When people ask how Artificial Intelligence is being used in clinical research, they often expect to hear about revolutionary new technologies transforming every aspect of clinical trials.

Instead, what we’re seeing is something much more practical, and perhaps much more important.

During our latest AI in Clinical Trials Working Group meeting, members shared how their organizations are moving beyond experimentation and beginning to operationalize AI. While the use cases themselves were valuable, the bigger story wasn’t what organizations were using AI for. It was how they were implementing it successfully. Several common patterns emerged.

Start with Low-Risk, High-Value Work

The organizations seeing the greatest success didn’t begin by automating complex clinical decisions.

Instead, they started with administrative tasks that consume valuable staff time but still allow for straightforward human review.

Participants described using AI to:

• Draft emails and routine communications
• Summarize meetings and lengthy documents
• Complete forms from protocol amendments
• Generate sponsor metrics and operational reports
• Create staff training materials
• Conduct market and literature reviews
• Brainstorm workflow improvements

These activities reduce administrative burden without replacing human judgment.

One participant noted that processing protocol amendments with AI saves approximately 30 minutes per amendment because the system extracts the information required for the IRB submission form. That may not sound revolutionary—but multiplied across hundreds of amendments each year, those savings become meaningful.

Human Oversight is a Key Design Principle

Perhaps the strongest consensus throughout the discussion was that AI should augment people, not replace them.

Organizations are intentionally designing workflows where AI handles repetitive work while humans remain responsible for decisions.

Examples included:

• Automatically rejecting high-complexity amendments that require expert review.
• Restricting AI agents from drafting first versions of research protocols.
• Requiring human review before patient-facing materials are submitted to the IRB.
• Using AI to generate first drafts rather than final products.

Rather than viewing human oversight as a regulatory burden, participants described it as an essential feature of successful AI implementation.

Build Institutional AI, Not Individual AI

Another interesting trend was the movement away from individuals experimenting with public AI tools toward institutionally supported AI environments.

Several organizations described developing internal AI assistants that combine enterprise AI models with institutional knowledge, policies, templates, and procedures.

One participant shared their organization’s “Research Concierge”, an AI agent that helps investigators locate templates, policies, regulatory guidance, and the appropriate administrative contacts. Importantly, the agent was intentionally restricted from writing essential research documents, instead directing users to the appropriate experts and resources.

Others described Retrieval Augmented Generation (RAG) implementations that allow AI to search approved institutional content while preventing sensitive information from leaving the organization’s environment.

The message was clear: organizations are beginning to invest in AI infrastructure rather than relying solely on individual users.

Governance Is Becoming More Practical

One of the most valuable discussions centered around governance—not as a theoretical framework, but as an operational process.

One organization described obtaining IRB approval for an internal AI agent by documenting:

• The purpose of the tool
• What information it could access
• What it was explicitly prohibited from doing
• Reference materials and trusted information sources
• The limitations of the system

This practical example demonstrated that governance can move beyond high-level principles into repeatable organizational processes.

Similarly, several participants described sponsor-approved approaches for using AI with study materials. Rather than uploading complete protocols, they work from high-level summaries, avoid intellectual property, and follow clearly defined boundaries regarding what information can and cannot be shared.

The Biggest Barrier Isn’t Technology

Ironically, the discussion suggested that today’s biggest challenges aren’t technical.

They’re organizational.

Different sponsors maintain different AI policies. Some explicitly permit AI-assisted workflows, while others prohibit study-related content from entering any AI system.

Institutions continue to develop internal guidance around approved tools, data handling, and acceptable use.

Staff are receiving various levels of AI training from their institutions and are often left to learn and experiment on their own. This results in a wide range of technical ability and interest in AI adoption. Some institutions are beginning to develop standardized AI training on how to use AI effectively, but also on when not to use it.

The result is a patchwork of governance approaches that makes implementation inconsistent across organizations.

Five Lessons Emerging Across Organizations

Although every organization represented was at a different stage of AI maturity, several common lessons emerged:

• Start with administrative workflows before regulated clinical activities.
• Design human oversight into every AI-enabled process.
• Establish clear boundaries around what information AI can access.
• Invest in institution-level AI capabilities rather than relying solely on individual experimentation.
• Develop governance alongside implementation rather than after deployment.

These lessons may seem straightforward, but together they represent a shift in how the industry is approaching AI.

Moving from Experimentation to Operationalization

Perhaps the biggest takeaway from the discussion is that clinical research is entering a new phase of AI adoption.

The conversation is no longer, “Should we use AI?”

Instead, organizations are asking:

• Which workflows should we tackle first?
• How do we implement AI responsibly?
• What governance is actually necessary?
• How do we build trust with sponsors, regulators, and research teams?

The most successful organizations aren’t necessarily those with the most advanced technology. They’re the ones building thoughtful implementation strategies that balance efficiency with oversight.

That may ultimately prove to be the most important innovation of all.