The growing role of AI in pharma
While there is significant excitement around the potential of AI to transform the drug discovery and development ‘funnel’, as illustrated below (figure.1), this transformation is not one that will happen overnight. The biopharma industry is slow to adapt, and the technology is not yet proven at scale. As one leading industry expert puts it - “[we’re] in the third to fourth inning of a nine-inning game”.
Nevertheless, Third Bridge experts say the industry has transitioned from a data-aggregation and testing phase to an execution period. As AI is increasingly deployed, the common refrain becomes:
- Can AI turbocharge drug discovery efforts?
- Can AI lower the cost of clinical trials and simultaneously increase the probability of success?
- What does biopharma’s adoption of AI mean for third-party service providers such as CROs (clinical research organisations)?
Figure.1 The traditional R&D funnel vs the AI-enabled “Ideal” pharma pipeline.
Source: Recursion Pharmaceuticals' S-1 SEC Filing
Upstream drug discovery: How is productivity being redefined?
One of AI's most mature applications is in generative chemistry - allowing researchers to optimize molecules for specific properties (such as binding tightly to a disease-causing protein) before any physical lab work begins. In some instances, algorithms are also being used to design entirely new molecules from scratch. However, as illustrated below, this only represents one of the multiple steps involved in drug discovery.
Figure.2 Applications of AI across the drug discovery process.
Source: Drug Discovery World
Generative Chemistry can take many different forms, although there are two predominant company archetypes: Pure-Play Computational and Hybrid Tech-Bio.
Pure-Play Computation players include Schrödinger, who has developed complex models including FEP+, which uses the fundamental laws of physics and chemistry to predict how a drug interacts with a protein. Some Third Bridge experts now consider this as the gold standard for accuracy, particularly when a target is novel and lacks the historical data required for other AI models.
Hybrid Tech-Bio players include Recursion Pharmaceuticals, which looks to integrate AI and ML [Machine Learning] with autonomous wet labs (robotic laboratories) to create a closed feedback loop. Recursion currently operates the BioHive-2, considered one of the most powerful supercomputers in the biopharma industry, which is used to train its models. This vertical integration has showcased tangible ROI, moving its REC-1245 candidate to the clinic in 18 months, compared to the 40-month industry standard.
Generative chemistry applications hope to not only increase the success rate of new drug discovery, but also lower the cost. Recursion and MIT’s newer AI models like Boltz-2 are approaching the accuracy of traditional physics methods for estimating binding affinity, but at a fraction of the computational costs - potentially commoditizing early-stage screening.
While the design phase has reached technical maturity, enterprise-wide adoption at scale remains hindered by internal data fragmentation and the ongoing translational gaps between in silico and human clinical outcomes.
Clinical development optimization
As assets move into clinical development, the focus shifts to how AI can improve trial efficiency and the probability of success (POS).
POS: Biology is complicated and extremely hard to predict. Hence the need for large, expensive clinical studies. Even then, the cumulative probability of success for a drug once it has entered human studies is low; as low as 10-15%1. In an attempt to increase this POS, the industry has been gravitating towards the concept of ‘precision medicine’, which relies on identifying molecular ‘signatures’ - aka biomarkers - that can then be used as a tool to find the ideal patient. Experts note that AI has a clear use-case in discovering new biomarkers, thus better matching treatments to patients. Not only does this increase the trial POS (particularly useful for phase 2, given this stage has the lowest POS - see figure 3.), but also increases the drug’s risk-benefit profile - a critical factor as the biopharma industry looks to sell into increasingly budget-contrained health systems.
Trial efficiency: Running a clinical trial represents one of the largest cost items, with a Phase 3 study accounting for up to 69% of overall R&D costs1. Thus, AI is seen as a key tool to control that cost. One example would include synthetic control arms; AI-driven biosimulation to create virtual control groups using Real-World Evidence, which can reduce or eliminate the need for placebo groups in trials - something particularly important in deadly diseases where using placebo, or sub-optimal therapy, may be seen as unethical. This would also reduce the number of patients needed in a study at an average cost of USD53k/patient for a Phase 32, thus lowering the costs as illustrated in Figure.3. Example companies developing such synthetic control arms include Medidata AI with large data providers like IQVIA and Veeva well positioned given their proprietary datasets.
AI can also be used to increase enrollment patient speed, and thus decrease the overall duration of the study, thereby potentially lowering the aforementioned average cost per patient. Specialized AI tools are being used to analyze electronic health records (EHRs) to find eligible patients faster, ultimately reducing the timelines.
Figure 3. Average Cost and POS by study phase.
Source: Pubmed
Downstream impact on CROs
There is clear investor concern that increased biopharma R&D productivity will impact the magnitude of outsourced spend on third-party services such as CROs. However, Third Bridge experts view AI as more of a tailwind because of:
Specialized outsourcing: Small and mid-sized biotech companies lack the internal infrastructure to build proprietary AI stacks, forcing them to rely on CROs. Moreover, they lack the resources to manage the transition from in silico discovery to clinical execution
Focus on high-value workflows: AI predominantly automates smaller, less profitable manual tasks (like data cleaning), allowing for more efficient employee utilization (CROs can be notoriously inefficient) and thus CRO margins expansion
Pipeline expansion: By increasing the productivity of early discovery, AI is expected to increase the total number of assets in development, thereby driving higher demand for late-stage, high-value clinical workflows
All insights in this article are based on information provided by Third Bridge experts.
For media inquiries, please contact: comms@thirdbridge.com
Transcript references:
1. AI in Early R&D – Downstream Impacts on LST, Bioprocessing & CRO Outsourcing (5 March 2026)
2. Generative AI Impacts on Drug Discovery & CRO & CDMO Demand (8 Jan 2025)
3. Schrödinger - Defending the Physics Moat in a Generative World & Potential Toxicity Solutions (4 Feb 2026)
4. Recursion Pharmaceuticals – Pipeline Prioritisation & Phenomics Validation (16 Dec 2025)
5. Recursion Pharmaceuticals – Clinical Validation, Generative Chemistry & Platform Benchmarking (3 Feb 2026)
6. Certara – Tier 1 Budget Shifts & Reality of AI in Integrated Drug Development (2 Feb 2026)
References:
2. https://www.ncbi.nlm.nih.gov/books/NBK611884/table/ch5.tab5/?report=objectonly
