The protocol landed on a Friday. By the following week, MotilityCount ApS had a validated electronic data capture system ready to collect patient data across 15 fertility clinics in what would become a 1,600-cycle randomized trial. Nine days, start to finish.
For context: industry standard runs closer to 10 or 12 weeks.
That Danish fertility trial is the kind of case study Harbor—a San Diego startup barely past its first birthday—keeps pointing to as it makes an audacious pivot. The company just signed a $1.93 million, three-year contract to run an entire clinical study for a medical device firm. Not just provide software. Actually run it, end-to-end, as a contract research organization.
The wager is straightforward, if ambitious: that owning the data infrastructure lets you automate nearly everything else—study build, monitoring, site support, the whole operational stack—and that speed compounds into a defensible edge. Whether a two-person team can actually deliver on a seven-figure CRO engagement is another matter entirely.
Harbor graduated from Y Combinator and announced the CRO expansion in a late-May launch post. The company currently supports seven clinical trials on its platform, generating around $15,600 in monthly recurring revenue from its software. Contracted bookings for the next twelve months sit at roughly $187,000. The newly inked CRO deal represents a different revenue model, and a different kind of risk.
Timing Is Everything
The announcement comes as the $82 billion contract research organization market scrambles to bolt AI onto decades-old infrastructure.
Medidata announced a decade of AI leadership in March 2026, highlighting generative tools designed to accelerate the notoriously tedious process of translating study protocols into live database builds. Parexel launched ParexelAI in mid-May. IQVIA unveiled its IQVIA.ai platform at NVIDIA's GTC conference. ICON partnered with Advarra's Braid AI system. Fortrea rolled out something called the FIT suite in April.
All of them are grafting machine learning onto systems built in another era. Harbor is trying the opposite approach: build the AI layer first, then wrap services around it.
What the Platform Actually Does
At its core, Harbor is an electronic data capture system—the software backbone that collects, stores, and manages patient data during clinical trials. The platform is designed for compliance with 21 CFR Part 11 (FDA's electronic records rule) and Good Clinical Practice standards, which is table stakes in this industry.
What sets it apart, at least according to Harbor's pitch, are two features that compress setup timelines dramatically.
The first is Magic Build. Upload a protocol PDF and the system auto-generates electronic case report forms, visit schedules, data monitoring events, edit checks, even paper source documents. The second is Magic Capture, which extracts data from uploaded source files—lab results, imaging reports, whatever sites scan in—and populates the database while maintaining field-level audit trails back to the originals for source data verification.
The platform also handles electronic patient-reported outcomes, randomization, medical coding via MedDRA and WHODrug, data monitoring, e-signatures, database lock, and exports. Harbor runs on Google Cloud and AWS infrastructure with isolated databases for each study. The company's subprocessor list includes Google's Vertex AI and Gemini models, OpenAI, AWS, Cloudflare, and various monitoring tools.
One customer testimonial on Harbor's site claims the system "cutting that entire process down to just two days." Another says uploading source documents and letting Harbor's AI extract the data reduced "transcription time by 90%."
Those are marketing claims, of course. But the MotilityCount case study offers a somewhat more concrete data point.
The Nine-Day Trial
MotilityCount ApS is running a multicenter randomized controlled trial evaluating the SwimCount Harvester device in IVF with preimplantation genetic testing. The study—registered as NCT07369362 on ClinicalTrials.gov—plans to enroll patients across 15 to 25 centers for around 1,600 IVF cycles.
According to a case study Harbor published, MotilityCount handed off the protocol and went live with a validated EDC in nine days. Other vendors had quoted 10 to 12 weeks. Harbor charged a flat $2,000 build fee. The trial is currently live and enrolling, though Harbor did not disclose how many sites are active or how far along enrollment has progressed.
That speed matters more than it might seem. The Tufts Center for the Study of Drug Development estimated the mean direct cost per day of a Phase III trial at around $55,716 in 2023 USD. Phase II trials run closer to $40,000 per day. Every week shaved off setup means a week closer to data—and a week of burn avoided.
Whether that velocity holds up across more complex protocols, larger site networks, or heavily regulated therapeutic areas remains an open question.
The CRO Bet
Harbor's recent expansion announcement frames the shift clearly: the company is moving from selling software subscriptions to operating as a full-service CRO. The $1.93 million contract will unfold over three years with a medical device sponsor. Harbor declined to specify the therapeutic area, trial phase, or patient population.
The thesis is that owning the EDC lets the company automate not just data capture but the operational layers stacked on top—study startup, site contracting, monitoring visits, query resolution, statistics, regulatory reporting. In theory, that should let a small team punch above its weight.
In practice? Harbor has seven trials running on its platform and a two-person founding team. Scaling from software to full-service operations is a fundamentally different challenge, one that involves site relationships, patient safety oversight, regulatory inspections, and the kind of institutional knowledge that doesn't automate easily.
The company hasn't disclosed whether it's hiring monitors, biostatisticians, or clinical project managers to support the CRO contract, or whether it plans to rely on automation and contractors.
The Founders
Albert Cai, Harbor's CEO, previously worked at Biolinq on clinical trials and regulatory affairs, shepherding a product through FDA's De Novo clearance pathway. Before that he spent time at Close Concerns, a diabetes-focused research and consulting firm. He studied biomedical engineering at the University of Michigan.
Nathan Leung, the CTO, came from stints at Google and Ramp and was the first employee at a prior YC-backed startup. He holds degrees in math from UCLA and computer science from Michigan. Leung recently completed ICH E6(R3) Good Clinical Practice certification, the latest revision of the international standard for clinical trial conduct.
The company was founded recently and participated in Y Combinator's program. No outside funding round beyond YC's standard investment has been publicly announced. LinkedIn lists the team size as 2 to 10 employees; YC's profile confirms two.
The Incumbents Aren't Standing Still
Harbor is entering a market where the established players have distribution, validation history, and customer inertia that's hard to overstate.
Medidata's Rave EDC has been around for over two decades and powers thousands of trials globally. Veeva's Vault EDC is used in over 2,000 trials across more than 600 customers, according to the company's recent proxy materials. Oracle Clinical One has deep penetration among pharma sponsors and academic medical centers.
All three have announced AI initiatives in recent months. Medidata's generative AI features are live in over 500 studies, the company said in a press release. Veeva has been quietly building AI-assisted protocol design tools. Oracle is embedding automation across its cloud-based clinical suite.
The large CROs—IQVIA, Parexel, ICON, Fortrea—have collectively participated in thousands of trials and employ tens of thousands of people. They've all launched AI initiatives. The Association of Clinical Research Organizations noted recently that its members participated in over 7,600 studies last year, and AI integration is accelerating across the sector.
Harbor isn't the only YC company working this problem. Phases, from an earlier batch, is building AI agents for trial operations starting with patient recruitment. But Harbor's approach—owning the data layer and layering services on top—sets it apart, at least structurally.
The Unbundling Thesis
Harbor's bet echoes a pattern that's played out across enterprise software over the past decade: whoever controls the data layer can automate the service layer, and speed becomes a moat.
Stripe didn't just provide payments infrastructure—it automated merchant onboarding, compliance, and fraud detection. Gusto didn't just do payroll—it bundled benefits administration and HR compliance. The wedge is software; the prize is owning the full workflow.
Clinical trials are ripe for that kind of unbundling, perhaps more than most industries. The process is fragmented, manual, and extraordinarily expensive. Sponsors typically contract with a CRO for operational execution, a separate EDC vendor for data management, another vendor for randomization and trial supply, yet another for patient recruitment. Each handoff introduces delay, cost, and error.
If Harbor can collapse those handoffs by owning the data infrastructure and automating downstream tasks, the economics could work even at a much smaller scale than incumbents. The question is execution.
What Happens Next
The $1.93 million contract is the test. Running a full CRO engagement with a two-person team and a platform that's been live in trials for less than a year will clarify whether the automation thesis holds when patient data starts flowing, sites need real-time support, and monitoring cycles begin in earnest.
The regulatory environment isn't getting easier. FDA finalized ICH E6(R3) guidance in late 2025, introducing stricter requirements for risk-based monitoring and electronic systems. Rolling effective dates extend into next year, which means any new platform needs to demonstrate not just speed but sustained compliance under tightening standards.
The broader strategic question is whether "AI-native" is a durable advantage or a temporary head start that evaporates as incumbents rebuild. Medidata, Veeva, and Oracle have the resources to acquire startups or reverse-engineer features. They also have the sales infrastructure to bundle AI tools into existing contracts at minimal marginal cost.
Harbor's edge, if it has one, is that it's building the stack from scratch without legacy technical debt or entrenched business models to protect. The risk is that it's a two-person team trying to compete in a heavily regulated, relationship-driven market where execution complexity scales faster than software can automate it.
For now, Harbor has a handful of live trials, one well-documented case study showing nine-day turnaround, and a seven-figure contract to prove the model works beyond the demo. Whether the thesis survives contact with a 15-center trial, patient safety events, and the grinding reality of clinical operations—that's the question the next three years will answer.
