Accelerate clinical trial start-up: Beginning with the protocol

Deloitte US built a CSP writing process in which AI agents do the heavy lifting while people stay in charge. It grounds its work in science, reuses approved content, and lets writers draft interactively. The tool is powered by Claude, implemented on AWS Bedrock and securely integrated into AstraZeneca’s existing AWS technology environment. Claude drafts detailed scientific sections, answers questions about the source documents, and refines drafts. Deloitte US designed the overall process, the step-by-step guidance, and the built-in review controls so it works for everyday use, with writers firmly in control.

How it works, end-to-end:

Research and gather: The tool can rapidly identify and extract relevant content from prior protocols and scientific literature, while filtering and drilling down to the most pertinent sources for a current study.

Synthesis and authoring: The tool can help teams draft open-ended narrative sections and merge them with standard protocol language to produce a high-quality first draft faster.

Iterate with functions: The tool supports collaborative authoring and review, helping enable quicker iteration cycles through AI-assisted research, drafting and quality checks—while keeping authors in control.

Key capabilities:

• Evidence-backed drafting: Pull from prior protocols and scientific literature with inline citations to speed review and strengthen defensibility.
• Reusable structure and from-scratch drafting: Use pre-approved templates or libraries as a foundation, then add custom-written sections—this maintains the required structure while producing tailored content faster.
• Guided, consistent authoring: Section-level guidance and prompts help improve consistency across authors and protocol sections.
• Interactive refinement: Ask questions of source documents and iteratively edit drafts to validate assumptions and refine language.
• Quality and governance controls built in: Human-in-the-loop controls and agent-led quality checks help reduce avoidable rework before broader reviews.
• Works with your existing tools: Takes in rich content like tables and figures in many formats, and exports to the platforms teams already use for collaboration and approvals.

AstraZeneca plans to use the tool for more studies. Building on this foundation, there is the potential to extend the same agentic authoring pattern—research, synthesis, and collaborative iteration—to additional templates and document types across more clinical trials.