Across the industry, we’ve normalized a “complexity tax” in clinical development, treating friction as if it’s an unavoidable operating cost. Before the familiar downstream problems emerge (compressed timelines, global coordination burdens, shifting protocols) clinical development faces a more fundamental challenge: finding the right patients for the right trials at the right moments. When enrollment suffers, so does everything that follows.
Research from the Tufts Center for the Study of Drug Development (CSDD) has reported that 76% of Phase I–IV protocols now require at least one amendment, up from 57% in 2015, with costs per amendment ranging from $14,000 to $535,000. A large share of those costs trace directly back to eligibility assumptions and enrollment pipelines that fail to reflect how real patients present in everyday care settings.
Trials are dynamic systems, not linear projects
Most trial operating models still assume a predictable execution: finalize a protocol, activate sites, enroll patients, lock the database, read out results. The reality is messier. The industry has gotten good at generating insights after the fact, but struggles to continuously coordinate action while a trial is in motion. Signals often arrive late, in separate systems, and require scarce analytic resources to interpret. We need a different approach, one that moves from static planning and reactive management to continuous orchestration, starting with identifying the right patients.
From “insight” to “action”: what agentic AI actually changes
Agentic AI refers to systems that can interpret goals, plan actions, and execute across tools and workflows, adapting as conditions change. Traditional AI generates insight: identifying what is happening and why. Agentic AI operationalizes and orchestrates that insight: determining what should happen next, routing work to the right owners, and learning from outcomes. While technology will certainly play an essential role in optimizing the clinical development process, it can’t be treated as a “set it and forget it” capability. In healthcare, this must be introduced with human oversight and full auditability full stop.
That orchestration capability is what unlocks enrollment efficiency, but only if the underlying data can support it. Agents alone cannot close the enrollment gap, and require access to real-world data that meets three bars: broad enough to reflect the diversity of sites and patient populations that actually exist; deep enough to include biomarkers, treatment history, and clinical detail beyond administrative records; and current enough (ideally updated weekly or better) to reflect where a patient is in their care journey today, not six months ago. Without that data foundation, you are optimizing a model against assumptions, not reality.
There is also a distinction the industry tends to collapse: the difference between a patient who is eligible for a trial and a patient who needs one. A patient may meet every inclusion criterion on paper, but if their current therapy is working, they will not — and should not — choose a trial. Matching on eligibility alone misses this. What is required is the ability to identify patients who are both eligible and facing a genuine gap in their current care: patients whose diagnosis, test results, and response to treatment suggest they may need a different path, and for whom a clinical trial represents a clinically meaningful option. That requires a real-time view of where each patient stands in their disease course, combined with predictive modeling that anticipates when their current therapy may no longer be sufficient. This is what separates a matching tool from a true clinical intelligence capability.
Agentic AI changes the way teams confront these issues by turning feasibility into an iterative, evidence-driven loop rather than a one-time gate. Sponsors and their partners can stress test protocol assumptions earlier using broader signals such as real-world evidence, operational history, and trial performance patterns drawn from datasets spanning millions of patients across diverse care settings. The result is eligibility criteria grounded in reality rather than projection, and fewer expensive amendments downstream.
How agentic AI changes operations
Beyond trial design, complexity persists in clinical management. Small delays compound: a site that struggles with staffing becomes an enrollment issue, then a timeline issue, then a budget issue, then an amendment issue. Agentic AI is well suited to coordinate this complexity at scale by connecting three capabilities that are too often separated:
- Early risk detection: Agentic systems monitor signals such as enrollment dips, site performance gaps, and activation delays to spot trends before they escalate and route interventions to the right owners.
- Adaptive intervention planning: Beyond flagging risk, these systems recommend actions and reduce the latency between signal and response.
- Continuous optimization: Agentic systems learn from intervention outcomes and update forecasts, site expectations, and resource plans as conditions change.
For CROs, this shifts differentiation toward measurable outcomes rather than staffing intensity. For sponsors, it means speed and confidence in site selection with fewer late-stage surprises. For patients, particularly those in historically underrepresented communities, it means the opportunity to be seen, matched, and enrolled in a therapy they might otherwise have been overlooked for.
Trust, governance, and auditability aren’t optional
If agentic AI is going to become the coordination layer for clinical development, trust is mandatory. In regulated environments, systems must be designed with consistent accountability and audit trails, even as automation increases. Human-in-the-loop oversight is not optional in healthcare. Teams must be able to understand why a recommendation was made, what data informed it, and what alternatives were considered. In patient-facing workflows like trial matching, AI-generated eligibility assessments require explainability, auditability, and safeguards against systemic bias that could unintentionally exclude certain patient populations.
Our industry imperative: keeping the patient at the center
Faster trials are not a vanity metric. They reflect an undeniable need: getting safe therapies to patients sooner. The organizations that succeed will be the ones that leave static plans behind and embrace continuous orchestration grounded in real-world data, powered by predictive intelligence, and designed from the start around patient need, not just patient eligibility.
Photo: metamorworks, Getty Images
Eron Kelly is Chief Executive Officer of ConcertAI, where he leads the company’s mission to accelerate clinical research and improve patient outcomes through AI-powered insights. Prior to ConcertAI, Kelly held roles at Inovalon, Amazon Web Services, Microsoft, and the U.S. Air Force, driving innovation at the intersection of healthcare, data, and enterprise technology.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.
