The clinical-trial bottleneck was never the science. It was finding patients. HealthKey (YC W25) points AI at the whole EHR - faxes and all - and only gets paid when someone enrolls.
A doctor's office is one of the strangest data warehouses in the world. Decades of MRIs, lab panels, encounter notes and referral letters - much of it still arriving by fax and living as scanned PDFs nobody will ever read again. Somewhere in that pile are patients who could get into a clinical trial today. Nobody has time to find them. That gap is the entire reason HealthKey exists.
HealthKey, part of Y Combinator's Winter 2025 batch, is an AI system that connects to a clinic's electronic health record, reads every patient's history, and checks it against the eligibility criteria of active clinical trials. The output is not a black box. It is a ranked list of likely-qualifying patients, each one accompanied by the specific evidence - an MRI finding, a lab value, a line in an encounter note - that a doctor can look at and verify in seconds.
The founder, Josh Sabol, did not come from medicine. He was a product manager at AWS working on supply-chain automation, and before that the first PM at Plate IQ, a B2B payments company that later raised a $160M Series B. What he brought to healthcare was a specific instinct: when skilled people spend their days doing slow, repetitive matching by hand, there is usually a machine that should be doing the reading.
Recruitment is the quiet killer of clinical research. A large share of trials miss their enrollment timelines, and many close having never found enough participants. The treatments are real. The eligible patients are real. They simply never meet, because the work of connecting them - reading a chart, cross-referencing a protocol, ruling a patient in or out - is done by humans, one record at a time.
HealthKey's wager is that the answer was already sitting in the chart. Every criterion in a trial protocol - an age range, a prior diagnosis, a specific lab threshold, a treatment history - can be checked against a patient's record. Do that for every criterion, across every patient, across every open trial, and a vetted candidate list falls out the other end.
That last step matters more than it looks. HealthKey does not try to replace the clinician's judgment; it does the reading and hands back the receipts. The doctor still decides who to approach. The difference is that the doctor is now looking at a short, evidence-backed list instead of an unsorted patient panel.
Here is the part that makes procurement easy. HealthKey charges success fees per enrolled patient. There is no upfront cost. If it doesn't find qualifying patients, the site doesn't pay. Pricing scales with sites, studies and subjects, which means a research team can try it without a budget fight and a large network can grow into it without renegotiating a license.
It is a rare kind of alignment. Most software bills you whether or not it works. HealthKey's revenue is tied directly to the outcome its customers actually care about - a patient in a trial - which is also the outcome a patient waiting for a treatment cares about.
HealthKey's customers are clinical trial sites, doctors' offices and their research teams. For a clinic, trial enrollment is a genuine revenue line that most practices never fully work, simply because identifying eligible patients is too labor-intensive. For a patient, being matched to a trial can mean access to a treatment that isn't otherwise available. HealthKey reports dozens of active, paying sites and a backlog of clinics and trials still waiting to onboard.
Patient-matching for trials is a contested space - names like Deep 6 AI, Paradigm, Mendel and the recruitment arms of companies such as Formation Bio all circle the same problem. The default competitor, though, is not another vendor. It is the status quo: a coordinator with a spreadsheet and a stack of charts. HealthKey's edge is less about claiming smarter AI and more about the shape of the offer - deep EHR reach including the messy fax layer, evidence handed to the doctor rather than hidden, onboarding in under a week, and pricing that only charges for results.
The company runs lean, with a team of around ten split between Austin, Texas and Chihuahua, Mexico. That it chose to live in the unglamorous parts of healthcare - eligibility criteria written like legal contracts, records trapped in PDFs, the fax machine that refuses to die - is a large part of why the pitch lands. The hard, boring reading is exactly the work HealthKey took on.
The mission Sabol describes is plain enough: match at-risk patients with cutting-edge treatments, and get them there faster. Whether HealthKey becomes the layer that clinical trial sites run on will depend on execution across thousands of protocols and millions of records. But the premise is hard to argue with. The patients were always there. Someone finally built the thing that reads all the charts.