The company teaching machines to read the part of your medical chart everyone else throws away.
A pharmaceutical analyst pulls a cohort of "severe asthma" patients. The claims data says 4,000 of them. Tidy. Defensible. Wrong. Because severity lives in a sentence a physician typed at 6pm - "patient remains poorly controlled despite triple therapy" - and that sentence was never a checkbox, so the database never saw it. Most real-world evidence platforms shrug and move on. Verantos stops, reads the sentence, and rewrites the cohort.
That, in one frame, is the whole company. Verantos is a Menlo Park healthcare-technology firm with a deceptively narrow obsession: making real-world evidence good enough to trust. Not impressive. Not big. Trustworthy - the kind a regulator, a reimbursement committee, or a clinician would actually stake a decision on. It is a less glamorous ambition than "AI will cure everything," and a far harder one.
Real-world evidence - RWE, in the trade - is the practice of learning from data generated by ordinary care rather than controlled trials. It is enormously promising and famously unreliable, because the data was never collected to answer the questions people now ask of it. The field's dirty secret is that scale is easy and validity is hard: anyone can assemble a hundred million records, and almost no one can tell you which of them are right. Verantos picked the hard half of that problem on purpose, and then made the difficulty its selling point.
The key clinical detail often lives in the narrative notes - the part most analytics platforms throw away.
Healthcare data is a junk drawer. Claims here, labs there, a discharge summary in free text, an imaging report nobody parsed. The industry's tidy answer was to use only the parts that fit neatly into columns - and to politely ignore how much got lost. Verantos took the opposite bet: that the messy, unstructured clinical narrative is not noise to be discarded but signal to be mined.
So it built AI to read the complete patient record and pull out what matters - disease subtypes, severity, outcomes - then it does the thing almost nobody else does. It measures how right it is. Every dataset ships with three numbers: accuracy, completeness, and traceability. Data quality stops being a vibe and becomes a figure you can audit.
Values shown are illustrative of the framework Verantos reports for its studies, not a specific published result.
Launched September 2022. Curates, links, and enriches the complete patient record - structured data plus clinical narrative - to generate RWE that aims at the standards in FDA draft guidance.
Disease-specific registries built from routinely collected real-world data, curated for accuracy and delivered in standardized OMOP CDM - ready for clinical development, market access, and pharmacovigilance.
A federated, de-identified, linkable data network spanning health systems and partners - the raw material for representative, high-validity patient cohorts at scale.
Dan Riskin is an unusual specimen: a practicing surgeon (MD from Boston University, residency at UCLA, fellowship at Stanford, where he still teaches as a clinical professor) who is also an engineer, and who keeps building companies at the seam between the two. Verantos, founded in 2015, is his second act in health data - the first was Health Fidelity. He has argued, repeatedly and persuasively, that the bottleneck in healthcare AI is not the model. It is the truth the model is fed. Get the evidence wrong upstream and every downstream cleverness compounds the error.
It is a conveniently humble worldview for a man who sells exactly the thing he says is missing. It is also, inconveniently for skeptics, correct. The history of healthcare analytics is littered with dashboards that looked authoritative and turned out to be measuring the wrong thing - a coding artifact mistaken for a clinical fact, a billing convenience read as a diagnosis. Riskin's whole career has been a long argument that you cannot automate your way out of bad inputs, only into them faster.
Strip away the jargon and Verantos is useful in a handful of concrete ways. A drug developer can stand up a high-validity cohort to study how a therapy performs in the real world - not the curated world of a trial, but the messy one of actual clinics. A market-access team can build the evidence a payer demands before it agrees to reimburse. A medical-affairs group can answer the questions physicians ask after a launch, when the trial is over and the real patients arrive. A pharmacovigilance team can watch for safety signals in post-market data with enough fidelity to act on what it finds. And a regulatory team can assemble a study designed, from the first row of data, to survive the scrutiny laid out in FDA draft guidance.
What ties those jobs together is the same quiet promise: every number traces back to a source you can inspect. That is the unglamorous superpower. When a regulator asks "how do you know," the answer isn't a shrug or a black box - it's a click back to the original clinical sentence. In an industry where a single contested data point can stall an approval for a year, traceability is not a feature. It is the product.
The first company to generate research-grade evidence at scale across all therapeutic areas.
The customers are the kind that don't sign easily: Amgen and Bayer among them, plus the regulators and academic medical centers who supply the data and read the results. The plumbing comes from partnerships - Datavant for privacy-protecting record linkage, CuriMeta for an extra ~7 million de-identified patients poured into the network.
Here's the twist on the usual startup story: Verantos's early science wasn't bankrolled by venture capital chasing a unicorn. It was funded by competitive federal research grants - an SBIR Phase II award of roughly $1.5M from NCATS at the NIH in 2018, followed by further NIH support for advanced clinical phenotyping. The science had to survive peer review before it ever met a sales deck.
The growth that followed was real enough to land Verantos on the Inc. 5000 (2022) and the Deloitte Technology Fast 500 (2023).
Dan Riskin founds Verantos to bring measurable validity to real-world evidence.
Awarded an SBIR Phase II grant (~$1.5M) from NCATS at the NIH.
Receives NIH grant to fund advanced clinical phenotyping in RWE studies.
Launches the Verantos Evidence Platform for high-validity RWE.
Named to the Deloitte Technology Fast 500 (#189 in North America).
Partners with CuriMeta, adding ~7M de-identified patients to the network.
Riskin details the accuracy/completeness/traceability framework on The Healthcare Technology Report.
Return to that analyst and the 4,000 "severe asthma" patients. After Verantos reads the notes, the number changes - some patients in, some out, each with a traceable reason you can click through to the original sentence. The cohort is smaller, or larger, but the difference is no longer the point. The point is that someone can now defend it in front of a regulator, a payer, or a physician who will use it to decide what happens to a real person.
That's the quiet trade Verantos offers the life-sciences world: not more data, and not louder claims about AI - just evidence with its receipts attached. In a field that has spent a decade dazzled by scale, building something you can actually trust turns out to be the contrarian move.