Clarum Wants to Read Every Document in the Deal - Before You Finish Your Coffee
A two-person San Francisco startup is teaching software to do the reading that private-equity associates dread. First it was due diligence. Now it wants to be the memory of the whole firm.
The private-equity data room is one of the least glamorous places in modern finance. It is a folder - sometimes thousands of files deep - holding the financial statements, contracts, customer lists and legal fine print of a company someone is deciding whether to buy. Someone has to read all of it. Traditionally that someone is a junior analyst who can get through a few dozen documents a day, fueled by coffee and the knowledge that a single missed footnote can sink a deal. Clarum, a two-person startup out of Y Combinator's Winter 2024 batch, was built on a plain observation: the reading is the bottleneck, and the reading is exactly the kind of work software has finally gotten good at.
The company's pitch, when it launched, was direct. Upload your files and your diligence questionnaire, and Clarum will answer hundreds of questions - transparently, with sources - in minutes rather than weeks. It reports cutting the time required for thorough due diligence by roughly four to five times, while claiming to deepen the risk analysis rather than thin it out. The premise is not that the machine has better judgment than a partner. It is that the machine never gets tired, never skims, and does not care that it is the fourth hundred-page appendix of the afternoon.
The reading was never the hard part - it was the volume
Ask anyone who has worked a deal and they will tell you the same thing: due diligence has looked roughly the same for decades. A room full of PDFs. A checklist. A stretch of nights. The constraint was human throughput. A person can only hold so much in their head at once, which means whole categories of question simply go unasked because answering them would take too long to be worth it. Clarum's founders framed the opportunity as a limit to be pushed past. As chief executive Anton Otaner put it, "We can push past these limits - imagine looking through billions of data points, all interconnected, in a few minutes."
That sentence is doing more work than it looks. The interesting shift is not that diligence gets faster. It is what happens to behavior once the reading is effectively free. When a task that cost three weeks now costs an afternoon, firms stop rationing it. They run diligence on deals they would have passed on, and they ask the fussy, cross-referencing questions - does this customer contract actually match the revenue in the model? - that used to be a luxury. Cheaper analysis does not just save hours. It changes the arithmetic of which deals a market can afford to look at.
"Imagine looking through billions of data points, all interconnected, in a few minutes."
What Clarum actually does all day
Rather than asking analysts to move into a new interface, Clarum works inside the tools they already live in - Excel, Word and PowerPoint. That is a quiet but deliberate design choice. The software drafts screening memos, builds returns analyses and financial models, assembles deep-dive presentations, reconciles income statements, and summarizes legal documents and the confidential information memorandums, or CIMs, that anchor a deal. It flags inconsistencies across files, compares and benchmarks deals against each other, and sorts the contents of a data room so a human knows where to look first.
Notice what is on that list and what is not. The list is reading, summarizing, reconciling and formatting - the grunt work. What is missing is the decision. Clarum is not trying to tell a fund whether to write the check. It is trying to separate the reading from the judgment, hand the reading to a model, and give the human back the part that actually requires being human. That division of labor is the whole product philosophy in one line.
There is a second, subtler reason to work inside the analyst's existing files rather than a shiny new dashboard. Trust in this world is built line by line. A partner is not going to accept "the AI said so" as a reason to move a hundred million dollars. So Clarum leans on transparency: answers arrive with their sources attached, traceable back to the exact document and page they came from. That auditability is not a nice-to-have. It is the difference between a toy that impresses and a tool a firm will actually put in front of an investment committee.
Where the diligence hours go - illustrative
The quiet pivot: from a tool to a layer
Something more interesting happened after launch. Clarum started as a due-diligence product and has been broadening into what it now calls "the intelligence layer for private capital." The reframing came, as good ones usually do, from watching what customers actually did. Firms did not just want a faster way to grind through one deal. They wanted to ask their own history a question and get an answer: why did we screen out that deal last year? What diligence gaps show up in the investments that later underperformed? Which of our portfolio companies quietly share the same concentration risk?
Those answers already exist inside a firm. They are just scattered across deal memos, partner notes, diligence files and returns data that no human can hold in view at once. Clarum's newer bet is that structuring that mess into a single, queryable model is worth more than automating any one task. It turns a fund's own accumulated memory - the thing that walks out the door when an analyst leaves - into something the whole firm can interrogate. Data you cannot search, in practice, is data you do not have.
A fund's best data is usually the data it already owns and cannot read.
The founders
Clarum was started in 2023 by two people who had already proven they could out-build a room. Anton Otaner and Tommy He met through a mutual friend at McGill University and decided to work together after winning first place in their university's hackathon - the kind of origin story that sounds like a cliche until you notice how many good companies actually start exactly this way.
Before Clarum, worked on blockchain infrastructure at Mint and led augmented-reality product development at EPHAS, an Ericsson One startup. Runs marathons in his spare time.
Background in theoretical math and computer science. Came from Tower Research Capital, where he worked on low-latency market connections and trade execution.
The pairing is telling. Otaner's résumé is a tour of hard, unglamorous infrastructure - blockchain plumbing, AR engineering - the kind of work that teaches you to respect messy real-world data. He's background is high-frequency trading, a world where reading enormous streams of information faster than the next firm is the entire game. Put a data-infrastructure builder and a low-latency quant on the problem of "read this data room instantly and find what matters," and the founding team starts to look less like a coincidence and more like a thesis.
Who it is for, and who else is in the room
Clarum's customers are private-market investors: private-equity firms, wealth managers and family offices. The company describes a founding cohort of general partners and limited partners it is building alongside, though, as is common at this stage, it does not name them. The buyer profile makes sense. These are firms whose competitive edge is analysis, whose data is intensely sensitive, and whose associates spend a punishing share of their week on exactly the reading-and-reconciling work Clarum automates.
It is not the only company to notice the opportunity. A wave of startups is aiming AI at financial and legal document work - names like Hebbia, Rogo and BlueFlame AI, alongside broader tools built for lawyers and analysts, and the ever-present incumbent: a room full of people doing it by hand. Clarum's wager for standing out is less about answering a single question faster and more about owning the structured layer underneath - the model of a firm's data that every future agent and tool would need to plug into. If diligence is the wedge, the durable business is the plumbing.
The distinction matters more than it sounds. A faster answer is a feature, and features get copied. A structured, permissioned model of a firm's entire history is closer to infrastructure - the sort of thing that gets more useful the longer a firm uses it and harder to rip out once it is embedded in daily work. Every memo run through it, every deal it benchmarks, makes the next query a little sharper. That compounding is the quiet moat Clarum is reaching for, and it is why the pivot from a diligence tool toward an intelligence layer reads less like a change of subject and more like a founder following the pull of what the product was always going to become.
Where Clarum sits in the stack
The business, and the honest unknowns
Clarum is a seed-stage company backed by Y Combinator, which ran it through the Winter 2024 batch under group partner Diana Hu. Its model is business software sold to investment firms - the kind of tool a firm adopts because it pays for itself in analyst hours and, more importantly, in deals that no longer slip through because someone missed a line. Public funding figures for the company vary by source, and Clarum has kept specifics like customer names and revenue private, which is normal for a company this young. Where the numbers are uncertain, they are best treated as approximate rather than settled.
What is clear is the shape of the ambition. Start with the most painful, most repetitive job in private capital - reading the data room - and use it as the entry point to something larger: becoming the system that holds and answers questions about everything a firm knows. Whether Clarum gets there depends on execution, trust and the messy reality of selling to institutions that guard their data closely. But the underlying observation is hard to argue with. In finance, the person who can read everything, remember everything and cross-reference it in seconds has an edge. Clarum is betting that person can be software.