What this covers
Patrick interviews Alex Mittal of FundersClub, a venture platform that combines a large investor network with a centralized, partnership-like investment bar. The conversation starts from the Power Law, the fact that nearly all venture returns come from a handful of companies, and works outward to how such companies get missed. Google's founders tried to sell search for a few hundred thousand to $1.6 million, and the buyers misjudged it too. GitLab's open-source model and fully remote structure first drew a no from FundersClub, which reversed after the founder call.
The middle of the conversation covers method. Mittal says quantitative screening works only for repeatable models such as enterprise SaaS and consumer marketplaces, where metrics like churn, LTV, CAC, and cohort retention are comparable. Useful data needs both scale and trust, a combination he says most firms lack. FundersClub's model predicts a company's Term Sheet valuation before the term sheet arrives. The more contested claims are that the Consensus Heuristic misleads when hunting for frontier opportunities, that "first principles" is often just a label for an opinion, and that the market matters more than the team. He ranks relentlessness as the top founder trait, citing Slack's and Wonderschool's pivots, and contrasts Instacart with Webvan.
The rest ranges across sectors. On crypto, he points to Coinbase growth that ignored Bitcoin's price and dismisses the Silk Road objection, then discusses trustless computation and Ethereum. He also covers compute swinging between CPUs, GPUs, and ASIC hardware, computational materials science at OTI Lumionics, self-driving trucks that start on highways, and why hot food delivery stays hard. Even with careful selection, he says, most seed investments fail.
Mittal argues venture returns come from a few outlier companies that look like bad ideas early, and that FundersClub systematizes finding them by pairing a large trusted investor network with data-driven, rule-based screening, while staying skeptical of consensus, first-principles claims, and hype.
- Venture returns concentrate in a handful of outlier companies, and early-stage outliers are systematically misjudged by founders, buyers, and investors alone
- Quantitative screening only works where repeatable business models produce comparable data, and it requires both scale and trust in that data
- Founder success depends heavily on market selection, with relentlessness and curiosity as the key personal traits
Venture returns come from rare outlier companies that consensus misjudges early.
- The very best companies do not look like the very best companies in their early days; the intersection of ideas that seem bad but are actually good is where outlier opportunities sit, and that intersection becomes obvious only slowly, if at all.
“the very best things do not at all look like the very best things in the very early days”
- Paul Graham observed that if Darwin had published On the Origin of Species today he might be fired, illustrating how controversial a now-accepted scientific idea was when first proposed.
“if Darwin had published the origin of species today, he'd probably be fired from his job because that's how controversial it was at the time”
- All of the returns of the venture industry come from a handful of companies, and this outlier effect is the gravity and the laws of venture capital that any firm must systematize around.
“Basically, all of the returns of the venture industry come from a handful of companies, and it seems like that's the gravity of the industry”
- The human brain is hardwired to treat consensus as good, which is usually right but is wrong when specifically selecting for frontier, boundary-pushing opportunities, so investors must deliberately ignore popular opinion and expert consensus.
“your brain is just hard wired to assume that consensus is good. And for many things, it is.”
- Even when the firm backs companies that later become multi-billion-dollar companies, the power law dictates that most seed-stage investments do not end this way, so outcomes are frequently unsuccessful despite careful selection.
“the power law dictates that frequently, most of the time in fact, it doesn't always end up that way”
- Vitalik Buterin sold 25 percent of his Ether stake at $10 per ether in 2016 for diversification, and waiting about 18 months would have yielded roughly 30 times more value, illustrating how hard future value is to quantify even for insiders.
“sold 25% of his stake of Ethereum at $10 per ether”
- Google's founders tried to sell their search technology to Excite and Lycos for a few hundred thousand to $1.6 million, and even the buyers misjudged the technology, showing innovators and acquirers both struggle to assess future value.
“the innovators themselves even that misjudge the future value of that technology”
- When Facebook had a $2 billion market cap, many observers thought it was overvalued, with reporters urging Mark Zuckerberg to accept Yahoo's acquisition offer, which shows how outliers are consistently misjudged even as time passes.
“when Facebook had a market cap of $2 billion, if you go look at the press from that time, most people did not think it was worth that much”
Quantitative data and models can systematize venture investment decisions.
- FundersClub uses a model that takes parameters for repeatable business models and predicts the valuation a company will receive on its term sheet before the term sheet is issued, and the model has been remarkably accurate.
“we can tell them what their valuation is going to be on the term sheet that they're going to get before they've gotten it. And it's been remarkably accurate”
- Useful private-company data requires both scale and trust: many VC firms have scale without the founder trust needed to get real KPIs, while thoughtful, trusted investors lack the scale to derive formulas from data points.
“Basically, you need a lot of data, but you also need trust, right? So you need scale and you need trust.”
- Quantitative venture analysis is feasible only for repeated business models such as enterprise SaaS and consumer marketplaces, where metrics like ACV, retention, churn, LTV, CAC, and cohort retention are comparable across companies.
“certain business models are repeating. Breaking apart SaaS, what are example metrics”
- Investors with quantitative data can apply it to portfolio companies as a non-dictatorial coaching tool, showing founders how their retention or churn compares to top-decile peers, which founders often lack.
“there's a value to investors if they can... applying that to their portfolio as a tool. Not in a dictatorial way”
- Moats are a key factor that cannot currently be quantified, so investors rely on a qualitative semantic tiering of companies into strong moats, weak moats, and false moats that people mistakenly believe are moats.
“A good example is moats. If we could quantify moats, that'd be awesome. We're working on it.”
A centralized investment standard lets a large network operate like a partnership.
- FundersClub's algorithmic process is designed to remove emotion from investing, but a partner's sleeplessness led the firm to override a no decision on GitLab, showing the process is almost but not fully emotion-free.
“Boris couldn't sleep that night and I got a phone call from him”
- FundersClub's network model uses a centralized standard for investment qualification, closer to a partnership in decision rigor, and the most successful firms in venture actually resemble partnerships more than network platforms.
“if you look at the most successful firms, they actually much more closely resemble the latter than the former”
- FundersClub has backed about 300 startups since Q4 2012, mostly at seed stage, using a network of over 20,000 accredited investors and several hundred entrepreneurs, and publishes its returns publicly every quarter as a voluntary standard.
“We publicly publish our returns and update them every quarter. There's no rule requiring us to do that.”
- About 70 percent of FundersClub's portfolio is pure software, another 10 to 15 percent is software touching the physical world, and only about 10 percent is hard-science investing, which the firm approaches very carefully.
“I would say that something like 70% of our portfolio is strictly pure software.”
- Predicting the next multi-billion-dollar generation-defining companies is too uncertain for a consensus-based decentralized approach, so FundersClub combines a large network with a centralized, standardized investment bar closer to a traditional partnership.
“It's really difficult to do with a consensus decentralized approach.”
- Alex Mittal and Boris Silver started FundersClub around 2012 to ask what a venture firm fully embracing technology would look like, combining the lessons of successful VC firms with a community and co-op model for sourcing, vetting, and adding value.
“what would a VC firm look like if it was almost more of a community and co-op powered model”
- True first-principles analysis is very difficult, and the phrase is often used as a precursor to an opinion rather than as a disciplined method, but if done correctly it can reveal the future.
“I see it thrown around, especially in this town as a precursor to an opinion”
Strong markets matter more than founder qualities for startup success.
- The most important founder trait is relentlessness, which paired with a strong market produces success, as shown by Wonderschool's pivot from Instagram merchant tools to preschool marketplaces and Slack's pivot from a failed game.
“the number one trait that matters the most is relentlessness”
- Instacart succeeded where Webvan failed on-demand delivery because Instacart's model was a pure software business with lack of physical assets, a totally different asset structure, which Mittal argues is why it was misjudged as a repeat of Webvan.
“Failing to understand the nuance of actually totally different business model, totally different asset structure, like lack of assets, actually, pure software business”
- A strong market can make a team that lacks many founder qualities successful, while even a perfect team fails against a weak market, so founders should select markets carefully because the market's force dominates.
“the market can make a team that maybe is lacking in many dimensions... incredibly successful because the market is on fire”
- Brian Armstrong of Coinbase exemplifies founder ability to learn and be curious, absorbing what he reads and asking inquisitive questions that get stored and reused, similar to what David Swensen described at Yale.
“he has a next level ability to soak up like a sponge, things that he reads and implement them”
- Amazon's dominance made e-commerce plus tech look like a winning bet after the branded e-commerce failures, creating a forcing function that opened multi-billion-dollar opportunities at every link of the supply chain from port to store to online sale.
“there's a multi billion dollar business opportunity for multiple businesses in each of those categories”
- On-demand hot food delivery failed because unit economics require computer-science-level algorithmic mastery and restaurant operations are inherently difficult, so consumer NPS can be high while the business remains very hard.
“it's incredibly difficult to the point where you need a computer science degree and a mastery of algorithms and data to get the unit economics correct on on-demand delivery”
Crypto has real utility that is separate from token price.
- Early crypto projects such as MakerDAO's DAI stablecoin, Golem's decentralized compute network whose usage rose five-fold in a year despite roughly 90 percent price decline, and DocuSign's Ethereum e-signature integration show early real-world adoption.
“Golem, the decentralized compute network, usage has 5xed in last year despite the drop in price, like 90% or whatever it has been”
- Mainstream press frequently cannot separate crypto price from utility, as shown by a Financial Times headline questioning Ethereum utility as the price fell, and this bias is easy for anyone to fall into.
“it captures the inability to decouple price from utility in the mainstream press' mind”
- In 2012 when Bitcoin traded near $6, the Silk Road association was irrelevant to whether crypto infrastructure was foundational, since US dollars are likely the most common currency for drugs and weapons too.
“The fact that the Silk Road exists is completely irrelevant to this decision of is this industry, is this technology potentially foundational for the future?”
- Coinbase user growth continued week after week regardless of Bitcoin price, which Mittal treated as evidence that the technology provides real utility independent of speculative price moves.
“no matter what the price was doing, their usage was growing week after week”
- Blockchain enables trustless transactions and trustless computations, a foundational capability that did not exist before, and Ethereum's decentralized trustless interaction has applications across many domains, though its eventual uses are unknown.
“with blockchain technology, you can actually have trustless transactions, trustless computations”
Unusual or boring-looking businesses can hide valuable outlier opportunities.
- GitLab's open-source model and fully remote company structure looked like bad business ideas at the time, yet GitLab became a billion-dollar company with hundreds of remote employees, and FundersClub reversed an earlier no after the founder call.
“this was the one investment where we reversed our decision”
- Boring-sounding businesses such as freight logistics, Flexport, and Shippo can be attractive, as Mittal contends that the firm likes boring things as well as crazy ones and that logistics software has large value.
“we like boring things too. Flexport until later on, it was really boring of a sounding company”
- Self-driving truck technology should first target the linear, predictable highway portion of trips where most value lies, rather than the hardest city-driving problem, and trucking is one of the most dangerous jobs with a high fatality rate.
“In a strange way, there was all this focus on the hardest possible problem at the earliest stages of the technology”
- Gecko Robotics builds robots that climb boiler walls in power plants, replacing scaffolding and manual inspections that are expensive, dangerous, and require shutting down plants for about a week, which shows boring-seeming opportunities can be valuable.
“Inside of every power plant that say is a gas or a coal power plant, there's boilers that once a year have to be shut down”
- Shippo, which helps e-commerce companies handle shipping and reverse logistics, showed double-digit weekly growth, revealing a lack of infrastructure for mid-market e-commerce players that Shopify's platform was helping to create.
“a company called Shippo, which was seeing some pretty eye-popping numbers on their growth, week over week, double digit growth”
Computation and new hardware open frontier hard-science opportunities.
- Materials science has historically been empirical, but computing power and new algorithms now allow theory-driven computation of materials, which accelerates discovery, as OTI Lumionics does with quantum and classical algorithms for display and lighting technology.
“The history of a lot of material science is more empirical-based”
- Compute hardware has swung from general-purpose CPUs to GPUs to specialized ASICs and back toward specialized hardware, as Bitcoin mining acted as a petri dish for extreme demand for efficient compute that now resembles machine learning demand.
“It's swinging back to specialized hardware.”
- Neurable's non-invasive brain-computer interface combines proprietary technology with a different algorithm to achieve roughly 20 times faster speed or lower latency, and it needs only about a minute of training to acquaint the device with a user.
“That combination paired with a different type of algorithm than has traditionally been used is leading them to have 20 fold faster speed or 20 fold lower latency”
- Investing in specialized hardware is risky because if the algorithm falls out of fashion the hardware becomes worthless, so the open question is whether custom hardware can be redesigned quickly as algorithms change.
“if you invest in hardware and that algorithm goes out of fashion, seems like you're kind of screwed”
- Notable Labs uses software-driven methods to test thousands of drug combinations, approved for one illness and not others, to find therapies that were previously infeasible at this scale.
“Notable Labs that is able to effectively test thousands of combinations of drugs”
FundersClub avoids token purchases and speculative crypto pricing, investing only in equity.
- FundersClub made zero blockchain investments in 2017 and 2018 despite being bullish, pausing when speculation peaked because the price was crazy, and it had invested in blockchain equity (not tokens) starting in 2012 with a peak in 2016.
“We did zero blockchain investments this year. We did zero new blockchain investments last year.”
- The duality of believing in a technology while declining to invest because price is irrational is a mental stance investors must accept, understanding both sides of a position and being comfortable with the weirdness.
“It takes almost like believing, understanding both sides and being okay with that weirdness.”
- The firm almost never anticipated e-commerce and logistics trends before the data revealed them, and it first invested in blockchain in 2012 before stopping when others piled in during 2017 and 2018.
“Almost without exception, we've never anticipated these trends before the data told us that they were happening”
- FundersClub has never purchased cryptocurrency tokens and invests only in equity of blockchain companies such as Coinbase and Chainalysis.
“we've never purchased tokens. We don't own any cryptocurrency. We've only ever invested in equity”