Merck

Clinical study reports (CSRs) are traditionally labor-intensive, requiring teams of medical writers to sift through thousands of pages of clinical data over months. Writers must resolve conflicting data, verify results, and construct narratives using highly specific regulatory language. Despite widespread interest in applying generative AI to streamline this work, companies faced challenges building a platform capable of generating full CSRs of regulatory quality.

Merck developed a proprietary internal generative AI platform leveraging large language models (LLMs) to expedite the creation of first drafts of CSRs. The platform integrates advanced table pre-processing with generative AI authoring, managing table mapping, data extraction, styling, and validation. A team of over 80 data science, AI, life science technologists, and medical experts across three continents co-developed the platform with McKinsey, and Merck trained teams in prompt engineering and data pipelines to oversee it.

The platform reduced CSR first draft creation time from two to three weeks down to just three to four days. In tested use cases, the time to produce a fully human-reviewed first draft dropped from an average of 180 hours to 80 hours. Error rates in draft CSRs were reduced by 50% across categories such as data, messaging, citations, terminology, and typography. Merck is scaling the platform across its late-phase pipeline by end of 2025.

• Rigorous human oversight by qualified medical writers remains essential even with a high-performing generative AI platform.
• Deep cross-functional collaboration (data science, AI, clinical, regulatory, and external partners) is required to build regulatory-grade document automation.
• Scaling gen AI in regulated industries requires significant workforce upskilling in prompt engineering and data pipelines, not just technology deployment.