You've Been Spending Recruitment Money at the Wrong End of the Funnel

Anyone running an oncology trial is intimately familiar with under-enrollment, a tale as old as time. More than 20% of cancer trials fail to hit their enrollment targets,¹ with insufficient accrual pegged as a leading reason why studies terminate early.² The industry's standard response? Spend more on recruitment itself: more sites, advertising, third-party recruitment vendors, and (now) artificial intelligence (AI). None of this spending is worthless, but almost all of it is based on a single assumption: that the patients are somewhere out there and we simply have to do more to get to them. This piece argues that same assumption is only half right—and the wrong half is exactly where the money earns the least.

Where the evidence points

In the largest meta-analysis to date examining why patients don't enroll, the single biggest reason isn’t that they refuse to participate but instead the lack of a suitable trial available at the place they receive their care (comprising ~56% of the shortfall). Another ~22% is patients who fail to meet eligibility criteria for trials existing nearby. The group who’s genuinely invited but goes on to subsequently decline is the smallest slice of all.³ In fact, a separate meta-analysis found that when patients are actually offered a trial, more than half ultimately participate.⁴

Putting these numbers together leads to an uncomfortable conclusion: the place your recruitment funnel leaks the worst isn’t at the bottom (persuading a patient to say “yes”) but instead at the top (the patient and trial never meeting in the first place). No amount of spending at the bottom can overcome this.

Your sites themselves aren’t leaking

For a cancer patient in the US, the route to learning about the existence of a trial runs almost entirely through one channel: the treating physician. Since no physician can search the thousands of constantly revised trials running nationwide, they refer to those they already know inside their own institution and network. The corresponding consequence? The set of trials a patient is exposed to gets compressed down to a small slice of one doctor's field of view.

Your trial might be a perfect fit for a patient three states away but effectively doesn’t exist at all if his or her physician can't see it. Another recruitment vendor cannot solve this problem either, because the problem isn't that you aren't pushing hard enough but that the information never flows where it needs to go.

Your best trials, hidden within the registry

A patient's self-service backup channel is the public trial registry, indexed mainly by a "tumor type" field. You know better than anyone, though, that modern oncology trials are increasingly organized around molecular targets via basket trials and tumor-agnostic designs—meaning a single KRAS trial might recruit patients with pancreatic, colorectal, or lung cancer (for example).

Consider a pancreatic cancer patient who searches "pancreatic cancer." In this case, he or she often won’t even surface the KRAS trial that fits since the tumor-type field does not—and cannot—enumerate every cancer for which that target appears. In other words, the most cutting-edge trials you design are (ironically) the ones most likely to be invisible in search. Part of the budget you pour into recruiting for a tumor-agnostic trial goes to fighting the registry's own indexing.

Where AI actually belongs

While AI is speeding up parts of recruitment—defining cohorts, streamlining pre-screening, and optimizing the enrollment workflow—nearly all of this effort is dedicated to the "help the trial find patients more quickly" side of things. That’s all well and good, but it assumes the patient is already inside your visible range and simply in need of quicker processing. The real loss happens earlier, when the patient never entered anyone's visible range at all. Throw all your AI efforts into accelerating the people you can already see, and you've installed a quicker machine at the bottom of the funnel while the hole at the top hasn't moved an inch.

What to actually do

The commonality across these recommendations is simple: move effort from the bottom of the funnel up to the information layer at the top by doing the following:

1. Reweighting your recruitment spend. Before the next budget goes to another vendor or advertising round, ask yourselves: Are the patients we lose mostly "reached but not persuaded" or "never met at all"? If the latter (as strongly suggested by the data for most trials), then every dollar spent on making trials more discoverable returns more than a dollar spent on boosting the persuasion rate.
2. Writing registry entries for discoverability. Registries index by structured fields, but what your trial is really looking for lives largely in the free-text description. Be deliberate here, making your trial description and eligibility criteria state the molecular target and cross-tumor applicability clearly enough to be parsed without allowing the narrow tumor-type field to speak for you. This matters most for tumor-agnostic trials; if your entry lists only a handful of tumor types, you're shutting out eligible patients with your own hand.
3. Treating eligibility criteria readability as a recruitment tool, not a compliance document. It’s simple: patients can't read criteria written in the professional language of study coordinators and thus can’t tell whether they qualify. Offering an easy-to-comprehend version (that doesn't replace formal screening) gets patients and their physicians to the screening conversation earlier on as better-informed parties, directly reducing top-of-funnel loss.
4. Having AI make the invisible visible, beyond processing what’s visible more quickly. The same natural-language capability you'd use to optimize existing list screening can instead rebuild the ability to match patients per what a trial is actually looking for—recovering the matches structured fields hide, AI's real (and still-untapped) recruitment lever.

Regulatory tailwinds

Regulators are moving in the same direction. Not only did the FDA roll out new information on boosting clinical trial participation (updating expectations for eligibility criteria, enrollment practices, and trial design) back in December 2025,⁵ but in late April 2026, the agency announced a real-time clinical trials initiative: two proof-of-concept trials, both in oncology, designed to report endpoints and signals to the agency on a continuous basis. Joining this was an RFI for a proposed pilot program testing how AI-enabled technologies can make efficiency, speed, and quality of decision-making better in early-phase trials.⁶ Regulators have already taken the premise here—that clinical research is bottlenecked by information more and more, with data methods apt to loosen the same—and applied it to how trials are run. The next logical step is to apply this same idea to how trials are found by patients, to how trials are found by patients—and moving first confers an advantage.

The bottom line

In the end, this is a judgment about return rather than being about goodwill. The industry has long absorbed the cost of under-enrolled trials, prematurely terminated studies, and the huge cost of delays. Its default reaction, likewise, has been to throw resources at the bottom of the funnel. On the contrary, the data says the largest leak is at the top before patient and trial ever meet. Whoever shifts efforts first—from "reach harder" to "make it easier for patient and trial to meet"—is the first to solve a perennially misdiagnosed problem. What needs fixing is the connection layer—not another, faster machine bolted to the bottom of the funnel.

About the author

Yuchen Gan holds a Master of Science from Columbia University and works on risk and data systems in regulated finance, where she builds explainable, rule-based systems for high-stakes decisions. She independently researches clinical trial discovery and matching.

References

1. Tran G, Harker M, Chiswell K, Unger JM, Fleury ME, Hirsch B, Miller K, d'Almada P, Tibbs S, Zafar SY. Feasibility of Cancer Clinical Trial Enrollment Goals Based on Cancer Incidence. JCO Clin Cancer Inform. 2020;4:35-49. doi:10.1200/CCI.19.00088.
2. Stensland KD, McBride RB, Latif A, Wisnivesky J, Hendricks R, Roper N, Boffetta P, Hall SJ, Oh WK, Galsky MD. Adult Cancer Clinical Trials That Fail to Complete: An Epidemic? J Natl Cancer Inst. 2014;106(9):dju229. doi:10.1093/jnci/dju229.
3. Unger JM, Vaidya R, Hershman DL, Minasian LM, Fleury ME. Systematic Review and Meta-Analysis of the Magnitude of Structural, Clinical, and Physician and Patient Barriers to Cancer Clinical Trial Participation. J Natl Cancer Inst. 2019;111(3):245-255. doi:10.1093/jnci/djy221.
4. Unger JM, Hershman DL, Till C, Minasian LM, Osarogiagbon RU, Fleury ME, Vaidya R. "When Offered to Participate": A Systematic Review and Meta-Analysis of Patient Agreement to Participate in Cancer Clinical Trials. J Natl Cancer Inst. 2021;113(3):244-257. doi:10.1093/jnci/djaa155.
5. US Food and Drug Administration. Enhancing Participation in Clinical Trials: Eligibility Criteria, Enrollment Practices, and Trial Designs — Guidance for Industry. December 2025.
6. US Food and Drug Administration. FDA Announces Major Steps to Implement Real-Time Clinical Trials [press release]. April 28, 2026; AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program; Request for Information. Fed Regist. April 29, 2026.