November 4, 2026 | 11:00 AM ET
Some clinical research work has resisted automation for a reason. Read a protocol. Extract what matters. Make a series of judgment calls that depend on context and experience. Enter it correctly into a software system. Many inputs, many outputs, and almost no tolerance for getting it wrong. That description fits eSource build — and it fits a dozen other workflows on both the site and the sponsor side.
Luke Snedaker, CEO of West Clinical Research, and Mike Wenger, Chief Innovation Officer at CRIO, tell the story of what it actually took to make AI trustworthy against that kind of problem. That story includes the early version that handed the whole job to AI and produced a build no site could rely on. It’s also about what came next: AI treated as one tool in the toolbox alongside deterministic code, wrapped in a human review workflow with audit trails and checks against rubber-stamping. It covers the part of the AI conversation that usually gets skipped: the compliance and security groundwork underneath it — enterprise agreements with foundational model providers whose terms prohibit training on customer data, and the data-handling questions any site or sponsor should be asking a vendor before a single study document is uploaded.
The takeaway isn’t a product. It’s a pattern. The goal was never a self-driving car with nobody behind the wheel. It’s a driver-assist system, where the person stays in control and the technology does a meaningful share of the work without ever taking on the responsibility. Attendees will leave with a way to judge where AI belongs in their own high-stakes workflows, where it doesn’t, and what has to be built around it before anyone should trust what it produces.
Key Learning Objectives:
- Distinguish the parts of a clinical research workflow that genuinely call for AI from the parts that should stay deterministic and understand why handing an already-solved problem to AI introduces risk for no benefit
- Apply the design pattern behind AI eSource Builder — many small, evaluated steps rather than one large prompt, with monitoring at each stage — to other document-heavy, high-stakes workflows on both the site and sponsor side
- Identify what has to be in place before AI output can be trusted in a regulated setting. This includes review-and-approve gates, audit trails, and monitoring that catches reviewers who start accepting everything. Understand the vendor diligence that belongs underneath it, including enterprise model contracts and terms of service that prohibit training on your data, and clear answers on where study data goes and who can see it