Bildnachweis: TensorSpark – stock.adobe.com , Muchlisin – stock.adobe.com.
The pharmaceutical industry is defined by well-known metrics: decade-long development timelines, billion-dollar costs, high failure rates. Against this backdrop, AI promises to make drug development faster, cheaper, and more successful. Yet, despite substantial investment, its impact is still hard to measure. One reason is that the term “AI” spans a wide range of methods with distinct goals. Another factor is the long timelines: while effects on development speed and cost emerge early, improvements in success rates take years to become visible. By Pia Siegele
The many faces of AI in drug development
I often come across phrases like “AI-developed drug” or “AI-native biotech” in discussions about how AI will transform drug development. These labels tend to flatten what is, in reality, a broad set of technologies and use cases. In practice, using “AI” in preclinical development can mean anything from identifying new targets, designing and optimizing new molecules and proteins, repurposing existing drugs, to supporting lab automation. In clinical development, it can mean using AI to define endpoints, select patient populations and trial sites, detect study risks early, or automate tasks such as clinical document drafting.
Accordingly, biotech and pharma players are taking distinct strategic approaches to AI. Some so-called “AI-native” biotechs like Insilico Medicine and Recursion Pharmaceuticals apply AI broadly across their platforms. Others focus on specific use cases; Owkin, for example, focuses on target discovery and biomarker identification, Schrödinger on small molecule design and computational chemistry, and Generate Biomedicines on biologics design. Large pharma players are starting to deploy AI across their full value chain but still tend to prioritize select areas. Eli Lilly’s collaboration with NVIDIA, for instance, initially focuses on drug discovery and AI-assisted laboratory experimentation.
In the long run, AI will become a standard tool across the drug development cycle, and differentiation between players will become increasingly limited. For now, though, the race is on to identify, develop and scale the highest-value use cases.
Improvements in speed (and cost) are visible
Where does AI currently create the most value? In preclinical development, the most visible gains are in molecule and protein design. Amgen reported that AI-driven protein design reduced antibody lead optimization time by 50% and cut the preclinical development time for complex multi-specific antibody drugs by up to 70%. Roche/Genentech likewise noted a 25% acceleration in the design of an oncology molecule and a >70% reduction in the time needed to design a small molecule for an immunology target.
In clinical development, AI-guided site selection stands out. Novartis, for instance, reduced its typical four- to six-week site selection process for a 14,000-patient late-stage cardiovascular trial to a two-hour meeting, while still closely hitting enrollment targets.
Insilico Medicine offers an early signal of what fully AI-integrated workflows can achieve, reporting timelines of 12-18 months from discovery to preclinical candidate nomination versus industry averages of 2.5-4 years. That said, this speed is not driven by AI alone and likely also reflects other factors such as shorter decision cycles.
AI is already pulling forward key value inflection points in (pre)clinical development, with further acceleration as adoption broadens. Given the high cost of development, this translates into cost savings. In addition, for successful drugs, faster development shortens time to market, extends effective patent life, and ultimately drives higher revenues.
The challenge of development success
The impact of AI on development success is still uncertain. Early signals come from a 2024 BCG study which analyzed the pipelines of “AI-native” biotechs. The study found that 21 out of 24 molecules had successfully passed Phase 1 trials, implying a success rate of 80-90%, well above industry averages of 40-65%. The sample size remains small, and the outperformance likely partly reflects more selective pipelines; for example, three molecules were repurposed drugs with known safety profiles. Still, the data points towards AI’s potential to improve drug safety.
Data on Phase 2/3 success rates is still scarce. Several of the above-mentioned molecules have already failed in Phase 2 or 3. Only once more AI-assisted compounds progress through later stage trials will it become clear whether AI improves efficacy and overall success rates.
That said, even if these success rates initially do not exceed historical benchmarks, this would not necessarily imply that AI lacks long-term impact. As noted earlier, AI can support drug development in multiple ways, and many first-generation AI-assisted drugs have mainly used AI for molecule and protein design. This application tends to optimize binding affinity and pharmacokinetic properties, which are essential but do not directly determine efficacy. The key question is whether AI models focused on disease biology, such as models for target and indication selection, will achieve improved efficacy.
Conclusion
AI in drug development is a powerful tool operating in a highly complex system. We at TVM are convinced of its long-term potential when applied in a targeted way, with a critical eye on real impact. In future Going Public articles, we will discuss key success factors and mid-term challenges for AI in this area and how it can become a tool to support better venture capital investment decisions.
Autor/Autorin
Pia Siegele
Pia Siegele joined the TVM team in early 2026. Prior to this, she worked at McKinsey & Company, where she advised life sciences companies on portfolio strategy and the application of AI and digital technologies in drug development. She holds degrees in biomedical science and informatics from the University of Edinburgh, with a focus on neuroscience and artificial intelligence, as well as a degree in geopolitical risk from the Johns Hopkins School of Advanced International Studies.
