The Impossibility of “Indexing” Seed

In Startup Growth and Venture Returns, I suggested:

[A]t the seed stage investors would increase their expected return by broadly indexing into every credible deal.

It was pointed out that we were putting a great deal of leverage on the phrase “credible deal”. At the time I brushed this critique off, because we defined “credible deal” broadly and concretely: all the Seed deals on AngelList. How could we know anything about other deals?

Now, I think the result is still true, it’s just not useful. Specifically, the set of startups the market considers “worth funding” is not something that exists independently of investor behavior: it is created by the decisions investors make about what to fund. Put another way, there’s a “credible deal gap” that exists between “invest in every startup” and “invest in every startup that the market decides is worth funding”.

A recent product from AngelList lets us, for the first time on the platform, peer into the vast latent pool of startups that the market does not decide to fund. The size and quality of that pool makes me think that “indexing” Seed is impossible [1].

RUVs and the Latent Pool

Roll-Up Vehicles (RUVs) are fundraising infrastructure that allows a startup to bundle together a group of investors into a single cap table line, by “rolling them up” into one special-purpose vehicle. Importantly, the rollups.com/ruv web page that serves as the product’s front door is open to any startup.

As you may expect, there’s a tremendous number of unsuitable proposals that you get from an open front door on the internet [2]. Consistent with its venture-capital mandate, the AngelList rollups team filters RUV applications down to venture-scale companies. Importantly: only those startups can “start” an RUV on Rollups.

We can directly compare the pool of RUVs started by startups to the pool of SPVs that are brought to the AngelList platform by third-party deal leads after they have selected and diligenced the opportunity. In 2025 at the Seed stage, the pool size for started vehicles is roughly the same size for each set - “well over 1,000”. The following table shows the differences in valuations, wiring, and returns of these opportunities for 2025 Seed deals, with returns data as of August 1, 2026.

Note that about 1/3 of Seed-stage SPVs fail to wire, but this could happen for many reasons that have nothing to do with the viability of the underlying company (e.g., the deal lead loses their allocation, or fails to raise enough money for their specific vehicle). In contrast, about 3/4s of companies that start an RUV don’t end up wiring, and we believe more than 90% of the time that indicates the company failed to fundraise at all [3].

RUVs have an open front door, and many founders who start one do not ultimately raise money. But among RUVs that do wire, the investment outcomes appear at least as strong as those of lead-driven SPVs on AngelList. Both groups have a $25M median pre-money valuation, while RUVs have a slightly higher markup rate and average return. The meaningful distinction is between rounds that attract sufficient investor support and those that do not, rather than between startups using RUVs and venture-backed startups generally.

The sharp increase in valuation between RUVs that start and RUVs that wire suggests that substantial selection is happening during the fundraising process. But it’s hard to reproduce that selection from the fundraising materials alone: the work Seed investors do appears to be based on information that is not trivially observable ex ante. I had an AI agent carefully read through a 100-deck-and-text sample to try and surface obvious traits about founders, business models, etc. [4] The most interesting thing it flagged was that successful RUVs frequently have a “Why Now?” slide in their decks, whereas unsuccessful ones almost never pose that question. But on the whole, there appears to be no “weird trick” to distinguish a startup that will be venture-backed from a startup that could be venture-backed — and our RUV dataset suggests the latter pool is perhaps 3x larger than the former.

The “hard work” of Seed investing may therefore be less about spotting an obvious winner in a deck and more about exercising judgment and performing work at the margin: deciding which companies merit investment, putting reputational weight behind that judgment, and taking steps to help form enough capital for the company to exist as a venture-backed business.

The Two Ways Passive Indexing Breaks at Seed

Now we can assert the tension that wrecks “indexing” at Seed:

If your capital is marginal, you change the opportunity set. The RUV example on AngelList suggests a “big check passive indexer” would become the marginal source of capital for many companies whose rounds would not otherwise close. Intuitively - but obviously without any data to back this up - these investments would perform worse, perhaps much worse.

Making those investments would also swamp your budget. Suppose, generously, that the unfunded latent pool performed exactly as well as funded startups conditional on receiving money. The “big check passive indexer” strategy could still be unworkable economically because the act of indexing expands the index by several multiples. The more completely you index the marginal Seed opportunity set, the larger the opportunity set becomes. (This is a characteristic that makes Seed “indexing” fundamentally different than the S&P 500, which only has 500 companies in it no matter how broadly you choose to invest.)

On the other hand, if your capital is non-marginal, somebody else's selection determines your opportunity set. That is, if you only invest in opportunities that would be funded without your money, then you are trying to free ride on someone else’s discretion. Trying to squeeze into someone else’s round necessarily means you’ll be adversely selected. We documented this in our check sizing research, where we found that a GP’s small Seed checks outperform their typical or large Seed checks, consistent with the ideas that the best rounds ration allocation so the investor gets a smaller check precisely when demand from other investors is strongest.

One theoretical escape hatch is to make the indexer’s checks so small that they never affect whether a company gets funded: for example, invest $1,000 in every Seed round that is already going to close [5]. But this assumes that a passive investor can actually obtain a tiny allocation in a representative sample of rounds. In practice, allocation is itself selective. The companies with the greatest investor demand have the least reason to accept very small checks, so an investor attempting to remain permanently non-marginal is likely to face the worst adverse selection.

Seed investing is not merely ranking a fixed set of securities. Investors help determine which securities come into existence at all. The large selection effect between started and wired RUVs suggests that the fundraising process itself is performing a real economic function: converting a broad pool of plausible startups into a much narrower pool that other sophisticated investors are willing to finance. The RUV is not creating this selection problem; its open entry point simply makes a normally hidden stage of seed fundraising observable.

A passive investor faces an endogenous-opportunity-set problem. If they supply marginal capital indiscriminately, the population expands beneath them. If they wait for other active investors to define the population, they will receive systematically worse access to the most sought-after opportunities.

Seed can support broad diversification, but it cannot be reduced to passive indexing. Investors must perform the active function of deciding which opportunities become actual venture-backed companies.

Closely Related: Alpha at Seed?

Someone recently asked me if I thought “skill” in Seed investing looked more like a power law or more like a normal distribution. It’s a good question; I think the answer is probably closer to a normal distribution, but with a catch.

The conventional interpretation of skill is selection ability: picking better winners than everyone else. If that ability were heavy tailed, you’d see the top couple Seed investors run small concentrated portfolios that consistently do exceptionally well (think: 20+x gross returns).

But that’s not what we see: instead, my data-informed opinion is that two of the best Seed investors are Naval Ravikant and Charlie Songhurst, both of whom built enormous portfolios (think: 1,000+ startups). Access may be power-law distributed even if pure selection ability has much lighter tails. Put another way, a small number of investors may have the ability to get access to a large number of good companies, while no investor may be able to consistently identify a tiny number of extraordinary ones.

However, we’ve established here that a crucial part of Seed investing is deciding which opportunities from the enormous set of latent ideas should become venture-backed companies. That means “skill” at Seed may actually be the ability to consistently push good companies to cross from the “unfunded” to “funded” opportunity set. Under that definition, “access” and “ability” aren't independent, and the most-skilled Seed investor is the person whose participation changes other investors' behavior.

Footnotes

1. I’m ignoring two simple critiques of the indexing idea here. The first is that tracking any true “Seed index” would require re-weighting between assets in a way that would probably be impossible in practice, since the market to buy and sell freely doesn’t exist the way it does for public stocks. The second is that it’s operationally hard to hold lots of investments; which I don’t think is true anymore. If your fund administrator balks at building a 1,000-startup portfolio, you should probably find a new fund administrator.
2. One of my favorite unlaunched submissions was a proposal to raise significant equity capital to build "sustainable" tourism housing in a part of the world most people couldn't find on a map, and where similar glam-yurts appear to rent for $50 a night...if you can figure out how to get there.
3. Specifically, just under 5% of RUVs that did not wire ended up with a successful investment from another vehicle on the AngelList platform. On startup counts alone we believe GPs on the AngelList platform participate in at least half the Seed market, which would put the frequency of “The founder started an RUV and raised from real venture funds but not friends and family” at 10% or less.
4. I used AI for efficiency and to ensure any biases didn’t cloud the interpretation of the decks.
5. Writing small checks to lots and lots of credible deals, of course, is exactly the strategy implied by Startup Growth and Venture Returns.