What’s the Bet: TRAC VC
Can a group of outsiders disrupt venture capital?
Author’s note: This is the latest in my “What’s the Bet” series. AI has the potential to change everything, and founders have to make company-defining choices in response. That is not easy. In this series, companies pay me to analyze their bet and explain it to you. I retain complete editorial independence. Today’s bet is brought to you by TRAC, a venture capital firm where machines make investment decisions instead of people. If you have any feedback, just reply to this email.
If you wanted to build the perfect venture capital fund from scratch, you probably wouldn’t start with a 76-year-old former hedge fund manager, a serial entrepreneur whose biggest hit was an e-greeting card company, an ex-FBI agent, and a data scientist in El Paso. But this same group of people has had two consecutive top decile funds in a row. It’s worth asking how their thesis is paying off.
TRAC’s bet radically against Silicon Valley norms. They believe that the venture capital industry can be beaten with math. Typically when you hear a firm saying something trite like that, they mean using data as a supplement to gut instinct. TRAC means using math instead of people. Their models make the investment decisions, and the humans’ jobs are to convince founders to take the money the machine wants to give them. And somehow, it is working.
I had many calls with the co-founders over the last several months to figure out how. Here's what I found.
How the Machine Works
We should start with the data.
Many other venture firms employ data scientists. Some of them are even pretty good! But what they do is fairly, uh, droll. The data scientist will build a model, present it to the investment committee, receive a cookie and a pat on the head, and then the managing partner makes the call based on their gut. The data is a (mostly ignored) input for the human decision-maker.
TRAC doesn’t work like this. As their data scientist, Dr. Steve Marek — a former lecturer in biomedical engineering who came through grad school at UT Austin — put it to me: “Our data is our investment team.” The models decide, the people don’t. Steve told me he can’t even comment on individual companies because they’re just numbers to him. He doesn’t look at the product, the pitch deck, or the founder’s LinkedIn. He just looks at the score.
The closest parallel in financial history is Renaissance Technologies, where physicists and mathematicians who didn’t care about Wall Street’s traditions built quantitative models that crushed human stock-pickers for three decades. When I brought up Rentech to Steve, he didn’t flinch. His response was that not having preconceived notions about how venture is “supposed to work” is actually an advantage. None of TRAC’s team came up through the traditional VC pipeline. And so far, that has been to their advantage.
The foundation of everything TRAC does is a proprietary data infrastructure built over five-plus years. They pull from CB Insights, PitchBook, Crunchbase, SimilarWeb, and other sources, then merge them into a single source of truth. No individual database has complete or accurate information so TRAC has spent the last six years building their own taxonomy, cross-referencing and cleaning data until they had something wholly unique. The result is a unified dataset that merges over 650,000 investor entities, 1.25 million company entities, and 2.1 million deal entities.
That dataset is what makes everything else possible.
Fund I was the proof of concept. Joe Aaron, TRAC’s co-founder, had spent years manually combing through PitchBook data and noticed something: when he looked up the cap tables of companies that became massive, the same individual investors kept showing up. You’ll note how it is not always the same fund. It was the same *people*. He realized that there was an elite group of investors that no one had recognized. He brought in Steve to formalize those patterns. Fund I launched with two algorithms — SuperTRACer and the Investor Group Quality score (IGQ). About two-thirds of the way through the fund’s deployment, Steve layered in three more. It (perhaps surprisingly) worked. Fund I became a top-decile performer. Two of the first eight investments became unicorns. After nearly six years, Fund I’s loss rate is 7%.
Fund II deployed all five algorithms from day one, each filtering for a different dimension of signal:
The most important is the Investor Group Quality (IGQ) score. TRAC identifies 286 individuals globally who sit in the top tenth of one percent of early-stage investors, ranked by track record in specific industry sectors. So while Sequoia may have a great reputation, TRAC told me that only 5 of their partners qualify as top individual pickers by TRAC’s scoring. (Just because an investor sits at a prestigious firm doesn’t mean they are any good.) The core thesis for this algorithm is that it’s hard to fool several great investors at the same time. When multiple top-tier individual investors converge on a single company, that signal is predictive.
From there, companies pass through SuperTRACer and three additional algorithms covering profitability projections, market momentum, competitive positioning, and industry growth rates. Only 100 to 150 companies a year will clear all five thresholds. Of those, the models predict roughly 1 in 5 will become a unicorn and 2 in 3 will at least return invested capital.
Frankly, I went in a bit skeptical that this would work. But the results are compelling. The team told me that Fund II’s loss rate sits at 5%, down from Fund I’s 7%. Fund I saw 19% of its portfolio heading toward power-law returns. Fund II is at 38%. The math keeps beating the gut instinct of other investors.
The Scoreboard
Joe Aaron has a law he likes to cite. He calls it Menachem’s Law, and the logic is pretty simple: a startup either raises a follow-on round, gets acquired, or dies. There isn’t really a fourth option. If a company cannot attract new capital after its initial raise, it will go out of business. That makes the follow-on rate, what TRAC calls the “graduation rate,” the single most revealing metric for evaluating a venture portfolio. A good fund gets 50% of its companies to raise again. A great fund gets higher.
TRAC Fund I achieved 76%.
According to Preqin, there are 503 early-stage funds in the 2020 vintage. TRAC Fund I ranks #1 on the graduation rate among all of them. And that ranking holds no matter how you slice the comparison group: among funds with 46 or more investments, TRAC is #1 out of 30. Among funds with more than 10 investments, #1 out of 267. Thirty-five of 46 Fund I portfolio companies have raised follow-on capital. Companies that went out of business represent less than 12% of Fund I’s assets under management. The fund turns six in May.
Fund II is still early but the pattern is holding. Vintage 2022, 59 portfolio companies, 30 of which have already raised follow-on capital — a 49% graduation rate, with the fund’s final initial investment only completed in December 2025. Against Preqin’s 698-fund 2022 vintage universe: among funds with 59 or more investments, TRAC ranks #1 out of 16 on graduation rate. Among funds with more than 30 investments, TRAC sits in the 96th percentile — #3 out of 76.
So does the machine actually work? Beyond the loss rates, the aggregate numbers tell the story. Fund I’s 46 portfolio companies grew combined revenue from $48M to $309M over about three years. Fund II’s portfolio has grown from roughly $40M to $389M in half the time.
So if this all looks so good, why isn’t TRAC drowning in yachts and second homes in Tahoe? The problem is getting founders to meet with them.
The Distribution Problem
Most venture funds have the opposite problem from TRAC. They can get meetings. They have the brand, the network, and the outrageous parties up and down the SF peninsula. What they struggle with is picking. Their portfolios are full of pattern-matched bets that looked right in the pitch and went sideways in the market. TRAC has the inverse problem with a selection engine that works and a brand that doesn’t yet match it.
Perhaps most challenging is that competition to get into deals is built inherently into their model. If a company has multiple A-tier early-stage investors on the cap table, then they’ll have a much easier time raising a follow-on round. By nature of their models, every time they send in a term sheet, the founder will likely also have emails in their inbox from Kleiner Perkins, Founders Fund, Lightspeed, and every other top decile fund. That’s a tough lineup.
TRAC writes $250K checks at seed, $1M at Series A, and up to $3M at Series B — and never takes more than 20% of a round. They’re not trying to lead which means a founder has to make space for them after a deal starts to come together.
This is where Fred Campbell and Brant Meyer come in. Fred and Brant are TRAC’s two investing partners, the humans whose job is to convince founders to take the money the machine wants to give them. To help increase their hit rate, they pitch the founders a unique dashboard.
Over the last five years, TRAC has built a proprietary intelligence dashboard for portfolio companies that uses their investment models to generate valuation modeling, competitor mapping, investor overlap analysis, and what Fred calls “the do-not-call list” — flagging investors who’ve backed competitors and might be fishing for diligence rather than genuinely interested. Founders use TRAC’s data to negotiate higher valuations with lead investors. In at least one case, a lead investor backed down and accepted TRAC’s valuation number over their own.
Fred and Brant claim that if a founder agrees to a Zoom with them — just one call where they walk through the dashboard — TRAC gets on the cap table more than 95% of the time.
Nobody else at their stage offers this, because nobody else has spent five-plus years building the data infrastructure to produce it. The dashboard transforms TRAC from “random small fund writing a small check” to “the only investor who will hand you a competitive intelligence product on day one.” Distribution problems yield to time and evidence in a way that selection problems never do. If you can pick, then eventually, the brand will follow. To help speed it up, you could also sponsor dashing writers to examine your business in their newsletter. If you can’t pick, no amount of brand will save you.
Where I Think This Breaks
I’ll be honest about the risk I see, because I’ve written about it before.
I wrote a piece for Every called “Venture Capital is Ripe for Disruption,” where I argued that the fundraising environment has become structurally hostile to new venture capital firms. Fund sizes have ballooned to the point where even billion-dollar exits are meaningless for the largest firms.
That structural pressure creates a gravitational pull toward massive fund sizes, because LPs want to write big checks into big funds that can deploy hundreds of millions a year. A fund managing $73M across two vehicles, no matter how good its returns are, sits outside the LP comfort zone. The institutional capital allocators at pension funds and endowments aren’t set up to evaluate a data science team. They want a name-brand GP with a billion-dollar fund and a 10-year track record. TRAC is five and a half years into a 20-year plan, and the structural forces of LP capital allocation are not on their side yet.
Additionally, the team is a bunch of outsiders, and in the clubby environment of Silicon Valley, that is a disadvantage that is hard to break.
What I Actually Think
After many conversations with this team, here’s where I land.
The established firms won’t adopt this approach because it would destroy the mystique. Joe Aaron, a cofounder, told me a joke: “The data scientist went into the investment committee meeting. He had only two questions from two partners. One wanted him to fix the printer and the other to help him with his Internet connection.” The industry isn’t ignoring quantitative methods because they don’t work. They’re ignoring them because acknowledging they work means admitting the magic is more fluff than fundamentals.
Three unlikely people in unlikely places built a machine that picks better than most venture capital firms. Now they have to convince everyone to care. Five and a half years in, with a #1 follow-on rate, a 5% loss rate, and a growing stack of unicorn receipts, the machine is making that argument harder to ignore.