2026 SCRS Global Site Solutions Summit: What Does Al-Assisted Patient Matching Look Like?

Kyle McAllister, co-founder and CEO, Trially, identifies transparency as the defining characteristic of good AI-assisted patient matching.

He describes a system built so that when AI reads a protocol and scans patient charts for trial fits, its reasoning is visible rather than opaque. He illustrates this with a recent study where familial hypercholesterolemia was a key inclusion criterion, information often noted only once at a patient’s first doctor visit and then buried in a chart for 15 years. Their AI located that detail, a finding genuinely hard for a human to independently verify, which is exactly why transparency matters: without it, a correct match can look like an error simply because it’s difficult to confirm.

By showing where and how the information was found, the system lets reviewers see the match was accurate rather than dismiss it. He contrasts this with earlier-stage startups essentially wrapping a general-purpose chatbot around the problem, arguing real accuracy instead requires a team with deep understanding of specific therapeutic areas and how a protocol’s sponsor-written language translates into how physicians actually talk.

Asked about structural barriers beyond funding, McAllister points to regulation around electronic medical record APIs. While regulation requires EMRs to provide an API for external vendors, there’s no equivalent requirement, or at least no well-enforced one, that those APIs actually function properly. Trially spends significant time working to connect into EMR systems, and inconsistent integration capability remains a recurring obstacle to helping sites leverage their own data for trials. He frames this as a problem regulation, not technology, needs to solve: knowing which vendors integrate well is valuable operational knowledge his team has built, but the underlying standards themselves need to catch up and mandate functional interoperability, not just the formal existence of an API.