Artificial Intelligence (AI) is reasonably well established in discovery and preclinical research, with success in areas such as target identification and molecular modeling. However, in clinical development, adoption has been fragmented and challenging to embed across the lifecycle.

So, what are the challenges, and what is the opportunity to catalyze? The short answer: the challenge is due to complexity. The longer answer spans data interoperability, cultural inertia, and regulatory uncertainty across an ecosystem that includes not only sponsors and regulators, but also investigators, patients, technology companies, and healthcare policymakers. The opportunity to catalyze lies in simplifying and streamlining that complexity to drive meaningful progress.

Although we are recognizing progress, the learning curve is steep and the realities of conducting regulated research and implementing changes that impact patient safety carry a massive responsibility.

Why clinical lags behind discovery

Discovery and preclinical domains involve smaller and more focused experiments and are organizationally less matrixed. The data types are simpler and the workflows more straightforward.

The clinical trials enterprise, by contrast, operates in a highly matrixed environment, where scientific goals must align with regulatory requirements and commercial imperatives while considering the challenges of global implementation. Any innovative technology introduced into this space must meet a high bar in terms of both performance and its ability to earn trust across stakeholders and geographies.

It is no surprise that discovery organizations got traction first. With clearer workflows and less organizational sprawl, it makes it easier to pilot and scale. Large sponsors and contract research organizations face persistent challenges integrating legacy systems, processes, and people — slowing the adoption of data standards needed to streamline data exchange. Building AI models with confidence at pace requires us to reduce the effort and cost of data integration.

Where AI is gaining ground

Generative AI has given rise to exploring regulatory document generation — for example, protocol and report development, along with consent document creation and translation. Wider sharing of historical documents for model training could reduce hallucinations and increase confidence in AI-automated tasks. These efficiencies help with workflow, but they do not address the primary bottleneck: patient recruitment.

AI’s greatest potential in clinical research lies in enhancing trial design to accelerate participant recruitment and retention. One means of accelerating is by reducing the friction for participation by simplifying critical study design elements, including eligibility criteria and procedures, thereby reducing participant and investigator demands. AI can help assess trial burden and inform study decisions. Additionally, clinical trials are becoming much more complex and data intensive. From 2001 to 2020, the number of trial endpoints increased by 215 percent and procedures by 140 percent, while data volume surged nearly 600 percent.

A second major opportunity is reducing the number of patients required. Model-informed design could help predict which populations are most likely to respond, allowing for more targeted recruitment; in this case, through patient enrichment. Supplementing or replacing control arms with data from historical trials or real-world data can simulate active controls. While not appropriate for every trial, reducing any enrollment burden is an opportunity for acceleration.

AI also supports the promise of in silico modeling, which is being explored in areas such as drug interaction studies and pharmacovigilance. Some cases rely on AI to simulate outcomes without direct patient exposure. Still, they require regulatory engagement and rigorous validation to ensure models are contextually reliable. The impact of early efforts will be shaped by selecting opportunities that are built on robust data sets along with transparency behind the models and algorithms.

The role of real-world data

Once used primarily after approval, real-world data (RWD) is finding growing relevance earlier in the clinical process. Sponsors are incorporating observational data into long-term extension studies, real-world safety tracking, clinical trial planning, and exploratory endpoint validation.

Yet turning this potential into consistent value remains difficult. Real-time access to electronic health records and claims data is still limited, particularly in the fragmented US healthcare system. In other regions, strict data privacy laws pose different but equally significant barriers. Fully unlocking the potential of RWD — for both trial design and AI model development — will require collaboration among sponsors, standards organizations, technology vendors, health systems, and policymakers.

The real barriers to scale

Progress depends on technical capability but also trust. That trust is contingent on three enablers: clean, interoperable data; a culture of responsible sharing; and transparency — both with patients about how their data is used and with researchers and regulators about how AI tools perform. Data readiness is still a sticking point. Today, preparing clinical data for AI training is labor-intensive due to inconsistent formats, lack of translation across data types, and legacy systems. Building momentum and scaling AI will require a broader industry commitment to standards adoption and investing more deliberately in harmonization.

Scalable AI will also require scalable collaboration. Federated data approaches — sharing insights without transferring raw data — help reduce security and intellectual property concerns while enabling more robust model training. Advances in privacy-preserving analytics and distributed learning are key to making this practical.

Human factors matter too. Even with infrastructure in place, uptake depends on willingness to experiment, validate impact, publish successes and failures, and evolve practices. Teams need tools, time, and support to learn as they go.

Finally, AI developers need a clearer view of user needs to ensure their tools are accurate and actionable.

Laying the groundwork for what comes next

At TransCelerate, our work has long been centered on establishing a foundation for innovation. Whether through championing clinical data standards, developing frameworks for responsible data sharing, or educating technology partners on early AI use cases, we are focused on enabling the conditions that will allow AI to work, not just technically, but credibly.

AI adoption in clinical development is still in the experimental phase regarding transformational use cases. As sponsors test new applications across the development lifecycle, the most significant headway will come from shared learning, especially when it helps others avoid repeating the same early mistakes. Pilots have revealed what is possible. The real test is whether we are ready to make AI foundational to clinical research. Better outcomes and faster innovation make that shift worth the effort.

This article was contributed by Rob DiCicco, Vice President of Portfolio Management at TransCelerate BioPharma.