A single protocol amendment midway through a Phase III trial can cost upward of $535,000. It's a figure that doesn't account for the cascading delays—pushed timelines, disrupted patient enrollment, stalled regulatory reviews. And it happens all the time. More than 60% of clinical trials require at least one amendment after they're underway, corrections that often trace back to flawed or overly complex protocol design at the start.
That's the problem Risklick, a Bern-based healthtech startup spun out of the University of Bern, has set its sights on solving. And if two recent partnerships are any indication, the pharmaceutical industry might finally be ready to let AI take the reins on one of its most stubborn bottlenecks.
The company just notched a completed pilot with Debiopharm, the Swiss oncology-focused pharma group that also happens to be a seed investor, and secured a strategic tie-up with Citeline, a heavyweight in clinical intelligence that maintains one of the industry's deepest databases on trial design and regulatory filings. Neither move is flashy on the surface—pharma partnerships tend to move quietly, deliberately—but together they suggest something more than polite interest.
Pilots That Actually Matter
Let's start with Debiopharm. In October 2025, the company announced it had completed a pilot applying Risklick's Protocol AI to oncology protocol creation. On its face, that's not terribly remarkable. Plenty of AI vendors claim they can speed up document drafting, and plenty of pharma companies run pilots that never see the light of production.
What made this one worth paying attention to: Debiopharm Innovation Fund had led Risklick's seed round back in April 2024 (the amount was undisclosed), and Debiopharm doesn't typically pilot technology from its own portfolio unless it sees a real path to deployment. The Swiss firm is too pragmatic for theater. Oncology protocols are notoriously complex—multilayered patient populations, intricate treatment regimens, endpoints that regulators scrutinize down to the decimal point. If Risklick's tool could hold up in that environment, it might actually have legs.
The companies didn't release quantitative results, which is frustrating but typical in pharma. Still, the fact that Debiopharm was willing to attach its name to the outcome suggests the pilot cleared an internal bar that matters.
The Data Problem AI Protocol Tools Couldn't Solve—Until Now
Where things get more interesting is the Citeline partnership, announced in mid-January 2026.
Citeline, part of Norstella, is a clinical intelligence powerhouse. It aggregates trial design patterns, regulatory submissions, post-market surveillance data—basically, if it's been filed with an agency or run through a clinical program, Citeline probably has it indexed. Risklick is now integrating that data layer directly into Protocol AI's authoring workflow, aiming to build something closer to an end-to-end system: one that doesn't just generate text, but pulls in historical precedent, competitive intelligence, and regulatory patterns as protocols are being drafted.
For Risklick, this solves a problem that has dogged early AI protocol tools. The company already claims to draw on a database of more than 800,000 clinical trials, but Citeline's contribution adds a different kind of depth—regulatory disclosure, competitive positioning, the kind of granular detail that matters when a protocol lands on an agency reviewer's desk or needs to stand out against a rival's program.
Dr. Poorya Amini, Risklick's CEO and a former researcher at the Clinical Trials Unit Bern, has been careful to position the platform as evidence-based rather than generative in the ChatGPT sense. Protocol AI references ICH M11 and the USDM (Unified Study Definitions Model), regulatory frameworks designed to standardize protocol structure across jurisdictions. It exports to Word, Excel, JSON—formats that fit comfortably into existing pharma workflows, which matters more than you'd think. Pharma IT environments are notoriously sticky. If a tool requires ripping out legacy systems, adoption stalls.
The Efficiency Pitch (With the Usual Caveats)
Risklick claims Protocol AI cuts protocol development time by 50%, reduces amendments by 20%, and speeds up implementation into electronic data capture systems by 40%. Those numbers come from the company's own materials and haven't been independently verified, a caveat that's worth flagging in a sector where AI vendors sometimes let enthusiasm outpace evidence.
That said, the broader case for AI in protocol design isn't speculative. Citeline's own data pegs the cost of a Phase II amendment at $141,000, and Phase III protocols now routinely run over 200 pages—a far cry from the streamlined documents regulators once expected. If AI can shave even 10% off that complexity, the ROI starts to look credible. Perhaps not the 50% Risklick is claiming, but credible.
Branching Into Medical Devices
In January 2025—a year before the Citeline announcement—Risklick quietly extended Protocol AI to medical device clinical trials. It's a different regulatory universe: CE marks in Europe, 510(k) pathways in the U.S., design considerations that don't map cleanly to pharma. The move suggests the underlying architecture is flexible, though whether contract research organizations and device manufacturers adopt protocol automation at the same pace as pharma remains to be seen. Device trials tend to be smaller, leaner operations. The urgency around protocol optimization might not be as acute.
A Crowded Field, But Not Yet a Winner
Risklick isn't operating in a vacuum. Faro Health offers ICH M11-compliant authoring tools with a similar AI pitch. IQVIA bundles protocol optimization analytics into its sprawling R&D suite. QuantHealth focuses on virtual trial simulation and protocol refinement. The question isn't whether AI will reshape protocol design—it almost certainly will—but which platforms can embed themselves deeply enough into pharma's operational stack to become infrastructure rather than pilots.
Right now, Risklick's traction depends on converting partnerships into revenue. The company lists Debiopharm and ISS AG (Integrated Scientific Services, a Swiss CRO) as partners, and the Citeline relationship opens potential doors to Norstella's pharma client base. Whether that translates into recurring contracts is the open question.
The team is small—LinkedIn suggests somewhere between 11 and 50 employees—and the company holds an Innosuisse certificate, Switzerland's stamp of approval for startups within its national innovation ecosystem. Risklick also participated in the DayOne Health 4.0 Accelerator in Basel during the 2023/2024 cohort, though accelerator pedigrees don't always predict commercial staying power.
The Trust Problem
Here's the thing pharma executives won't say out loud but think constantly: Can we trust AI with documents that regulators will dissect word by word?
It's not a technical question. It's a cultural one. Pharma R&D is risk-averse by design, trained by decades of regulatory scrutiny and billion-dollar failures to mistrust shortcuts. AI-generated protocols might save time. They might reduce errors. But they also introduce a layer of abstraction between the clinical team and the document that will define a trial's success or failure. That abstraction makes people nervous.
Risklick's answer, for now, is to emphasize standards compliance and integration with existing workflows. The Citeline partnership helps—linking to a trusted data source adds legitimacy. The Debiopharm pilot helps too, offering at least one proof point from a credible pharma player. But adoption at scale will require something harder to manufacture: comfort. The kind that only comes when regulators have seen enough AI-assisted protocols to stop raising eyebrows, and when enough R&D chiefs have seen the efficiency gains stick.
With the partnerships in place and a product that at least one pharma investor was willing to pilot in oncology—perhaps the most unforgiving testing ground imaginable—Risklick has bought itself a window. The real test begins now.
