket, with its nano-scale microcircuitry and blazing screen and global connectivity, the pill’s just an inert lump. The phone has AI; the pill’s just… artificial. You don’t give it any more thought than you do the glass of water you wash it down with as you pop it in your mouth.

󠀠

You’re missing what’s really happening.

󠀠

The medication that you just took is the result of a decade-long process that began with a molecule in a research lab and ended just now when it hit your bloodstream. Just as much as your phone, the pill is a miracle of world-class minds, cutting edge research, advanced manufacturing, global supply chains and an obsessive commitment to quality and safety.

Most of all, that pill is the sum of thousands and thousands of individual decisions that have guided its journey from the moment a scientist conceived the idea, through unimaginable layers of tests, trials, protocols and approvals, to the moment you popped it out of its blister pack. 

Now the company that made that pill, Novartis, wants to use AI to enhance every stage of the journey. Same rigorous science, same uncompromising focus on quality, just better decision-making. And faster.

New arts


The name Novartis was coined in 1996 when the Swiss multinational pharmaceuticals company took on its current form. The name came from the Latin novae artes, meaning ‘new skills’. It’s an apt name for a company dedicated to producing truly innovative medicines, and it’s taken on an extra dimension now as Novartis looks to develop even newer skills in AI.

Overseeing the clinical development process is Shreeram Aradhye, Novartis’s President, Development and Chief Medical Officer. A trained physician, he is absolutely the sort of doctor you would want treating you if you were sick, bubbling with good humour and enthusiasm for his subject. ‘I’m excited to have this conversation,’ he says, when asked to discuss how AI is being embedded in the development of new medicines. ‘And what I’m most excited about is not the technical part of AI, it’s the human part of the engagement with this technology.’

Shreeram describes the company’s purpose as ‘turning molecules into medicines and getting them to patients.’ It’s a neat little summary for a process that is one of the most complex on the planet. From the point that a promising compound is identified, it takes on average a decade to get it to market. Development can cost hundreds of millions of dollars, sometimes even billions. And all that money only buys you a one-in-ten chance of success: 90% of drugs that enter development fail.

󠀠
󠀠

Novartis has already embraced AI in the earliest stages of that process through its strategic collaboration with Isomorphic Labs, the London company that uses AI for drug discovery. Established by Demis Hassabis, who also founded the pioneer AI research lab DeepMind, the company proved its bonafides in 2024 when Hassabis won the Nobel Prize for Chemistry for his work on using AI to predict the structures of proteins. The collaboration with Novartis has the potential to help find promising targets for drug development in ways that couldn’t be imagined before.

󠀠

But because of the timescales, the targets Novartis or Isomorphic identify now will take a while to reach patients. Much quicker to be felt will be the ways AI impacts Novartis’s productivity in drug development: essential aspects of the process like new trial protocols, regulatory submissions, and managing large patient data sets. The more immediate gains, in other words, aren’t at the wild frontiers of drug discovery. They’re in improving the processes that get a candidate product from the lab to the market.

󠀠

That’s why Shreeram’s excited.

Transformative innovation


One of Novartis’s key corporate tenets is the idea of ‘transformative innovation’. The company doesn’t aim for small, incremental improvements; they focus on dramatic advances that can significantly improve patients’ quality of life by curing, treating, or even preventing diseases in genuinely novel ways. 

But how do you transform a system as complicated as drug development? Complex systems are, by their nature, hard to change. It requires delicate judgement to decide how much to bite off at once. You can go big, try to imagine the future and leap towards it in one gigantic bound - but the track record of this kind of big bang approach is, to put it mildly, mixed. Monolithic technology programmes almost never work inside a complex system.

Small changes - the ‘easy wins’ and ‘low-hanging fruit’ - have a higher chance of success, at least locally where they’re deployed. But they probably won’t make a measurable difference to the overall performance of the organisation. A lot of companies’ AI investments fall into this category, discrete initiatives with no mechanism to join them up, that only reinforce siloed decision-making.

Testimonials

"We spent a year systematically looking at how AI can enable our clinical trials. We can shift the performance of multiple individuals closer to the performance of the best, and I’m convinced that’s where the real value is going to come."

Shreeram Aradhye
President, Development and Chief Medical Officer, Novartis

Shreeram’s having none of that. ‘We replaced the word “divisions” with “continuum”,’ he says. ‘Commercial organisations all have to work together along this continuum, and all of them can be AI-enabled to be more efficient at the individual level, and at a team level.’

The best way to use AI to change a complex system, as Shreeram recognised, is to break down that continuum into modular components that are valuable in their own terms, but which connect together into something much more transformative. And the key challenge here is to decide what form those individual components or increments should take. Particularly with AI, it’s easy to be misled by the outputs that the models create. AI specialises in forecasts, clusters and classifications of data. It summarises documents and generates new ones. All these capabilities are impressive but none of them, in their raw form, are high-value business outputs.

󠀠
󠀠

The most successful applications of AI in an organisation tend to be those where the software’s outputs are carefully shaped as inputs to high-value decision-making, supported by intelligent technology. So when you’re looking at how to break down the business continuum into meaningful constituent parts that can be enhanced by AI, the right increment is the decision. Specifically, the decisions that create (or indeed destroy) the most value. Then you can use the technology to improve the speed, quality and accuracy of the decision-making processes.

󠀠

It’s easy to think of this as an arithmetical process: that the output is the sum of all the decisions made. But the right calculation is actually combinatorial. The decisions don’t operate independently. They depend on one another, along the whole continuum, up and down the system. The outputs of one often comprise the inputs to the next, and you can’t understand the consequences of one decision without knowing what a set of related decisions were.

󠀠

When it comes to Novartis, few parts of the business have as many complex, interlinked decisions as the process of testing a new medicine.

But can you prove it?


‘A medicine,’ says Shreeram, with his knack for capturing complex processes in quotable one-liners, ‘is simply a molecule plus evidence.’ Making sure that evidence gets gathered through a series of ever-larger and more ambitious trials is one of his core responsibilities. But as these trials can last months or even years, there’s plenty of time for things to go awry. 

‘A medical trial isn’t like launching a rocket and letting it go,’ Shreeram says. ‘It’s the equivalent of a ship that has to navigate across the complexity of the ocean.’ And there’s not much room for error. A molecule might be effective, but if the trial isn’t designed and managed well then the data might not show it. So Novartis turned to AI to support the decisions that help them chart a course through that complexity.

Trials that used AI in the design process have been more likely to finish ahead of schedule. Sites that AI identifies as being suitable for a particular trial have strong potential to recruit patients faster than ones that weren’t selected by the AI, and produce the diversity of patient population that the experimental protocol demands.

󠀠

‘We spent a year systematically looking at how AI can enable our clinical trials,’ says Shreeram. It starts with the scientific protocols that underpin any trial. These are a sequence of decisions that shape everything that follows. What primary and secondary goals should we set for the study? What type of experimental design should we use, and how should we collect the data? What should be the eligibility criteria for participants? What dosage, frequency and duration of treatment should we specify? How often should we bring patients in for assessment?

Shreeram describes a tool Novartis developed called Protocol AI. ‘It was our effort to say: “Can we enable the person who writes the protocol? Can they be augmented by an AI tool that gives them access to knowledge and information about comparable trials that have been designed for the same medical condition? Can they learn from the previous performance of similar trials to understand the impact of different design features, and what their future implications might be?”’

󠀠
󠀠

Once the protocols are established, the hard work of recruiting participants begins. This, too, is a decision-intensive process. ‘We have hundreds of sites across multiple countries,’ says Shreeram. ‘And there used to be a very complicated process where we had to send the request to the countries, find out if it was feasible, and it took weeks to go back and forth.’

󠀠

In choosing which sites to use, investigators have to consider whether those sites can provide enough participants for the trial, whether they’ll be able to meet the protocol requirements, even whether other pharmaceutical companies will be competing for the same patient populations. More and more decisions, each intersecting with the others.

󠀠

And when the trial is over, the results have to be written up for submission to regulators. Again, Novartis are deploying AI to improve the process. ‘We use GenAI to generate first drafts of reports so the medical writers can focus on the actual interpretation of the content to position it in the best and most accurate way possible,’ says Shreeram.

Innovate differently


Each of these is a compelling use-case, and they’re already starting to show results. 

Trials that used AI in the design process have been more likely to finish ahead of schedule. Sites that AI identifies as being suitable for a particular trial have strong potential to recruit patients faster than ones that weren’t selected by the AI, and produce the diversity of patient population that the experimental protocol demands. Those sites also get approved faster. ‘What used to take weeks now… doesn’t,’ Shreeram marvels.

Incrementally, each of these is making a difference. But to fully achieve the kind of transformative innovation that Novartis aspires to, the company is working with Faculty to tie it all together with an Intelligent Decision System called Frontier. This system uses a ‘computational twin’, a sophisticated digital simulation that connects data sources, operational processes and machine learning models together in an interactive, virtual replica of the clinical trial process.

In this controlled environment, Novartis can experiment with different ways of linking up its various AI initiatives to make sure they are all working in synergy to deliver the greatest benefits. This means systematically deconstructing the decision-making process to see where the key decision points are, what data will be required to inform them, and where future AI investments will have most impact in enabling those decisions.

Ultimately, it will let Novartis connect decisions across different functions (such as clinical, operational, supply, strategy, regulatory) and at different levels (for example by patient, site, cohort, medical condition, program, or the whole portfolio of trials). It will provide a scalable framework that allows future investments in data and AI to connect seamlessly to what’s already in place.

󠀠
󠀠

‘The focus now will be on ensuring that we scale the few things that we know definitely work, but not necessarily spend as much energy in coming up with hundreds of more new things,’ says Shreeram, reaching for a botanical metaphor. ‘We're shifting from a meadow of wildflowers of AI innovation, to more of a curated garden.’

󠀠

John Gibson, Faculty’s Chief Commercial Officer, has a different analogy. ‘We think of it as a keyhole surgery approach. No bottom-up, multi-year data transformations. No baskets of AI use cases looking for a business user. It’s a pragmatic approach that drives AI-enabled change in a sustainable way.’

󠀠

‘The goal is not to fail fast,’ says Shreeram. ‘The goal is to succeed as quickly as we can.’ When an organisation augments decisions with well scoped and safe AI technology, it’s able to raise the average distribution of decision-making quality across complicated processes. ‘We can shift the performance of multiple individuals closer to the performance of the best, and I’m convinced that’s where the real value is going to come.’

And in all this, Shreeram never loses sight of the bottom line. ‘I'm not focusing on efficiency. I'm talking about value, and I always say the value is not just the amount of money saved. It's value to patients, too, so that they can benefit from earlier treatment and faster access to new medicines.’

Novartis have already started accelerating that vital outcome with AI-enabled processes. Connected together in the service of human decision-makers, they will transform it.

The lesson in summary
Build in increments that are individually valuable & collectively transformative.
  • Unambitious AI programmes usually underwhelm. Don’t just focus on low-hanging fruit. Or on single use cases. That’s not how you make a measurable difference to the overall performance of an organisation.
  • Over-ambitious AI programmes usually underdeliver. It is very risky to try and imagine the whole future at once and seek a big bang. Monolithic technology programmes almost never work.
  • The right balance is to think big, but build forwards in modular steps. Each individual module should be quick to implement and valuable on its own terms. But modules should also be designed so that when connected together, they transform a whole process end to end.
  • The operational decisions that determine how a business process runs make good targets for individual modules. Improvements to the speed, quality and execution of decision-making are one of the most reliable ways in which AI can improve the overall performance of a business process.
  • Connecting these individual decisions together allows them to break out of organisational silos and better account for upstream and downstream interactions. This shifts focus from what is best for each local part of the process to what is best overall.
background placeholder

Did you enjoy this story? There are nine others just like it, told from the perspectives of nine of our other inspiring customers, in the full book 'Ten Lessons From Ten Years of Applied AI'. Just leave your details below to get instant access to your copy of the book.

󠀠

󠀠

󠀠󠀠With contributions from:

Ten Lessons From Ten Years of Applied AI

Download the eBook

Get instant access to ten examples of AI solving the world's biggest challenges, told through the stories of ten of our most brilliant customers.