Fixing Clinical Trial Matching - From Bottleneck to Breakthrough

AI-Supported Clinical Trials Matching_April2026.pdf

Download PDF • 280KB

A Conversation between Rebecca Clyde, PhD, Botco.ai, and Henning Mann, PhD, HM.BioConsulting

Authors: Rebecca Clyde, Desiree Goubert, Henning Mann

Clinical trial matching is entering a new phase where advanced AI is embedded directly into clinical workflows, enabling a meaningful shift in speed and accessibility. In a recent discussion with Rebecca Clyde, Co-founder and CEO of Botco.ai, we discussed how this approach connects patients, physicians, and trial sponsors while reducing the time required for matching from several hours to roughly one minute. OncoTrial Match is currently part of an IRB study at the University of California health system, organized by a team at the UCLA David Geffen School of Medicine.

Rebecca Clyde:OncoTrial Match™ is an AI clinical trial matching service for doctors. Patients who are not getting what they need from existing treatments want to join trials, but it’s difficult to find appropriate ones within their radius and based on their disease and biomarkers. On the other side, pharmaceutical and biotech companies need patients which are the right fit for their trials, a significant and critical challenge for the success of substantial investments of dollars and time into those trials. Physicians sit in the middle, but they don’t have the time to do the matchmaking. It can take three to four hours per patient.

The idea was: what if AI could reduce that to a 60-second effort? We worked with the UCLA oncology team to design the experience. They told us it has to be mobile, usable while walking between appointments, take less than a minute, and automate the referral process.

As AI systems are increasingly applied in real-world clinical environments, their effectiveness is defined by how well they align with the physical and cognitive workflow of physicians. Clinical trial enrollment remains one of the biggest bottlenecks in drug development, with many eligible patients never identified. This raises the obvious question about who the system is designed for.

Henning Mann: Is your solution directed to the patient’s side or the physician’s side?

Rebecca Clyde:This is primarily for the physician’s side. Doctors need something simple, mobile, and fast that automates referrals. They don’t want to go back to a desk and handle emails or paperwork.

The system also has to match patients based on trial criteria and willingness to travel.  So, we built a simple AI agent where doctors enter a few points, mostly button-based, and immediately see trial options. Once they select one, the referral is automated.

Henning Mann:With physician adoption dependent on speed and simplicity, the reliability of underlying data becomes critical, especially given the fragmented and hard-to-navigate nature of trial databases. Where does the trial information come from?

Rebecca Clyde:We curate it from clinicaltrials.gov, that platform is what physicians currently spend hours navigating.

We take that data, restructure it, and combine it with our predictive analytics agent. Instead of manual searching, the system matches based on a few inputs and the patient profile.

Henning Mann:

It is surprising that this doesn’t exist yet. Transforming static registries into usable clinical tools reflects a broader trend in AI - making existing data actionable. This often starts with understanding real user behavior. How did you get the idea?

Rebecca Clyde: By talking to customers. We already had a patient support chatbot, and physicians told us they needed support too, specifically for referring patients to trials. That’s where we focused development.

Henning Mann:Integration into clinical environments becomes the defining factor to move solutions from concept to practice. Usability within real workflows determines whether adoption happens at all. How does your solution integrate into existing hospital systems and physician workflows without adding friction? What is the threshold for doctors to just start using it?

Rebecca Clyde:Many physicians - especially surgeons - don’t have a dedicated workspace at the hospital. They’re either in surgery or doing rounds, moving from room to room. They use shared devices, log in, log out, and anything they were working on is gone. So, if they’re in the middle of researching a clinical trial, they can’t reliably come back to it.

Additionally, everything is happening through email, and the reality is they only have one to two minutes between patients, sometimes literally walking from one room to another or heading to another hospital, running from clinic to clinic, taking care of patients, and anything new has to be squeezed into that.

That’s also why doctors often don’t go along with many technical implementations, they’re too cumbersome and don’t reflect how physicians actually operate in a clinical setting. Especially for surgeons, systems are rarely designed around how they physically move through their day.

So, we designed this to go with the doctor, wherever they are. It sits in their pocket. They can quickly text something like: “a patient with recurrent ovarian cancer, second line of therapy, platinum sensitive”, and our solution will show suitable trials within 50 miles almost instantly. And that’s it.

It had to be mobile, usable while walking, and independent of a laptop.

Henning Mann:Once embedded into high-pressure clinical settings, trust becomes essential. Matching systems must deliver both speed and precision to support confident decision-making. Physicians are already overworked and under pressure. How do you ensure accuracy in matching? It matters which trial gets assigned.

Rebecca Clyde: We’re tracking two key metrics: referral rate and accrual rate.

Referral rates are currently very low,often under 5%,because doctors don’t have time. We aim to significantly increase that. At the same time, trials often only reach 10–20% of their enrollment goals.

Accuracy is critical, so we focus on up-to-date data and precise matching using clinical criteria and geography. We narrow results based on what patients are realistically willing to travel.

Henning Mann:As AI systems scale in clinical use, continuous improvement becomes central. Performance gains increasingly depend on feedback loops from real-world usage. With AI systems now operating directly within these constrained environments, continuous improvement becomes a defining feature of performance and usability. Does the system learn over time from referral outcomes and trial enrollment data? Would that even make sense?

“The key change was restructuring the data around how physicians think, not how the FDA organizes it. Clinicaltrials.gov reflects regulatory reporting, but physicians evaluate trials differently.”

— Rebecca Clyde, Co-founder & CEO, Botco.ai

Rebecca Clyde:Yes. Already during the pilot phase, we have made significant improvements to the chatbot.

We worked closely with oncology teams to understand how they as doctors from varying levels of experience think about clinical trials, and how they interact with our system. That meant interviewing them, watching them in their setting, and understanding how they mentally go through the process of referring patients into trials. We also received direct notes and feedback about the system.

We observed how they search on clinicaltrials.gov and extracted the most important parameters that needed to be part of the interaction to serve up the best trial in the least possible time. Further, we met with various oncology consortiums across different health systems to test and collect feedback during the early days of the implementation. This ensured that we had a wide range of input going into the rollout.

Because those were the two constraints: "accuracy and time”, it has to be precise, but it also has to happen quickly and with just a few questions. So, the focus became: what are the minimal questions that consistently lead to the best possible match?

Henning Mann: Refining outputs over time requires a strong technical foundation, where architecture determines how effectively systems retrieve, structure, and apply complex clinical data. How would you describe the core architecture of the system?

Rebecca Clyde:We built our own RAG system and a dedicated database derived from clinicaltrials.gov, refreshed daily. Retrieval Augmented Generation, or RAG, is a way to make AI models more reliable by letting them pull in information from external, trusted sources such as clinicaltrials.gov. This means it does not just depend on what the model learned during training, it first looks up relevant information, adds that context to the user’s question, and then generates an answer based on this information.

The key change was restructuring the data around how physicians think, not how the FDA organizes it. Clinicaltrials.gov reflects regulatory reporting, but physicians evaluate trials differently. We had to flip that entire structure to make it usable for clinical decision-making.

Henning Mann:With matching becoming more efficient, its effects extend beyond clinicians to sponsors, influencing how trials are planned and executed across development programs. With improved matching efficiency and broader applicability, the discussion naturally expands to how this may shape trial execution and sponsor strategy. Do you see this influencing how sponsors design trials, knowing that matching and recruitment can be more efficient?

Rebecca Clyde:Companies invest significant resources into trials, and recruitment is a key factor in whether trial progresses. Improving matching and referral supports faster enrollment and helps trials reach their targets.

As matching becomes faster and more precise, sponsors gain greater visibility into where eligible patients are located and how quickly they can be identified and referred. This can support more informed decisions around site selection, geographic distribution, and inclusion criteria. At the same time, improved referral flow increases the likelihood that trials reach their enrollment targets within expected timelines, which is directly tied to overall development progress.

Henning Mann:Even with improved matching, patient access remains uneven. Geography and travel constraints continue to shape participation in clinical research. What about patients in rural areas?

Rebecca Clyde:In rural markets, patients face some necessity to travel. We are working with patient advocacy groups that are helping us take this capability to the mid-west where a more rural population will be serviced. We're collaborating with them on programs that can ensure patients outside of large metro areas can also be referred to clinical trials and gain access to them. 

Henning Mann:In distributed and complex environments, safeguards are essential. Systems must integrate validation layers to ensure appropriate trial assignment. How do you prevent errors, given that physicians rely on the system?

Rebecca Clyde: There are multiple safeguards. Even if a mismatch occurs, the trial site ultimately determines eligibility. There are always humans in the loop, physicians and principal investigators, so that inappropriate referrals are filtered out before enrollment.  Additionally, multiple guardrails around privacy, security and accuracy were implemented. First, we had to build a system that never compromised the patient's privacy. That means that the trial matching occurs without needing to expose a patient's personally identifiable information. When it comes to guardrails around accuracy, we iterated extensively with the clinical teams to ensure a high confidence level with every referral.

Henning Mann:Reliable matching feeds directly into enrollment performance, which remains a key determinant of trial success and overall development timelines. Recruitment is a major risk factor for companies.

Rebecca Clyde:Exactly. Companies invest tens or hundreds of millions into trials. If they can’t recruit enough patients, the trial may fail before generating results. Our goal is to increase qualified referrals so trials can reach enrollment targets faster and reduce that risk.

Henning Mann:Handling sensitive patient data, privacy and compliance is a foundational requirement, especially in regulated healthcare environments. How do you handle patient data and HIPAA requirements?

Rebecca Clyde:We avoid including direct patient identifiers in referral packets. Instead, we use proprietary, tokenized proxies. Trial sites can re-identify patients internally through their own systems. This allows us to stay compliant with HIPAA and privacy requirements. Additionally, of course, the privacy guardrails we implemented assure the needed privacy, security and accuracy.

Henning Mann:As learning systems evolve, attention naturally shifts to how they handle the complexity of clinical trial criteria, where eligibility can be highly nuanced. Clinical trial criteria can be extremely complex - how do you handle edge cases or borderline eligibility?

Rebecca Clyde:If we present a trial and explain why we think it’s a match, the doctor can agree or disagree. If they disagree, they can skip it and see the next best match. We keep presenting the next best option.

If needed, we can expand the parameters, such as increasing the geographic radius, based on input from the physician. We intentionally keep the initial criteria narrow to find the best possible match.

Henning Mann:Once your solution demonstrates reliability in one domain, the same architecture can extend into additional therapeutic areas where similar matching challenges exist. Do you see this model expanding beyond oncology into other therapeutic areas?

Rebecca Clyde:Neurology will be our next area of focus. I’m fielding a lot of requests from neurology trial sites, both for multiple sclerosis trials and Alzheimer’s trials. Pediatrics and rare diseases are also on the roadmap.

Henning Mann:With a growing data infrastructure, new applications may emerge, including broader connections between patients, trials, and research ecosystems. There may be an opportunity to also extend this into research sample allocation.

Rebecca Clyde:That’s an interesting idea. In parallel, we’re addressing another major issue: patient attrition. About 40% of patients drop out of trials.

Additionally, we also built a second product called AskGRACE—Guided Recommendation Assistant for the Cancer Experience. It focuses on keeping patients engaged during trials. 

Henning Mann:This is very interesting. Sustained participation is as critical as initial matching, placing greater emphasis on continuous engagement throughout the trial journey. How does AskGRACE work?

Rebecca Clyde:While OncoTrial Match™ is for physicians; AskGRACE is for patients. It checks in daily, supports adherence, answers questions, provides resources for side effects, and escalates serious issues to the clinical team. The idea is to both increase adherence and ensure patients stay in trials through completion.

Henning Mann:All of this is very impressive! As solutions demonstrate clinical and operational value, deployment and scaling become the next step, supported by funding and institutional adoption - how are you progressing with fundraising?

Rebecca Clyde: We’ve raised nearly $6 million so far through a seed round and extension. We’re now deploying with large cancer centers and preparing for a Series A raise later this year.

Henning Mann:Thank you very much, Rebecca.

This discussion illustrates how advanced AI is adding game-changing efficiency to critical processes that are executed every day but which have long been carried out in time-intensive and often fragmented ways. This discussion highlights how clinical trial matching, a cornerstone of drug development and of literally vital importance to both sides, patients and drug developers, can be transformed by aligning intelligent systems with real-world workflows, reducing effort from hours to minutes while improving access and precision. By structuring complex data for rapid decision-making and supporting both physicians and patients throughout the trial journey, solutions like OncoTrial Match™ and AskGRACE demonstrate how efficiency gains translate directly into better trial execution. As these capabilities continue to mature and expand across therapeutic areas, their influence will extend further into trial design, enrollment strategies, and overall development timelines.