Most pharmaceutical companies are investing heavily in artificial intelligence for drug discovery, but the majority of executives doubt AI will actually make drugs more likely to succeed in clinical trials. A new survey from Citi found that while 72% of respondents said their companies were either scaling or had fully scaled AI across research and development, only 24% expected AI to deliver a significant improvement in drug success probability.

Why Are Pharma Leaders Hesitant Despite Widespread AI Adoption?

The disconnect between AI investment and confidence in outcomes reveals a fundamental tension in the pharmaceutical industry. Companies recognize that AI can accelerate the discovery process, helping researchers identify promising drug candidates faster than traditional methods. However, speed alone doesn't guarantee better medicines. John Yung, head of Asia healthcare research at Citi, explained the core problem: "The biggest risk to the AI-powered drug discovery thesis is not that AI fails to accelerate discovery. The stronger risk is that acceleration does not translate into better drugs or higher probability of success".

"It's only through that translation that companies can turn AI-driven speed into profit, and patients into real beneficiaries," said John Yung.

John Yung, Head of Asia Healthcare Research, Citi

The survey included responses from pharmaceutical executives across the industry, and the findings paint a sobering picture. While 56% of respondents expected only a moderate uplift in drug success rates, another 34% identified a more troubling bottleneck: human biology, clinical judgment, and execution will remain limiting factors regardless of how much AI is deployed. In other words, even if AI can process data faster and identify novel drug targets more efficiently, the fundamental challenges of developing safe, effective medicines remain stubbornly human.

What Are the Real Barriers to AI-Driven Drug Success?

The pharmaceutical industry faces several interconnected challenges that AI alone cannot solve. Drug discovery is only the first step in a long, expensive pipeline. Once a promising compound is identified, it must undergo preclinical testing, clinical trials across multiple phases, and regulatory review. Each stage introduces uncertainty and the possibility of failure. AI can help optimize earlier stages, but it cannot eliminate the biological complexity that causes many drugs to fail in human trials.

• Clinical Translation Gap: AI excels at finding patterns in existing data and predicting molecular properties, but predicting how a drug will behave in a living human body remains extraordinarily difficult. Animal models and early-stage human trials often reveal unexpected side effects or efficacy issues that computational models missed.
• Regulatory and Execution Risk: Even if AI identifies a promising drug candidate, regulatory agencies like the FDA require extensive evidence of safety and efficacy. The approval process depends on human judgment, clinical trial design, and the quality of data collection. AI cannot replace these gatekeeping functions.
• Biological Complexity: Human biology is far more complex than any current AI model can fully capture. Genetic variation, environmental factors, and individual patient differences mean that a drug effective in one population may fail in another. AI can help identify these patterns, but understanding causation requires deeper biological insight.

The survey results suggest that pharmaceutical executives understand these limitations. No company reported having no plans to adopt AI, indicating universal recognition that the technology is essential for competitive advantage. However, the cautious outlook on success rates reflects realistic expectations about what AI can and cannot accomplish in drug development.

How to Align AI Investment With Realistic Outcomes

• Focus on Specific Use Cases: Rather than deploying AI broadly across all research functions, companies should identify specific bottlenecks where AI can demonstrably improve outcomes, such as target identification or lead optimization, and measure impact carefully.
• Invest in Data Quality: AI is only as good as the data it learns from. Pharmaceutical companies should prioritize data standardization, validation, and integration across research teams to maximize the value of AI systems.
• Combine AI With Human Expertise: The most effective approach pairs AI's pattern-recognition capabilities with human researchers' deep biological knowledge and clinical judgment. Rather than replacing scientists, AI should augment their decision-making.
• Plan for Longer Timelines: Companies should set realistic expectations about when AI-driven drug candidates will reach patients. Even accelerated discovery timelines must account for regulatory requirements and clinical trial phases that cannot be shortened.

The pharmaceutical industry's cautious optimism about AI reflects maturity in how the sector approaches emerging technologies. While earlier hype suggested AI would revolutionize drug development overnight, executives now recognize that innovation in medicine requires both technological advancement and patience with biological reality. The next phase of AI in pharma will likely focus less on grand promises and more on measurable improvements in specific discovery tasks, combined with honest acknowledgment of the limitations that human biology and regulatory requirements impose.

As the FDA continues to build its own AI capabilities, including the recent appointment of Jared Seehafer as the agency's first deputy commissioner for Technology and Artificial Intelligence, regulatory frameworks will evolve to better accommodate AI-driven drug development. However, these regulatory advances will not eliminate the fundamental challenge that executives identified: translating speed into success. That translation will require not just better algorithms, but better understanding of how to bridge the gap between computational prediction and clinical reality.