AI in clinical trials: practical uses and limits

Artificial intelligence (AI) can help clinical research teams draft protocols, manage data and plan trials. Sponsors, contract research organizations (CROs) and technology partners are assessing where it fits into everyday workflows. Its practical impact depends on the task, the available data and the review required before teams can act on its output.

Start with the problem

Clinical trials are highly regulated, data-intensive and reliant on multiple stakeholders, systems and workflows, meaning that AI can provide practical benefits only when applied within a strong data infrastructure and guided by appropriate human oversight. Simply adding AI to a process can increase cost, complexity and governance challenges without improving operations.

A more effective approach starts by identifying the problem (whether a process is slow, repetitive, inconsistent or difficult to scale), then assessing whether AI is the right solution. In some cases, process redesign, better data integration or improved workflow management may be more effective. When AI is appropriate, effective implementation can reduce manual effort, accelerate document preparation or help identify operational risks earlier.

Clinical research teams must assess emerging technologies against requirements for patient safety, trial integrity and regulatory compliance. Protecting patient data requires clear controls over how staff enter information into AI-enabled systems. Safeguards include de-identification, secure access and vendor oversight.

Build on reliable data

The usefulness and suitability of AI in clinical trials depend heavily on the quality and structure of the data available, as well as the specific AI tool and governance controls in place.¹ Modern trials generate information from many different systems, including electronic data capture platforms, clinical trial management systems, randomization tools, safety databases, patient-reported outcome platforms, wearable devices and operational dashboards. Historically, these systems have often been fragmented, creating isolated snapshots of trial activity.

AI tools can struggle to identify patterns when trial data is fragmented across sites and systems. Early attempts to introduce AI or analytics capabilities to individual platforms may have provided value within those systems, but they often limited the wider insights that could be generated across the trial. Aggregating data across platforms allows organizations to apply AI to a broader dataset. This can support analyses that draw on information from several parts of a trial.

Incomplete, inconsistent or poorly structured datasets require attention before teams can use them reliably with AI. For clinical trials, where data integrity is central to regulatory confidence and patient safety, data quality, standardization and governance remain essential prerequisites for effective AI adoption.

Identify practical use cases

AI can help in protocol drafting by gathering information, organizing content and generating first drafts, reducing the time teams spend assembling repetitive sections or searching for relevant background materials. Experts still need to check, refine and contextualize the output.

A similar pattern is emerging in downstream document generation, operational planning, training materials and information summarization. AI can help to create a starting point, but clinical, regulatory and operational experts remain responsible for determining whether the content is accurate, appropriate and aligned with the study’s objectives.

Patient recruitment presents more complex challenges. Connecting trial teams to eligible patients across health databases, sites and institutions involves several constraints. Privacy requirements, consent considerations, fragmented healthcare systems and differences between public and private data environments all limit what can be achieved.² AI may still improve recruitment over time, aiding with feasibility analysis, identifying trends and refining outreach strategies, but it is unlikely to provide a simple solution.

Administrative and operational processes may therefore offer some of the most achievable near-term gains. Tasks such as document triage, invoice review, data validation and workflow summarization are often more standardized than clinical decision-making, making them better suited to controlled AI implementation.

Account for specialized data

The type and amount of data available within a therapeutic area also affect AI adoption, as large, structured and representative datasets are easier to use for training and validation. This creates natural advantages in fields with extensive trial activity, such as oncology, where substantial clinical trial data can increase the potential for AI model development. Specialized areas such as dermatology and rheumatology can be more challenging, as trials in these fields may generate smaller datasets and many endpoints depend on clinical assessment, visual interpretation or patient-reported experience.

Subjective data can still be valuable, but it requires careful handling and consistent collection, and the challenge is often whether there is enough usable, standardized information to train and validate AI tools effectively. Historical data may also be difficult to use, as clinical trials conducted decades ago may not have captured information in formats suitable for modern AI applications, even when the scientific content remains relevant.

As a result, some specialized fields may need more time to build the datasets required for advanced AI use, while broader operational processes, such as document generation, data flow management and administrative efficiency, may offer more immediate opportunities.

Define oversight and accountability

AI’s growing presence in clinical trial operations is making governance increasingly important, particularly as sponsors expect CROs and vendors to use modern tools while also requiring reassurance that they are being applied appropriately and securely. AI use raises questions about confidentiality, data protection, validation and accountability. Most clinical trials are likely to involve multiple organizations with different levels of AI maturity, varying appetites for innovation and risk, making alignment essential.

Policies must define where AI can be used, what data can be entered into AI-enabled systems, how outputs are reviewed, and who remains accountable for final decisions.³ Teams also need to consider whether a task is appropriate for AI assistance. AI may assist with image analysis, highlight patterns or summarize complex information, but clinical judgment, patient communication and final accountability must remain with qualified human professionals.

AI will continue to change clinical research, but its progress is likely to be gradual. Technology is advancing quickly, while regulatory expectations and clinical trial processes move more cautiously. For each proposed use, teams need to define the task, establish that the data is suitable, resolve privacy concerns and assign responsibility for reviewing the result.

Guillaume Gigon is VP of Technology and AI at Indero. He brings over 15 years of expertise in clinical trials, leading innovation in technology, system validation and data systems.

References

1. U.S. Food and Drug Administration. Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products. Draft guidance, January 2025.
2. European Medicines Agency. Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle. Published September 30, 2024.
3. International Council for Harmonisation. Guideline for Good Clinical Practice E6(R3). Final version adopted January 6, 2025.