- David Jones
- Jan 14
- 5 min read
Updated: Feb 18
Insights from Paulius Ojeras at COG Bay Area – Redefining Efficiency, Quality, and Oversight with AI Teammates
In clinical trial outsourcing and clinical operations, the rise of artificial intelligence is reshaping how sponsors, CROs, and biotech firms approach vendor selection, quality oversight, and operational efficiency. This article distills practical insights and lessons from a COG Bay Area session featuring Paulius Ojeras, Head of Clinical Operations at Perceive Biotherapeutics, focused on pragmatic adoption of AI in clinical study delivery.
The session revolves around practical adoption of artificial intelligence (AI) solutions—from automation and efficiency gains to quality management and oversight in sponsor-driven trials. Leaders from across the clinical development ecosystem attended, seeking actionable frameworks for advancing their outsourcing models and minimizing common pitfalls.
AI Adoption: Finding the Balance Between Technology and Operations
Clinical operations remain highly human-intensive, despite advances in digital technology. Paulius Ojeras opened the session by highlighting the steep challenges faced in scaling and optimizing clinical outsourcing, especially under mounting regulatory expectations and complex trial designs.
A growing industry metric—the Augmented Intelligence Coefficient (AIQ)—measures how effectively organizations blend AI into human workflows. However, as Paulius Ojeras noted, “the AQ score in clinical operations in general it’s close to zero as of the moment.” This disconnect means that, while AI holds promise, the sector’s operational guardrails must be clear and robust to ensure real-world utility.
Five Levels of AI in Clinical Operations
To clarify the rapidly evolving AI landscape, Paulius Ojeras described a framework for assessing solutions by complexity:
Pure Automation: Rule-based triggers; minimal complexity.
Predictive Tools: Machine learning models digesting large datasets, with humans making decisions.
Smart Assistance: Contextual AI suggesting actions, still requiring human validation.
Agentic AI: Autonomous teammates capable of both planning and execution, requiring only human oversight.
Cooperative Agentic AI: Groups of agentic AIs collaborating—akin to a virtual CRO team.
Most biotechs and CROs are currently experimenting between levels 1–3, leveraging AI as an assistant rather than a true autonomous partner. Levels 4–5 remain frontier territory, but “the real revolution is unfolding” in how these agentic systems could transform outsourcing, scalability, and quality.
Rethinking Clinical Outsourcing Models with Agentic AI
Traditional approaches to clinical trial outsourcing typically fall into one of three categories: contracting with a CRO, building in-house execution teams, or piecing together expert contractors. Paulius Ojeras emphasized that the agentic AI model introduces a “fourth way”—deploying AI teammates to execute individual or grouped tasks.
Key operational advantages highlighted include:
Scalability: AI can process vast, parallel workloads, solving bottlenecks in study startup and document management.
Speed: Automated systems deliver instant, rule-consistent results.
Quality: Reduced human error, with AI reliably flagging inconsistencies or missing data.
Yet, for many clinical leaders, doubts persist around vendor hype and unclear definitions of what “AI solutions” actually do. Paulius Ojeras cautioned against “just another sales pitch,” underscoring the need for sponsors to critically vet solution providers and their technical capabilities.
Real-World Case Study: Transforming TMF Management with AI
To illustrate principles in practice, Paulius Ojeras shared a detailed case study from Perceive Biotherapeutics on trial master file (TMF) management—a process notorious for document backlogs and resourcing challenges.
Situation: After phase I, the chosen CRO was underperforming, prompting a shift to an in-house operational model. TMF management emerged as an optimal area for testing agentic AI, given its rules-driven, repetitive nature and the ability to revert to manual backup in case of system failure.
AI-driven TMF workflows achieved:
Filing Accuracy: When AI refiled phase I documents previously QC-checked by humans, 25% had issues missed by manual review—ranging from missing dates to incorrect file locations.
Efficiency: Real-time feedback on missing or erroneous documents, enabling immediate resolution by site teams.
Scalability: Seamless, concurrent handling of large document volumes—effectively eliminating typical startup-phase bottlenecks.
Quality by Design: AI generated dynamic expected document lists, automatically created placeholders, tracked expiration dates, and prompted missing items.
Importantly, this approach did not just automate individual steps—it fundamentally reimagined the process. “Paulius Ojeras” described it as moving from workflow automation to “fully automated, quality-by-design operations, with humans supervising rather than intervening.”
Essential Guardrails for AI Integration in Clinical Trials
While the benefits are significant, Paulius Ojeras was clear: successful AI adoption hinges on disciplined, thoughtful approaches to technology, vendor selection, and internal readiness.
Questions to Assess Vendor Readiness
Sponsors and clinical ops teams should rigorously evaluate AI vendors using criteria such as:
Is the solution “ready to use” or custom-built?
Are live deployments available—or just pilots and demos?
What is the vendor’s financial runway (stability)?
Can you trial the system before committing?
How is the AI team trained, and what safeguards exist (e.g., mitigation of hallucinations and assurance of consistency)?
According to Paulius Ojeras, currently “95% of the AI companies are crashing,” making due diligence on vendor stability and clinical research expertise vital.
Consistency & Oversight in Highly Regulated Environments
Consistency of outcomes—especially with autonomous AI agents—is both difficult and essential in regulated settings. Robust validation (e.g., 98% concordance across independent agents for TMF management tasks) is needed before entrusting high-stakes deliverables.
Additionally, sponsors must rethink process flows for end-to-end automation, ensure clear accountability for AI-driven decisions, and maintain “critical analytical curiosity” in reviewing outputs.
Training, Change Management, and Human–AI Collaboration
As the audience remarks and follow-up questions highlighted, success isn’t just about technological capability. Internal training, prompt engineering, and cross-functional collaboration are core to unlocking AI’s full potential.
Prompt Engineering: Teams must learn to write clear, relevant prompts to guide AI tools. “Garbage in, garbage out”—the quality of input determines the utility of output.
Change Management: Adoption should be matched by education efforts, with clinical staff encouraged to engage critically with AI-generated results.
Collaborative Implementation: Vendors should support sponsors by co-developing prompts and processes, especially where clinical or scientific nuance is required.
What This Means in Practice
Rigorously vet AI vendors for both technical and clinical operations expertise.
Demand live deployments and real-world references—avoid decisions based on demos or pilots alone.
Prioritize fully validated solutions that minimize human intervention and maximize consistency.
Build internal capacity for prompt engineering and AI literacy through staff training.
Rethink process design—look for opportunities to automate from end-to-end, not just at the task level.
Ensure clear accountability structures for AI-driven actions, especially in regulatory or safety-critical workflows.
Treat AI integration as a change management challenge: support adoption with education and transparent oversight.
Key Takeaways
Agentic AI offers transformative potential for clinical trial outsourcing, especially in scalability, quality, and speed.
True impact comes from reimagining processes—not just automating existing steps.
Consistency, validation, and oversight are critical; poorly managed AI solutions risk adding complexity, not efficiency.
Successful adoption hinges on strong vendor selection and pragmatic operational guardrails.
Internal education and change management must keep pace with technology advances to ensure meaningful, reliable outcomes.
Selected Quotes
“Clinical operations remain very human-intensive despite the exploding complexity, really unsustainable timelines and mounting regulatory… the AQ score in clinical operations in general it’s close to zero as of the moment.” — Paulius Ojeras
“Where agentic AI would be making the greatest impact is really the activities which require high productivity, high speed and scalability… any workflows which we have well described could be fully automated.” — Paulius Ojeras
“Don’t buy the pilots, demos or slides. There is many of those currently in the industry because everyone wants to jump into this hype… look for references for exactly specific solution, not just the name. And try it before you buy it.” — Paulius Ojeras
“We are so used to buying the tools which creates an extra level of people which will need to set up, manage and supervise… versus buying the real solutions.” — Paulius Ojeras
The PBC Group and COG event series: https://thepbcgroup.com
For further curated sessions from the COG Meeting series, visit the above link for agendas, blogs, and live event information.
Comments