Ariel Katz

The through-line

Ariel Katz has spent his working life on one problem stated in one sentence: nobody can find out who is doing what in science and medicine. It began as a student frustration, that you cannot answer "tell me all the latest biotech research going on at Harvard right now" without clicking through websites and publications [1], and that he could not get into a lab at Binghamton while being qualified enough to do consciousness research at Columbia [3]. The response was a platform where researchers post opportunities and students apply, "think about like indeed for research opportunities" [3], and his summary of the impulse is the cleanest statement of his method: "i wanted a sandwich i couldn't get a sandwich i needed to build a place to get me a sandwich" [3]. H1 is the same problem at global scale. He describes it as "LinkedIn meets zoominfo meets ZocDoc for doctors" [1], holding "everything you want to know about every doctor in the world" [1][3], and frames the purpose as connection rather than data: the mission is "to create a healthier future", the vision is "connecting parties in the healthcare ecosystem", and the mechanism is "providing a source of truth of doctor information for the world" [2].

What has shifted is the frontier, not the thesis. Early on the argument was about coverage and aggregation, about being the first to profile roughly ten million healthcare professionals globally [1][3]. More recently the argument is about what AI can now resolve inside that data, and about the way drug pricing regulation is redrawing the map of where clinical research physically happens [4]. Throughout, he refuses to treat financing or revenue as the score: "we barely skimmed the surface of our mission" [2].

On why doctors never adopted LinkedIn

His starting observation is that professional networking software was built for people whose careers look nothing like a physician's. "There are two segments that don't use LinkedIn doctors Healthcare professionals and people in entertainment", because they do not need an online network in the way other workers do [1]. Doctors do not move jobs often and do not switch tracks, "a pediatrician doesn't become a medical oncologist pediatrician's pediatrician", whereas people in sales, marketing, PR or HR can move anywhere and are constantly contacted by recruiters [1]. Because clinicians are scattered across institutions, there is no central directory and no way to answer a query as simple as showing all the oncologists in a given city [1]. So LinkedIn's use cases do not translate, and H1 built something different: a network where a doctor can claim an H1 profile, sitting on top of a full record the doctor did not have to enter [1].

The same failure shows up in the tools patients actually touch. Health plan find-a-doctor pages, he argues, give you neither reliable contact details nor any sense of what a physician clinically focuses on, and a CMS study found roughly 46 percent of directory entries inaccurate [3]. Searching instead is worse: you can Google the best oncologist in Boston and get something bias-ridden, and you cannot meaningfully Google the best oncologist in Chennai, Mexico City or Saskatchewan, or the best pharmacist who understands HIV [1]. Even conventional search surfaces clinicians who merely mention a condition and cannot help at all, which patients only discover in the appointment [2].

On building a source of truth

He is explicit that no single feed produces a usable profile. Public hospital sites give you first name, last name, phone number and specialty, "that's not enough" [1]. So H1 buys data, including government and insurance data on the types of patients a physician sees, pulls publications from PubMed, studies from clinicaltrials.gov, and signals from Twitter and LinkedIn, then adds what doctors and the community contribute directly [1]. For a single physician that means "hundreds of different sources coming in for that one doctor and we mush it all together" [1]. Self-reported input is welcome but never taken on trust, since "sometimes people want to make themselves seem bigger or smaller than they really are maybe they're humble maybe they're arrogant", so everything is validated [1].

The structure matters as much as the volume, because the questions users ask are relational: whether two doctors trained together, sit on the same medical society board, share patients, or whether a physician belongs to a hospital inside an IDN insured by particular payers and partnered with a particular pharma company [1]. A lot of this already sits in a graph, some remains relational, and the goal is to move everything into what he expects will be the largest healthcare graph in existence [1]. The endpoint he keeps describing is a specific search experience that does not exist yet, asking who is the best doctor for my eczema, my hip replacement, or to run my bladder cancer trial in a named city, which is why funding goes into natural language processing, data science, search and engineering [2]. And it must hold globally: "John Smith in Tampa that's not so hard what about in Uzbekistan that's a harder one" [2].

On who actually uses it, and what they pay for

He maps demand by the question each customer is trying to answer. Pharma, biotech and medical device companies need to know which doctor to work with across the whole lifecycle: which investigator can run a trial, which physicians medical affairs teams should educate about a therapy and which can change medical practice, and eventually who to market an approved drug to [1]. Insurers ask two questions, which doctors to insure, since "insurance companies make money when people are healthy not on people that are sick", and how to show members useful information for choosing a physician [1]. Doctors themselves come for the latest science, comment on it, and claim and update their profiles, with that commentary propagating to everyone else [1]. The commercial model follows the value: always free for patients and doctors, which he considers "good for the world good for healthcare", with annual subscription licences sold to pharma, biotech, medical device, insurers and hospitals [1]. Powering insurers' find-a-doctor pages so patients land on the right physician is a distinct initiative he treats as a major push [3]. He has also discussed the specific case of AI reshaping how pharma engages thought leaders [10].

The displacement argument is concrete. Trial teams used to work from clinicaltrials.gov, Google, press releases and internal files, none of which tells you the race, ethnicity or income level of a physician's patients, or whether they actually recruited anyone on their last study, "can't get that publicly so you have to come to H1 to answer those two questions" [1].

On why picking the right investigator is hard

Selecting a trial site sounds trivial and is not. If the protocol covers atopic dermatitis in 65 to 85 year olds, the investigator has to actually see patients in that band [3]. If the trial is meant to be representative, it cannot be run entirely by and on white males, and if it is not confined to the United States the same evidence has to exist worldwide [3]. Language is part of the match: recruiting Hispanic patients requires knowing which physicians speak Spanish, just as a patient who only speaks Spanish needs to know that before booking [3]. He is blunt that the industry usually gets this wrong, that pharma companies generally do not find the right doctor and consequently do not recruit patients [2].

The cost of getting it wrong is measured in burn. Every day a trial runs without recruiting, a sponsor is losing money, "big money hundreds of millions" depending on the therapeutic area [3]. When Russian sites shut down, sponsors called H1 asking where the patients, doctors and hospitals were so they could relocate studies [3].

On AI as an administrative fix, not a discovery story

Katz sorts the AI landscape into two piles and is candid about which one he is in. There are the drug discovery companies chasing the next GLP-1, "the flashy exciting ones that you might hear about", and then there is everything operational [4]. His view is that, as with hospitals, "the biggest bang for your buck comes with the administrative overhead reduction", because by the time you are running a trial you already believe the science and are spending hundreds of millions to test it [4]. What AI now makes tractable is selecting the right doctor and the right research site, knowing where patients are and reaching them directly, and knowing while you draft the protocol exactly where you will run it and when those patients will be available: "some of those basic questions are now solvable with AI that were not solvable before" [4].

He thinks the industry's caution has flipped. The old posture was to let bigger companies experiment first and copy what worked, and "if you do that, you will be like in the dark ages behind" [4]. He sees sponsors past the experimentation stage, the FDA adopting AI for reviews, and a normally slow industry moving quickly [4]. His practical advice starts from admitting the confusion is universal, since even a technology company cannot keep up with successive model releases, "as humans, we can't keep up with it" [4]. The answer is a decision or evaluation framework built on defined questions, followed by aggressive internal implementation: "you need to move aggressively in this day and age or you will be left behind", and the debate about whether AI is real or will be adopted by enterprises is "long past" [4].

On regulation redrawing the map of where trials happen

He argues that drug pricing policy is quietly relocating clinical research, and that almost nobody is saying so publicly. Under most favored nation pricing, American prices are set near the average of peer countries, which gives sponsors an incentive to stop launching drugs in lower-priced markets so those markets do not drag the average down [4]. The operational consequence follows directly: "they're quite literally going to not do clinical trials there", and in conversations with top 20 pharma companies he hears more appetite for giving a drug away free or skipping studies in a country than for taking the price hit [4]. He calls this "the dark side that nobody talks about": cheaper drugs in America, and countries that end up with no access to therapies patients need [4]. The IRA adds a separate pressure, pushing sponsors toward concurrent studies so a drug launches with multiple indications on label rather than adding them sequentially in the Keytruda pattern [4]. Since H1's products are used to decide where trials should run globally, he watches this change in real time [4].

On judging the company by impact, not by capital

Katz treats financing as an output. Valuation, money raised, revenue and client count are "not a North star at all"; the North Stars are impact metrics [2]. He names two: enabling a hundred million patient-doctor interactions within two years, a quarter of the United States finding the right physician through H1, and half of American clinical trials using H1 to ensure diverse patients and diverse physicians are enrolled [2]. The logic connects the second to the first: more representative medicine builds trust among minority communities, produces more successful trials, and therefore delivers therapies that might otherwise not exist [2]. Asked whether H1 is succeeding, he separates the financial answer from everything else, noting that money is "a very shallow way to look at an organization" [1]. The posture is deliberately unsatisfied. A hundred million is "a drop in the ocean" against seven billion people, and after a $100 million round his reaction was that he had just tied his shoelace as the starting gun fired, echoing Kobe Bryant's "job's not finished yet" [2].

On raising money and what it buys

His account of financing is unsentimental about process and specific about purpose. Investors framed each round as cards on the table: more money means more cards you can flip over, not all of them work, and you need enough of them to hit the milestone that justifies the next round [1]. Between the A and the B a couple worked, and again between the B and the C, and those cards were product line expansion, geographic expansion and continued investment in what users want, with most capital going into engineering, product and data [1]. He originally thought $20 million was a lot and was told it was far too little [1]. Companies like H1 run a deliberate deficit in order to grow faster, and the spend breaks into data science, engineering and getting the product in front of as many people as possible [2]. He notes wryly that use of proceeds is a question later-stage investors barely ask during a raise, since conversation centres on growth of the business [2].

The Y Combinator decision was equally pragmatic. With a few million in revenue and a signed term sheet already in hand, he initially refused on the grounds that H1 was too late stage, and joined only after being told he would get better terms, which proved correct when the round came in well above the earlier valuation [1]. The lesson he took from YC is a two-part definition of the job: "your job as a CEO before product Market fit is to get product Market fit your job as a CEO after product Market fit is to get your thing in front of as many human beings that need it as quickly as possible without breaking the company" [2]. He has also talked through how to hire as a company scales, and about working too hard [11].

On timing, luck and execution

Katz is unusually willing to attribute H1's opening to circumstance. "Half the game is luck and half the game is timing", and the timing was that after the ACA, CMS made far more information public about physicians, including which doctors work with which pharma companies and what kinds of patients they see, after which other countries followed [3]. He credits Aneesh Chopra and notes he only understood the pattern in hindsight: the government innovated, and H1 started when "the iron was hot" [3]. Competitors including Definitive Healthcare and Komodo Health began two or three years earlier, which he characterises as a real but small head start within a single generation of companies [3]. What H1 contributed was scale, a global platform of roughly ten million healthcare professionals that nobody had attempted [3]. His arithmetic for the rest: "the idea is one percent 49 luck and timing 50 execution and so we just executed better" [3].

That same instinct shaped his choice of problem after selling Research Connection. Teammates wanted to build a social app for athletes; he reasoned that he was "in the one percent of knowledge around profiling of research activity and understanding how research gets funded" and that building there was the higher-probability bet [1]. Running Research Connection had already taught him how research gets funded, that NIH's roughly $40 billion a year was flat to slightly declining, that non-profit funding was not growing, and that industry was where the gap had to be filled [1].

On what doctors think of the science

The newest layer of the profile is opinion rather than fact. Through Faculty Opinions, which H1 partnered with and acquired, invited biologists and physicians write recommendations on new publications and new medicine, saying whether it is good or bad and why [3]. It is invite-only and deliberately prestigious, seeded with the most eminent names in a field, including eight Nobel laureates, who then invite their peers, so a topic like eczema is covered by "the creators of eczema cream type doctors" rather than ten random clinicians [3]. He distinguishes it from UpToDate, which he sees as curated content written by medical writers and closer to a publication you search; Faculty Opinions is "more like rotten tomatoes for medicine and science", or more precisely a New York Times book review, since it carries the critics and not the audience [3]. The patient-facing point is simple: when your doctor prescribes a drug, you have no idea what other doctors think of it [3].

Takeaways

• The core product claim is coverage plus trust: "everything you want to know about every doctor in the world", assembled from hundreds of sources per physician including public sites, purchased government and insurance data, PubMed, clinicaltrials.gov and social platforms, with self-reported input always validated [1][3].
• Physicians never adopted LinkedIn because they rarely change jobs or specialties and are not courted by recruiters, so healthcare needed a directory built around what a doctor treats rather than where they want to work next [1].
• The biggest near-term AI gains in clinical trials are administrative, not scientific: picking the right investigator and site, locating patients, and drafting protocols around where those patients actually are [4].
• Most favored nation pricing will push sponsors to stop launching drugs and stop running trials in lower-priced countries, a consequence he says is discussed in pharma boardrooms but not publicly [4].
• Adopt AI with a defined evaluation framework and then implement aggressively; waiting for larger companies to experiment first now leaves you behind [4].
• Success is measured against impact North Stars, a hundred million patient-doctor interactions and half of US clinical trials using H1 for diverse enrollment, with revenue and valuation treated as outputs [2].
• Trial recruitment failure is expensive in the hundreds of millions, which is why matching investigator patient panels, age ranges, ethnicity and language to the protocol is the decisive operational question [3].
• Roughly 46 percent of health plan physician directory entries are inaccurate according to a CMS study, which is the gap H1 aims to close by powering find-a-doctor pages [3].
• H1's opening came from post-ACA CMS transparency that made physician data public, a break he sums up as "the idea is one percent 49 luck and timing 50 execution" [3].