Leicester City won the 2015-16 Premier League with 42.4% possession, the lowest figure of any English champion in Opta's records. They ranked second to last in the league in passing accuracy. In most matches they looked like the worse team on the field. They finished ten points clear of Arsenal.
The one attacking number where they ranked near the top was expected goals, where they came second, a fraction behind Arsenal. Expected goals, or xG, weights every shot by how likely a shot from that spot, at that angle, under that pressure is to end up in the net. A tap-in from six yards counts as most of a goal. A hopeful swing from 35 yards counts as almost nothing.
So a team that barely held the ball and could not string passes together generated the second-best set of chances in England. The statistic everyone watched on the broadcast said mid-table. The statistic almost nobody watched said champions.
That gap is the whole idea behind how we build a portfolio at Team Ignite Ventures, and it took me a while to notice that our own slogan gets it slightly wrong.
A small mistake in the phrase
We have written before about taking more shots on goal in a power-law world. In soccer, that phrase is redundant. A shot on goal and a shot on target are the same thing: a shot that would go in if the keeper did not stop it. The phrase migrated into business writing from hockey and basketball, where a shot on goal just means a shot attempted, and somewhere in the migration it lost its teeth.
The sloppiness matters, because the sloppy version is the version people argue against. "More shots on goal" sounds like a defense of spraying capital at everything with a deck. The precise version says something much more demanding. Get the ball into a position where the shot has a real chance of going in, and then do that as many times as your organization can without lowering the quality of any individual attempt.
Those are different strategies. One is easy. One is an organizational design problem that takes years.
The chances are what predicts the result
The data is unusually clean on this point. A 2026 study in BMC Sports Science, Medicine and Rehabilitation looked at six seasons of the Turkish top division, 2,728 team observations across 1,364 matches that ended in a win or a loss. Each additional shot on target was associated with a 58% increase in the odds of winning, with an odds ratio of 1.582 and a tight confidence interval. Shooting accuracy separated winners from losers more strongly than raw shot volume did.
Two honest caveats. The relationship is associative, and good teams get more shots on target partly because they are good teams. And a single league is a single league. But this finding shows up across World Cup samples, the Greek league, and the English reserve leagues with enough consistency that I would put high confidence on the direction and moderate confidence on the magnitude. Chance quality predicts results. Possession, passes, and territory are decoration by comparison.
Confidence level on the analogy holding in venture and I will come back to what would prove it wrong.
Venture plays this game with one rule changed
Soccer is interesting because goals are scarce. One deflection settles a match. A team can dominate for 90 minutes and lose to a set piece. Under that kind of scarcity, creating more real chances is worth an enormous amount, because each chance is a fresh draw from a low-probability distribution.
Early-stage venture has the same scarcity, then adds a rule that would break soccer entirely.
Not every goal counts for one point.
One investment returns three times the money. Another returns thirty. Another returns three hundred. Once in a great while, one returns a thousand or more, and that single line item pays for every miss in the fund and then keeps going.
Imagine playing soccer where almost every shot misses, most goals count for a point, and once every few hundred attempts someone scores a goal worth a thousand points. Nobody in that sport would build a strategy around clean sheets and possession. You would organize your entire club around manufacturing quality chances at volume, because your season is decided by whether you take enough good shots to be standing in front of the net when the thousand-point one goes in.
That is the asset class. Team Ignite is built for it on purpose.
The strongest argument against us
Here is the case a good investor makes against this strategy, and it deserves to go first because it is mostly correct about most firms that try it.
High deal count in venture is usually a symptom of weakness. Capital is abundant and access is scarce. The best rounds are oversubscribed before they are announced, and allocation goes to firms the founder wants on the cap table. So a firm doing a hundred deals a year is often the firm that cannot win an allocation in the twenty deals that matter, taking the leftovers instead. Volume looks like strategy and functions as adverse selection. The deals available to you in unlimited quantity are, by definition, the ones somebody else passed on.
The second objection is arithmetic. Small checks across hundreds of names produce small ownership. Run round numbers. A couple hundred thousand dollars into a company priced in the high teens of millions buys you roughly one to two percent, and dilution across subsequent rounds cuts that meaningfully by the time anyone exits. A ten billion dollar outcome then returns tens of millions of dollars on that line, not hundreds. The concentrated manager's conclusion follows naturally: your ownership is too small to matter, so you need an outcome so rare you will never see one.
The arithmetic in that objection is right. The probability judgment inside it is where I think it fails.
Everything depends on which distribution you are drawing from and how many draws you get. Twenty-five draws from the industry-average pool almost never contains a decacorn. Four hundred draws from a pool that has already survived one brutal filter and then a second one is a different question entirely, and it is the question our portfolio simulations exist to answer. Holding underlying quality constant, the probability of catching the extreme right tail rises sharply with position count. That is not a claim about being clever. It is a claim about sample size against a fat-tailed distribution, and you can run it yourself in our Power Law Simulator with your own assumptions about, entry price, dilution, and outcome rates
There is also an asymmetry with no equivalent in soccer. A missed shot in soccer costs you nothing except the counterattack. A bad check in venture costs you exactly one unit of capital, and that is the floor. A shot you never take, in a company that becomes the outcome of the decade, costs you the entire fund's return. Bounded downside on the attempt, unbounded downside on the abstention. When the penalty for absence dwarfs the penalty for error, you take more attempts.
The adverse selection point still stands as the thing an investor should press hardest on, and I will get to how to test it.
Expected goals with no ceiling
The useful version of xG for venture is not "how likely is this to work." It is "how likely is this to be enormous, weighted by how enormous."
Soccer's xG is bounded. The best shot in the sport, an open net from two yards, is worth a maximum of one goal. The gap between a great chance and a poor chance is a factor of maybe twenty.
Venture has no ceiling on a single shot. The gap between a decent investment and a great one is not twenty times, it is thousands. Which means the shape of the problem is different from anything a soccer manager faces. In soccer, a fixed number of possessions per match caps how many chances you can create, so efficiency per possession is everything. In venture, the constraint on chances created is organizational, and organizations can be rebuilt.
So the real competitive question for a firm is narrow and slightly uncomfortable. How many high-quality shots can you manufacture without degrading the quality of each shot?
That is a sourcing problem, an access problem, a selection problem, and an operations problem at the same time. Firms that solve two of the four end up with volume and no quality, or quality and no volume. Two firms in particular have solved all four, and it is worth being precise about how.
Y Combinator built the biggest funnel in the business
Y Combinator states that more than 10,000 companies apply every three months and that it accepts roughly one percent. YC now runs four batches a year, so the machine processes north of 40,000 applications annually to fund somewhere between 150 and 200 companies per batch. Recent batches have run below one percent. Summer 2025 was reported at 0.6%, the lowest on record.
That is an enormous funnel followed by an enormous filter, and the two are inseparable. YC has funded well over 5,000 companies, including more than 100 now valued above a billion dollars, and every one of those outcomes makes the next application cycle stronger. More successful alumni attract more applicants. More applicants deepen the selection pool. A deeper pool produces more successful companies. YC has said explicitly that larger batches make the network more valuable to the founders inside it, since each founder gains more peers, more early customers, and more people who have already solved the problem in front of them.
Scale is an input to that system, not a byproduct of it.
a16z built the other kind
Andreessen Horowitz attacked the same problem from the opposite direction. It pioneered the platform model, building a large organization of operators around the investing partners: recruiting, go-to-market, marketing, policy, finance, legal. The firm describes itself as having over $100 billion under management and the largest team of operators in venture. Independent estimates of its regulatory assets under management sit closer to $90 billion after its January 2026 raise, which is a distinction worth knowing but does not change the point.
The point is that all that infrastructure functions as an access engine. If you are connected to thousands of founders, executives, customers, engineers, and co-investors, you get more looks, and better ones. Successful founders refer other founders. Executives spin out companies. Operators see emerging problems a year before investors write about them. Customers tell you what they are buying before the numbers show up in a deck.
The organization produces signal, signal produces investments, investments expand the network, and the network produces more signal.
What we do
We are not going to outspend a16z or rebuild YC. Our advantage is narrower and, I think, more durable at our size. We sit on top of several high-quality funnels that already exist, and then we add a filter of our own.
YC is the clearest example. Instead of starting with the tens of thousands of companies that apply, we start after one of the most competitive screening exercises in the world has already run. Then we screen again, targeting roughly the top decile of what we evaluate.
Look at 2026 so far. Winter 2026 had 199 companies in the public directory and we backed 16. Spring 2026 had 193 and we backed 17. That is 33 investments out of 392 companies, a little over eight percent, from a pool that was itself drawn from more than 20,000 applications. Those are not 33 startups pulled off the internet. They came through YC's filter, then ours.
YC is one source of shots. The rest of the system is five loops that feed each other: our AI and data infrastructure, our founder community, our investor community, the broader Team Ignite network of operators and advisors and executives, and the Ignite Podcast. The podcast is the least obvious of the five and one of the most useful, because it builds a real relationship with someone eighteen months before they ever send us a deck. Trust creates access. Access creates better data. Better data improves selection. Good investments expand the network. Then the loop runs again.
Hundreds of portfolio companies. Thousands of co-investors. Sixteen thousand people in the community. Those numbers are only interesting as surface area. Every credible person in the network can surface the next company, the next customer, the next executive, or the next category before it has a name.
The throughput constraint moved
There used to be a real operational argument against all of this, and it was a good one. A partnership can only read so many decks. Partners can only take so many calls. Past some volume, deal count and diligence quality traded directly against each other, and the trade was unavoidable.
That constraint has moved. We see roughly 10,000 early-stage companies a year across our sourcing funnel, and our systems rank, extract, compare, research, and flag before a human reads anything. What comes out the other side is a structured queue where human attention goes to the decisions that require judgment: the founder, the context, the negotiation, the final call.
Team Ignite runs as a solo general partnership with agentic automation doing the reading. People want to talk about that part, and it is genuinely the reason a portfolio of this shape is operable today when it would have been organizationally miserable ten years ago. It is also not the interesting part. The interesting part is the network that tells you which of the ten thousand deserve a human hour.
What would prove this wrong
The honest test is not deal count. Deal count is the input we control, which makes it the least informative number we publish.
If the adverse-selection critique is right about us, it will show up in two places. Our companies will struggle to raise strong follow-on rounds, and the investors leading those rounds will be unremarkable. A portfolio that gets marked up by good firms at seed and Series A is drawing from a healthy distribution. A portfolio that stalls at the first extension round is a portfolio full of leftovers, no matter how large it is.
The second test is graduation rate against YC's own base rates for the same batches. If we are picking well, our slice should outperform the batch we picked it from. If it tracks the batch, we are an index fund with extra steps, and an index fund should charge index fees.
Those are the numbers I would want to see if I were on the other side of the table, and they are the numbers we track.
Play the game the distribution gives you
If venture returns were normally distributed, none of this would make sense. You would concentrate, pick twenty companies, optimize ownership, and hope you are an exceptional picker. That is a coherent strategy for a normal distribution.
Early-stage venture is brutally power-law distributed. A tiny number of companies create most of the value in the asset class, which means missing one extraordinary company hurts more than being slightly wrong about forty ordinary ones. Concentration is a bet that your judgment is precise enough to find that company in twenty or thirty attempts. Ours is a bet that judgment plus attempts beats judgment alone, and that the second lever has been sitting there underused because the operating cost used to be prohibitive in a pre-agentic AI era.
PitchBook's Q2 2026 Global League Tables ranked Team Ignite as the third most active early-stage investor in the world, behind Andreessen Horowitz and Y Combinator. That ranking is evidence the funnel is running, not evidence the strategy works. The strategy works if the outliers are in there.
Because in soccer, the team creating more good chances usually has an advantage over a season. In venture, the same advantage compounds into something much larger, since every soccer goal is worth one and every venture goal is not.
Occasionally one shot decides everything.
If this is close to how you already think about the asset class, the deck walks through the whole model: where the shots come from, how the filter works, what the position count is, and the math underneath it. You can view it here. For the underlying portfolio construction work, the Power Law Simulator and our original research on volume in a power-law world are both worth an hour.