Snapshot
Artificial intelligence in clinical trials is no longer only a theoretical discussion. In Suvoda’s recent webinar, Using AI to responsibly accelerate clinical trials under regulatory scrutiny, Priyanka Sharma, Senior Vice President, Software Engineering, moderated a discussion with Andrew McVeigh, Chief Architect, and Jadon Sargeant, Senior Product Owner for RTSM (randomization and trial supply management) and Sofia (Suvoda’s AI assistant), on how AI adoption is moving from experimentation toward implementation.
Clinical trials are complex, highly regulated, and built on trust. Sponsors need AI that can improve speed and consistency, but they also need confidence that the technology is safe, controlled, and appropriate for the clinical trial environment.
As Andrew noted during the webinar, AI technologies are “game changers,” but they must be designed carefully because clinical trials are “a highly regulated and absolutely critical industry.” That regulation exists for good reason. Clinical trials test experimental therapies in human patients, where mistakes can affect patient safety, data integrity, and the advancement of medical knowledge.
In clinical trials, the first responsibility is to patients and clinical staff. That is why Suvoda’s approach keeps humans in the loop for critical decisions and uses AI in ways that support expert judgment rather than bypass it. Andrew described this directly: “We have a principle of human in the loop. This is designed to augment people.”
That principle is especially important because large language models can be powerful but are not inherently deterministic—meaning they do not always produce the same answer or the correct answer. At Suvoda, our AI solutions are each built around a powerful deterministic core layer, which means the AI is not making decisions on its own or operating without structure. Instead, it works within defined workflows, rules, permissions, and data checks that produce predictable, traceable outputs. In everyday settings, users can often spot when an answer seems uncertain or incorrect. In clinical trials, that is not enough. AI needs the right safeguards, review steps, and auditability, so uncertainty is controlled by the system, not left for users to manage on their own.
The takeaway for sponsors is straightforward. Responsible AI depends as much on design boundaries as it does on capability.
Key principles include:
Another important theme was teachability. AI can move quickly, but in clinical trials it needs to be guided by people with domain expertise and depth of knowledge in this arena.
Jadon described the importance of combining the knowledge built over decades in clinical trials with technology that can synthesize information quickly. As he put it, “The knowledge comes from the humans, but the speedy execution comes from the AI side.” AI can handle repeatable, time-consuming tasks that don’t need humans.
That idea is central to Sofia, Suvoda’s AI assistant. Sofia uses a conversational interface, but it is not simply left to search freely across a large set of data. It maps user questions to a structured library of questions and expert-authored answer logic. Subject-matter experts can teach the system how to handle specific processes, and that knowledge can then be used consistently.
Clinical trials need pragmatic AI, clear guardrails, auditability, regulatory readiness, and must balance large language models with deterministic, expert-authored logic.
For Suvoda, those guardrails include reliability, teachability, and security. The speakers discussed part of what that entails: zero-day data retention, not training models on sponsor or user data, role-based permissions, and protections designed to prevent inappropriate data access.
Jadon gave a specific example related to blinding. If a user is in a blinded role, Sofia can only access the data points available to that role. If the user asks something that would unblind them, Sofia cannot access those data points. As Jadon explained, “It’s not like [the AI] knows those data points and is told not to tell you. It can’t access them.”
That distinction is important. In clinical trials, responsible AI should not depend only on the model’s response. It should be constrained by the system’s underlying permissions, study rules, and data protections.
The core guardrails include:
Sofia and agentic RTSM are examples of how these principles can work in practice.
Sofia helps users ask questions about RTSM data and workflows through a conversational experience. In one example from the webinar, Jadon described asking why a drug shipment failed. Instead of manually checking several possible causes, such as depot inventory, release status, expiration, or address issues, Sofia can follow expert-instructed checks and help surface the answer. That example illustrates two principles at once: the checks were authored by clinical trial experts (teachability), and Sofia runs them the same way every time a user asks that question (deterministic logic).
Agentic RTSM applies AI to one of the most operationally intensive parts of clinical trial startup: the study build process. RTSM systems manage critical trial functions such as randomization, drug supply, inventory, dispensing, and patient progression, which means study configuration must be both fast and carefully controlled. Andrew described how Suvoda is using AI to help translate expert-authored business logic into deterministic scripts, which are structured, rule-based instructions that can be reviewed, tested, and validated. In this authoring loop, AI helps accelerate the historically time-consuming configuration of the randomization and trial supply management system while keeping human review and validation in place. By reducing study build timelines by up to 80% and bringing studies from kickoff to UAT in as little as two weeks, agentic RTSM can help sponsors move trials forward faster without removing the oversight clinical trials require.
The capability is being used in an early adopter phase with real studies and customer feedback, rather than being developed only in isolation. The approach also reflects the deterministic core principle: the system does not interpret how to configure a study. It uses Suvoda's low-code/no-code configuration tools to execute precisely, producing the same output for the same inputs every time. And because auditability matters as much as speed in a regulated environment, every action the system takes is traceable—so sponsors and regulators can see exactly what was configured, when, and why.
Both examples reflect the same broader philosophy: AI should be used where it can reduce manual burden, improve consistency, and accelerate work, while preserving the controls required in clinical trials.
AI is becoming more familiar to people, more capable, and more expected in clinical trial technology. But the path forward is not simply to move faster. It is to move with the right structure around the technology.
For sponsors, that means asking practical questions. How is data protected? Where does human review occur? What can the AI access? How are permissions, auditability, and regulatory expectations addressed?
The webinar discussion made clear that responsible AI is not just a feature set. It is an approach to design. Sofia and agentic RTSM are examples of how Suvoda is applying that approach—building AI that earns trust through actions and choices that sponsors can inspect and defend. The larger principle is broader: clinical trial AI should reduce friction and improve experience, but it must also hold up to a sponsor's or regulator's hardest questions about what the system did, why, and how it was governed.
As AI continues to move from experimentation to operational use, the opportunity is significant. So is the responsibility. Clinical trial technology should use AI in ways that are useful, controlled, and grounded in the expertise of the people who understand clinical trials best.
Priyanka Sharma |
Andrew McVeigh |
Jadon Sargeant |
ON-DEMAND WEBINAR