Protocol AI: How We Built CRF Auto-Generation

Every clinical trial starts with a Protocol — a detailed document specifying the study design, endpoints, visit schedule, and data collection requirements. Translating that protocol into case report forms (CRFs) has traditionally been a manual, weeks-long process. We built Protocol AI to change that.

The traditional CRF design process

In a typical study, CRF design works like this:

1. The medical team writes the protocol (4–12 weeks)
2. A data management team reads the protocol and designs CRF pages (2–4 weeks)
3. The sponsor reviews CRF drafts and requests changes (1–2 weeks, often multiple cycles)
4. CRFs are built in the EDC system (1–2 weeks)
5. User acceptance testing is performed (1 week)

Total: 8–20+ weeks from protocol to production CRFs. And that’s if everything goes smoothly.

What Protocol AI does

Protocol AI reads your protocol document and automatically generates a complete CRF package. Here’s how it works:

• Protocol parsing. The AI reads the protocol PDF and extracts the visit schedule, procedures, endpoints, inclusion/exclusion criteria, and data collection requirements.
• CRF generation. Based on the extracted requirements, it generates CDISC-aligned CRF forms with appropriate field types, edit checks, and visit mapping.
• Standards compliance. Forms are generated following CDASH standards and your organization’s library of approved CRF modules.
• Human review. The generated CRFs are presented for medical and data management review — they can be modified, approved, or regenerated with different parameters.

What we learned building it

Building Protocol AI taught us several lessons about applying AI to clinical trial operations:

• Protocols are surprisingly inconsistent. The same concept might be described five different ways across five protocols. The AI needs to understand intent, not just match keywords.
• Standards matter enormously. Without a strong CDISC/CDASH foundation, AI-generated CRFs would be technically correct but operationally useless.
• The 80/20 rule applies. AI can generate 80% of CRF content accurately from the protocol. The remaining 20% — complex conditional logic, sponsor-specific conventions, and edge cases — still needs human expertise.
• Speed creates value even when perfection isn’t possible. Getting a 90% complete draft in hours instead of weeks compresses the entire study startup timeline.

Results so far

Across studies using Protocol AI:

• CRF design time reduced from weeks to hours
• 30% fewer review cycles due to better first-draft quality
• Improved CDISC compliance in CRF design
• Faster study startup — first patient in sooner

See Protocol AI in action — upload a protocol and watch it generate CRFs in real time.