Damien Ng, senior research analyst at Julius Baer, argues that AI can’t guarantee successful drugs. But it could help identify potential failures earlier.
The pharmaceutical industry celebrates breakthrough stories. A new therapy reaches patients. A life-threatening disease becomes treatable. What receives far less attention are the countless ideas that lead nowhere.
For every successful medicine, thousands of potential drug candidates disappear along the development pathway. Years of research can culminate in an uncomfortable realisation: this was the wrong molecule or the wrong biological hypothesis. Only around one in ten drug candidates entering clinical trials ultimately reaches approval, after a process that can take more than a decade and consume billions of dollars.
One of the industry’s greatest challenges is determining which scientific ideas have the best prospects of success. This is why the most important impact of artificial intelligence in biopharma has little to do with replacing scientists or automating laboratories. Its greatest contribution lies in helping researchers make better decisions before committing years of work and enormous pools of capital. Drug development is fundamentally a prediction problem.
Faster isn’t always better
Much of the discussion around AI focuses on speed, from faster drug discovery to faster molecule design. Although those benefits are real, they may miss the bigger picture as a faster journey towards the wrong destination delivers little value.
Scientists seek to understand whether a biological target plays a meaningful role in disease. However, they also need to assess how a potential molecule will interact with that target and whether the treatment can deliver meaningful benefits for patients. As a result, every decision involves uncertainty.
For decades, researchers have relied on a combination of scientific expertise and experimentation to navigate that complexity. Today, they can draw upon unprecedented quantities of biological data. Genomics, proteomics, imaging and real-world clinical records provide a far richer understanding of human health than any previous generation of scientists could access. But the difficulty is that no human team can fully comprehend datasets of this scale and complexity.
This is where AI begins to demonstrate its practical impact. Algorithms can sift through millions of data points in search of meaningful signals. Research teams are using AI models to uncover relationships between genes and disease pathways that previously remained hidden within vast datasets. DeepMind’s AlphaFold, for example, predicted the structures of more than 200 million proteins, providing scientists with a powerful resource for understanding disease biology and identifying drug targets.
The technology is also beginning to influence commercial drug development. Companies such as Insilico Medicine have advanced AI-assisted drug candidates into clinical trials, demonstrating how computational approaches can help identify and prioritise promising therapeutic opportunities.
This potential to improve the selection of drug targets matters because many failures in drug discovery and development can be traced back to decisions made at the very beginning. Pursuing the wrong target frequently leads to disappointment years later, after substantial investments have already been made.
Anyone who has spent time around pharmaceutical research & development knows that scientific setbacks are not the exception but rather part of the process. This makes the goal early recognition of failure as opposed to elimination of failure completely. The meaningful impact of AI’s role in ruling out weak biological hypotheses early on, particularly before entering expensive laboratory and clinical programmes, extends beyond saving money. Instead, resources can be redirected towards more promising opportunities, improving the overall productivity of innovation.
AI’s ability to identify meaningful patterns also applies to clinical trials. Patient populations are rarely homogeneous. Two individuals carrying the same diagnosis may have profoundly different underlying biology. AI can help detect those differences, enabling more precise patient selection and generating clearer evidence about which therapies work and for whom.

Reshaping competitive advantage
The growing adoption of AI across drug discovery and development could also reshape the sources of competitive advantage in biopharma. AI may democratise some aspects of discovery by making sophisticated analytical tools more widely available. But it may simultaneously increase the value of unique biological and clinical datasets. While algorithms can often be replicated, proprietary patient records and decades of accumulated research cannot. As AI capabilities become more widely available, access to distinctive data may emerge as an even more important source of competitive advantage.
Organisations with unique datasets and deep disease expertise are building competitive advantages that may prove difficult to replicate. The next winners in biopharma will be those that transform data into insight. From there, better predictions can pave the way for better medicines.
None of this suggests that algorithms will replace researchers. Scientific breakthroughs still require human curiosity and judgement. Biology remains too complex, and healthcare too consequential, for blind faith in machine-generated outputs.
AI’s real potential is far more practical. In an industry where every decision can carry a price tag of hundreds of millions of dollars, even modest improvements in predicting success and failure can have enormous consequences. The future of biopharma may therefore depend on making better decisions earlier, identifying which paths are worth pursuing before years of time and capital are committed.
Developing new medicines will always involve uncertainty. AI may improve the odds by helping researchers make better-informed decisions about which ideas to pursue, but it cannot guarantee success.



