About three-quarters of clinical trial sites now use eSource in active studies, according to a 2024 RealTime Reports survey, yet the panel that CRIO and Clinical Leader convened last month found sites still wrestling with a more basic question: do you buy AI tools, or do you build them? The honest answer from operators who are already running AI day-to-day is that the question itself is premature for most sites.
The panel, moderated by CRIO Chief Innovation Officer Mike Wenger, included Aneesh Vaze of Clinical Research Philadelphia, Sam Stein of ALSA Research, and Nick Spittal of Velocity Clinical Research. Their consistent advice: sequence matters more than platform choice. Sites still running paper-based records need to move to electronic systems first, establish a basic AI governance policy, and give staff enterprise LLM accounts before touching anything custom-built. Skipping that foundation and going straight to homegrown tooling tends to create more cleanup work than it eliminates. For CRIO customers specifically, backend data access through Google BigQuery and Looker makes it possible to layer financial forecasting and custom dashboards directly onto operational data without re-entering anything, which is roughly the infrastructure a site needs before custom AI workflows are worth attempting.
Where AI is already earning its place, the applications are narrow on purpose. Drafting and updating investigator CVs came up as a concrete example: using an enterprise LLM account, describing the manual process clearly, and iterating through rough drafts produced faster results than trying to automate the whole workflow in one pass. A human reviewed every output. That pattern, small scope, human in the loop, dominated the panel’s practical recommendations. The failure mode runs in the opposite direction: automations that ingest documents, chat logs, and system data simultaneously while producing emails, reports, and dashboards in one step are harder to trust and harder to repair when something breaks. Inclusion and exclusion criteria were flagged specifically as a poor fit for current AI tools, since the criteria vary study-to-study and inconsistency in the input reliably produces incorrect interpretations in the output.
ICH E6(R3), which does not contain explicit AI governance requirements but provides a risk-based framework meant to accommodate new technologies, puts the documentation burden squarely on sites to justify whatever tools they use. That regulatory reality makes the panel’s sequencing advice more than operational preference: a site that deploys AI before its data infrastructure is clean has limited ability to demonstrate that its outputs are auditable. The concrete marker to watch is whether a site can answer, for any AI-assisted output it produces, exactly which data fed the model and which human reviewed the result.
Source link: https://clinicalresearch.io/blog/the-site-ai-playbook-your-top-questions-answered/
Moe Alsumidaie, MBA, MSF, is founder and Chief Editor of Vanguard Publications, which publishes Clinical Trial Vanguard, Pharma Vanguard and BullScope, and Head of Research at CliniBiz. He has two decades in clinical trial operations and data science, with earlier roles at Genentech, Abbott Vascular and Stanford University Medical Center, and is a guest lecturer in clinical trial sciences at Rutgers University.
