Carta Healthcare.

01The Nurse, The Screen, The Quiet Software

Somewhere in a hospital you have heard of, a nurse abstractor is sitting in front of a screen at 9:14 on a Tuesday morning. She is looking at a patient chart that has, at last count, 412 unstructured notes attached to it. She used to read all of them. Now she scans a clean summary on the left, glances at the highlighted source sentences on the right, and approves the row. The chart took twelve minutes instead of forty-five. She has another forty-two to clear before lunch. The software in the middle of this scene does not introduce itself. It just keeps the queue moving.

That software is Carta Healthcare. It does not look like a moonshot. It is one.

Caption: The highest praise software can earn from a clinician is the praise it gets when no one notices it.

02The Problem They Saw

The American medical record is a strange document. It is exhaustive and incoherent at the same time - a stack of forms, lab results, narrative notes, dictated paragraphs, and the occasional faxed PDF that no one asked for. Roughly eighty percent of what is in there is unstructured. To turn it into something a clinical registry can use, somebody has to read it, decide what counts, and re-type the answer into a field.

That somebody is almost always a nurse. The work is slow, expensive, and remarkably hard to automate. It is also non-optional: clinical registries are the backbone of how hospitals benchmark surgical outcomes, satisfy regulators, and qualify for value-based payments. If the data is bad, the conclusions are worse.

The size of the chore

For a single cardiac procedure, manual abstraction can run an hour per case. Multiply across registries, across service lines, across a health system the size of UPMC or Mass General Brigham, and the dollars become indistinguishable from the dollars a small product company raises in a Series A. The chore is a budget line.

Caption: Says the AI company. Which is the whole point.

03The Founders' Bet

In 2017, three founders - Matt Hollingsworth, James Matheson, and Anna Chukaeva - decided the way to fix the abstraction problem was not to replace the nurse. It was to give the nurse a much better partner.

This was a counterintuitive bet at the time. The fashionable pitch in healthcare AI was full automation, the kind that ends sentences with "no humans in the loop." Carta's founders went the other direction. They argued that for clinical data - where a wrong answer is a wrong answer about a patient - the only durable model is what they would later call Hybrid Intelligence. AI does the heavy reading. Expert humans validate. The two iterate.

It is the kind of position that sounds modest at a pitch meeting and competent in a hospital procurement review.

Caption: The user, not the marketing team. There is a difference.

04The Product

Carta sells two main products and one operating philosophy. The products are called Atlas and Lighthouse. The philosophy is that hospitals should stop paying for chart review by the hour.

Lighthouse

Lighthouse is the abstraction engine. It rapidly analyzes structured and unstructured data inside a patient chart, surfaces the answers a registry asks for, and shows its work - the exact source sentence in the exact note - so a human can verify in seconds rather than minutes. The customer-reported result is an inter-rater reliability score of 98 to 99 percent. In a domain where consistent human-to-human agreement is often a percentage point or two lower than that, this is a quietly outrageous number.

Atlas

Atlas is the registry workhorse. It uses large language models and Carta's clinical playbook to take abstraction from capture to submission - the part registry coordinators dread the most, where regulatory formats meet idiosyncratic field rules. Atlas handles the format. The coordinator handles judgment calls. The submission goes out on time.

Caption: The kind of numbers a CFO underlines twice. Source: Carta Healthcare and Series B1 announcement, May 2025.

05The Proof

Health-tech is a category where pitch decks promise rocket ships and pilots last forever. Carta has the harder evidence: customers who keep paying, and investors who are also customers.

UPMC Enterprises led the Series B1. UPMC also runs the software. MemorialCare and Memorial Hermann are on the cap table and on the customer list. Mass General Brigham Ventures invested in 2025; Mass General Brigham itself sits among the deployments. This is the kind of overlap that, in any other category, would feel cozy. In healthcare, where institutions only buy what they can defend in a board meeting, it is corroboration.

Caption: A polite way to say: the autonomous-AI-only pitch is not going to survive a chief medical officer's inbox.

06The Mission

The official line is that Carta exists "to harness the value of clinical data to improve patient care." The unofficial line, audible if you spend enough time around the company, is closer to this: the next generation of healthcare cannot be built on data nobody trusts, and the only way to build the trust is to combine machines with humans who know what they are reading.

It is a small philosophical claim with enormous practical consequences. A clinical trial that recruits the wrong cohort wastes years. A registry that misses a complication misshapes a national guideline. A health system that abstracts inconsistently across two hospitals cannot benchmark its own surgeons. Data is not a feature in healthcare. It is the floor everything else stands on.

Carta's bet is that the floor has been quietly rotting for a long time and that nobody felt empowered to mention it.

Caption: Most companies dress this up. Carta sells it on the label.

07Why It Matters Tomorrow

The reason a quiet abstraction company keeps showing up in funding announcements alongside the loud generative-AI ones is the same reason hospitals keep signing renewals: the demand curve for clean clinical data is going almost straight up.

Real-world evidence is now table stakes for pharma. AI models for diagnosis and triage are only as honest as the labels they learned from. Clinical trials need patient cohorts identified in days, not quarters. Every one of those workflows starts at the same place - a stack of unstructured notes - and ends in the same place: a row in a database somebody can defend.

If Carta is right, it has positioned itself as the layer between the two. If Carta is wrong, the rest of the AI-in-medicine pitch decks have a much bigger problem than they think.

The scene again · same hospital, lunchtime

Back to the nurse abstractor. It is now 12:51. Her queue is empty. There are forty-two more charts ahead of her this week, and she will read them by Friday afternoon, and on Friday afternoon she will go home on time. Her registry submission for the quarter is already drafted. Her name does not appear in any press release. The software that helped her does not, today, want to be famous. It wants the next chart.

That is what Carta Healthcare is selling. Not a revolution. A reliable Tuesday.