Key Takeaways
- AI/ML adoption in drug development is accelerating but still limited in full implementation. While investment and regulatory support are growing, only 11% of companies have fully implemented AI/ML solutions. Larger organizations are more likely to adopt due to greater resources.
- AI/ML offers significant efficiency gains, especially in regulatory and clinical trial execution tasks. Reported benefits include an average 18% cycle time reduction, with standout time savings in areas such as patient monitoring and regulatory documentation preparation.
- Trust, data quality, and legal concerns are major adoption barriers. Despite promising results, two-thirds of companies report low confidence in AI/ML data accuracy, and concerns over data privacy, governance, and intellectual property remain top challenges.
Not surprisingly, the use of artificial intelligence (AI)-enabled activity and solutions supporting drug development has skyrocketed. For years, anecdotal evidence from vendors and impassioned professionals has pointed to the promise of AI/machine learning (ML) and its expected growth. Recent analyst reports project a 27%-30% compound annual growth rate during the next decade in total investment in AI/ML solutions supporting drug discovery and development, culminating in nearly $20 billion by 2033.
Now, regulatory tailwinds and an increasing body of empirical evidence are facilitating adoption and quantifying more nuanced usage and impact on performance.
Regulatory encouragement of the use of AI in drug development has promoted adoption during the past year, and recent announcements suggest it will accelerate. The FDA Center for Drug Evaluation and Research had been witnessing rapid growth in the number of new drug and biologics applications submitted using AI/ML elements—from only three submissions in 2018 to 170 submissions in 2023. In response, in early 2025, the FDA issued a draft guidance entitled “Considerations for the Use of Artificial Intelligence to Support Regulatory Decision Making for Drug and Biological Products.”
Adding to the growing body of empirical evidence on AI/ML adoption and its impact, the Tufts Center for the Study of Drug Development (Tufts CSDD)—in collaboration with the Drug Information Association—conducted a global assessment among pharmaceutical and biotechnology companies and contract research organizations (CROs). The primary aims of the study were to gather AI/ML use cases.
For the purposes of this study, AI was defined as a dedicated field of science—combining computer science, statistics, and engineering—that uses algorithms or models to reason, learn, and make decisions and predictions. ML is considered a subset of AI that allows models to be developed through the use of training algorithms and data analysis, without models being explicitly programmed. And generative AI is a type of AI that can create new content—including text, images, audio, and video—based on data and prompts.
This brief article provides a high-level summary of key takeaways from the assessment. A detailed manuscript will soon be published in Therapeutic Innovation and Regulatory Science.
A global assessment
The Tufts CSDD team conducted the online assessment between May 2024 and August 2024. A total of 302 respondents primarily from clinical operations, clinical development, and data sciences/data management functions— from 79 distinct sponsors and CRO companies —completed the assessment: 51% of respondents are based in North America; 26% in Europe; and the remainder in the rest of the world. Companies responding were distributed across three broad groups based on annual clinical trial volume: 39% conducted fewer than 25 clinical trials; 23% conducted between 25 and 100 trials during the past twelve months; and 38% conducted more than 100 clinical trials.
Respondents provided a total of 36 unique case examples of AI/ML use in drug development: 12 (33%) cases featured AI-enabled activities and experience in clinical trial design and planning; 20 (56%) cases presented AI-enabled activities in trial execution; and 11% featured AI-enabled activities supporting regulatory submissions.
Overall reported implementation
As of late 2024, only 11% of the nearly 80 companies responding reported fully implementing AI/ML to enable and support clinical trial activities. An additional 22% reported partially implementing AI/ML solutions.
AI/ML adoption appears to be a function of company size due in part to the resources available to larger organizations. A significantly higher percentage of large companies (those conducting 25 or more clinical trials annually) reported partially or fully implementing AI/ML-enabled solutions (36%) compared to smaller companies with lower annual clinical trial volume (25%).
A typical AI/ML investment is reportedly more than a million US dollars. On average, companies reported investing $1.1 million to implement an AI/ML-enabled activity. Higher levels of investment were reported for AI/ML solutions associated with clinical trial execution activities—with average investments three to four times higher than those associated with planning and design and regulatory submission activities.
The highest average reported investment ($3.2 million) was for AI/ML solutions supporting data quality and data cleaning. The next highest areas—an average of $2.2 and $1.9 million—were investments in AI/ML solutions supporting trial master file (TMF) management and investigative site selection, respectively.
Areas of use and reported impact
More than one-third (35.2%) of sponsor companies and CROs report partially and fully implementing AI/ML activities related to clinical trial execution, with the highest use (50%) reported in the analysis of genetic data and the next highest, 48%, in patient narratives.
For planning and design activities, 29% of sponsors and CROs reported partially and fully implementing AI/ML, with the highest reported adoption (52%) in the identification of diverse patient populations and the next highest, 40%, in risk and quality assessment planning.
On average, three-out-of-10 sponsors and CROs report using AI/ML-enabled tools to support regulatory submission activities. Nearly four-out-of-10 (38%) companies report fully or partially implementing AI/ML for TMF management and 32% to support clinical study report writing.
Although AI/ML-enabled solutions use in drug development is relatively nascent, companies are already reporting significant time reductions and efficiencies. Of the 36 clinical trial case examples assessed, an estimated 18% mean cycle time reduction was reported when AI/ML approaches were leveraged across planning and design, clinical trial execution, and regulatory submission use cases. This included not only deployment of a given solution but also the incremental human review time associated with it.
Regulatory submission activities—a domain that is particularly time- and labor-intensive— were associated with the highest reported overall time savings when AI/ML enabled solutions were used (see Table 1 below). Clinical study report writing and drug identification and labeling information had mean time savings of approximately 10% off typical durations when AI/ML solutions were deployed. Gathering, organizing, compiling, and harmonizing documentation to complete regulatory submissions received the highest mean estimated time savings—63% off the typical duration when AI/ML-enabled approaches were used.
Within the domain of clinical trial execution activities, AI/ML-enabled approaches supporting patient monitoring received the highest reported average time savings (a 75% time reduction). Use of AI/ML approaches supporting patient enrollment assessments delivered a reported mean 45% time reduction.
AI/ML-enabled approaches used to support planning and design activities delivered a reported 13% average time reduction (see Table 1). Use of AI/ML approaches supporting the identification of targeted patient communities received the highest reported average time savings—a 68% time reduction—of all use cases in the planning and design domain.
Implementation challenges
At this time, trust in AI/ML-generated output was among the top implementation challenges reported. The majority—two-thirds of sponsors and CROs—indicated low confidence in the accuracy and quality of the underlying data used to train the solutions.
Intellectual property and legal concerns, as well as concerns about data governance and data privacy, were also reported as areas of high resistance and top challenges hindering adoption.
Ken Getz, MBA, Executive Director and Research Professor, Tufts Center for the Study of Drug Development, Tufts University School of Medicine
