Rewriting the Economics of Drug Discovery with Ben Taylor

In many ways, the human body is an undiscovered country. 

Recursion [RXRX] CFO Ben Taylor points to the limited advances made in pharmaceutical research thus far: “If you take the entire pharmaceutical and biotech industry for all of the time that they’ve operated, all of the hundreds of billions of dollars that have been invested, all the good people that have been working on it and all of the drugs that are currently approved or in clinical trials, you only cover about 10-12% of the entire genome… the rest of that biology is basically unexplored from a therapeutic perspective.”

But this lack of knowledge, in turn, represents an opportunity – one that firms such as Recursion are hoping to unlock with artificial intelligence (AI). This represents a new model of drug discovery, leveraging AI to discover and test novel drugs with the aim of shortening the pipeline from discovery to regulatory approval and treatment, all while improving patient outcomes. 

Taylor has worked in healthcare investment banking for more than two decades, working with global names such as Goldman Sachs [GS] and Barclays [BCS]. Prior to his current role, Taylor served as CFO at Exscientia, an AI-driven drug discovery peer acquired by Recursion in August 2024. He also served in executive roles at Aetion, a data-driven healthtech company, and Tyme Technologies, an oncology-focused biotech firm. 

In the most recent episode of CMC Aureon Sessions, Taylor sits down to discuss the massive potential of AI-driven drug discovery, what being AI-native means in the biotech space, and why Recursion’s team and partnerships are just as integral as the technology it uses. 

AI-powered acceleration

While Taylor notes that the traditional drug discovery process is largely manual and time-consuming, he likens the current developments in the sector not to industrialisation, but to the development of computers. In terms of the scale and complexity of the problems being tackled, “we’re going from people putting together vacuum tubes to how things are microfabricated.”

In fact, the level of complexity inherent in biology or chemistry research makes the fields uniquely suited for AI integration. “These are problems that we can’t really comprehend, even with traditional computing, so this is where AI is opening up completely new ways of tackling problems that have been plaguing us for decades.”

While the question of whether or not to use AI is a matter of controversy in a number of other sectors, in biotech it’s a no-brainer, Taylor says. “Everyone should be using AI. It’s just a better tool.”

The way companies deploy AI, however, is a different matter. “How you use AI, the depth that you’re able to use it, how you’re able to problem solve all of it – that is a very wide variance. If you look at the large pharma companies, they’re really using it for workflows or early research.” 

For an AI-native company like Recursion, however, the technology is central to the process of drug development and discovery. As Taylor notes, the first step in any process requires answering one key question: “What is the AI way that we could solve this problem?”

Despite the tumult caused by agents and generative AI in other sectors, the focus in biotech is largely machine learning, modelling and other AI-assisted ways to deal with large data sets, Taylor says. “We’re not doing sort of quick consumer things [like chatbots].”

Taylor argues that these tools help shorten the road to commercialisation for the sector at large. “A lot of the areas in tech where you are going from inception to commercialisation you have a two-year, a three-year time horizon,” Taylor notes. “Currently, in the biotech industry, that’s like 10 to 15 years.”

Using AI, Recursion has managed to shorten the drug development timeline considerably. “We’ve been able to take what’s normally about a four- to five-year process and bring it down to 17 months.”

That then has a knock-on effect, reducing the number of molecules tested and the resulting cost. On the clinical side, a better understanding of patient populations helps drive faster enrolment and expand the pool of eligible patients.

This, in turn, speeds up the clinical trial, which Taylor notes is the biggest expense in pharmaceutical R&D. “We’ve been able to see a 30-60% improvement in the enrolment speed, directly affecting the time and cost of the trial.”

The impact of these improvements is evident sector-wide. As seen in the graph below, data from McKinsey & Co shows AI-driven site selection reducing trial duration 16-39% across treatment areas such as Alzheimer’s, respiratory conditions and oncology.

Ultimately, more efficient tools allow companies to explore more solutions. “Being able to create and analyse data in a new way is how you get to a new understanding of diseases and patients. That’s going to be a massive long-term unlock. We’re just putting our toe in the water of what we could be doing.”

Key partners

Turning that promise into approved treatments, however, depends as much on how Recursion structures its business and partnerships as it does on the technology itself. 

Prior to merging in 2024, Exscientia and Recursion were AI-native peers in the drug development space. While the former focused on chemistry and the latter biology, Taylor explains, “both companies had developed their own internal pipelines, partnerships and a fairly large platform. By bringing them together, we created the ability not only to find novel targets, but also to create the drugs that could be used to hit those targets.”

That approach shapes how Taylor frames Recursion’s competitive position. He does not see the firm as competing with other AI-native biotech firms, but rather as blazing an alternative path to traditional drug development. “Our competitors are not other people using AI. It’s the traditional way of doing drug discovery and development, which is where almost all of the resources in the industry go right now.”

The merger helped underscore a fundamental part of Recursion’s business model by adding French pharmaceutical giant Sanofi [SNY] as a major partner. Preexisting partnerships with Roche [RHHBY] and Bayer [BAYRY] further support its business model, which leverages the financial support of larger partners and the expertise of Recursion’s staff to guide the drug development process.  

Recursion’s income stems primarily from milestones with its partners. Taylor underlined seven milestones achieved with Sanofi and Roche totalling more than $500m. “For Sanofi alone, we can do up to 15 programmes, and each one of those programmes has the potential for $343m in milestones and low double-digit royalties.”

The way Recursion structures its agreements helps ensure that these milestone payments transform directly into profit. “During the first couple years, we’re not spending any money out of our own pocket. We get paid in advance for direct costs and, within the first two to three years, we try and hit milestones that will start to drive profitability into it. By about the third year of a given project, we hope to be past our operating obligations. Then, once we reach something called development candidate, all of a sudden that payment would be all profit to us.”

While the company has yet to commercialise a drug, Taylor points to what he calls “green shoots” – opportunities for growth. One recent key win was Recursion’s REC-4881 drug – used to treat familial adenomatous polyposis, which causes potentially cancerous polyps in the digestive tract – reaching Phase 2 trials. 

While their partners might provide the resources and the direction, Recursion guides many of its collaborations in-house: “We agree to what the goals of the project would be, and then we do all of the work ourselves internally.”

To decide what projects to focus on, Taylor explains, Recursion looks at two key factors. 

First, “what do you have good data for?” Over time, data collection in a lab setting leads naturally to the questions that can be answered using that data. “You start to learn the language of the data,” he explains. “Then you can get much more intelligent and targeted with what data you create and what questions you ask.”

The second is expertise, both from Recursion and from its partners. “With Roche or Sanofi, we’re going to work with them very closely to agree to a target … what our partners really bring is that expertise.”

In the long term, Taylor sees therapeutics remaining the company’s primary focus. “It’s just how we will be traded and grow as a company, but the platform or the partnership side of it definitely balances us out.”

Biotech’s AI advantage

Despite the technologies it uses, Recursion’s aim is the same as any major pharmaceutical company, Taylor explains. “The goal is to get approved medicines over the line and really start treating more patients with them.”

However, as a “multi-parameter problem”, Taylor explains, biotech is ideally suited for AI. “Right now, there’s about a 95% failure rate in the drug discovery and development industry … [but] those problems aren’t all coming from one thing.”

Taylor outlines two general types of problems. The first involves teasing actionable information from massive troves of data – usually to solve problems of biology. Recursion has over 40 petabytes of internal data from their lab work, supplemented by an additional 25 petabytes supplied by partners.

Here, being AI-native helps “take a lot of the noise out of the system”, he says. He cites an example of analysing transcriptomic data – data related to the RNA in a cell or tissue sample. “We were able to much better predict [outcomes] using far smaller data sets because the data quality that we had is so high. Everything that we’ve done since inception has been created for machine learning and for integration into our system.”

The second problem involves a scarcity of data. “With a lot of chemistry problems, the data is not going to be in existence, and the number of potential solutions are effectively infinite.”

In that case, Recursion combines predictive and generative models to help create and test virtual molecules. 

A key tool in the process is BioHive-2, a supercomputer built in partnership with Nvidia [NVDA]. 

Recursion has implemented AI agents for workflow optimisation. That said, Taylor underlines the importance of a strong foundation for those models. “You actually lose fidelity by adding an agent to a system that doesn’t work,” he explains. 

However, any AI system Recursion does use is focused on enhancing the human talent the company already has. 

“Underlying it all is the people,” Taylor says. “It has taken more than a decade to put together the right group of people because you really want them to understand medicine and sciences as well as coding. 

The focus has been on multidisciplinary talent. “Those are going to be the people who can take AI and really maximise its uses. They’re not being replaced by it – they’re being empowered by it.”