Greg Rosen

The through-line

Rosen's whole argument rests on a single admission most investors will not make: venture has gotten harder, and he does not know how to get better at picking. "Maybe I'm just not good enough at this job. I don't know how to be 50% better at picking, let alone 2x, let alone 10x. I just think it's like a fool's errand to think that you're magically going to be become better pickers" [1]. If seeing, picking and winning are the three variables that produce enterprise value, and competition has risen while none of the variables change, then accepting flat inputs is accepting worse returns [1]. So he attacks the one variable he believes is actually movable: seeing. Everything else in his model, the refusal to lead rounds, the check sizing, the calendar, the outbound machine, follows from that choice. He calls it a "venture humility score," a deliberate inversion of the venture arrogance score, and defines it plainly: "we don't know how to have a basket of 20 companies in you know a fund or you know in a year and pick the most important ones given how competitive and hard it is. So for us we're saying we need to do more we need to see more" [1].

The second through-line is that collaboration is a structural commitment, not a temperament. Staying neutral requires permanently forgoing the most obvious route to scale, and Rosen treats that trade as non-negotiable rather than case-by-case [1].

On why collaborative venture has to be anti-scale

"Collaborative venture is hard. Um, and almost by definition, it has to be anti-scale in certain dimensions" [1]. The easy path to scaling capital is to start leading rounds and write bigger checks, which is precisely what dissolves neutrality. Rosen describes the moment as binary: a firm making 80 investments a year that leads two or three of them poisons every subsequent deal share, because the counterparty is left wondering whether they are seeing everything or only the deals the firm did not want for itself [1]. "The second you start leading those rounds, you just can't be Switzerland" [1]. Asked whether a genuinely exceptional, differentiated seed opportunity would justify breaking the model once, his answer is flat: correct, it is not worth it [1].

He is clear-eyed that this is a trade rather than a free lunch. The standard knock is that you need to own a lot of the companies that work, and he concedes the point while reordering it: "that is true, but also you got to be in the right companies. And it's much harder to be in the right companies. And so what we do is we trade ultimate ownership percentage points for a more collaborative, flexible model that lets us just see better companies and hopefully get into more companies" [1]. Roughly a quarter of deals, maybe slightly more, come from sharing with other investors, and the reason people share openly is that BoxGroup will not snipe the deal or compete for the lead position, plus they want to be the preferred partner introduced when a portfolio company raises its next round [1].

On scaling velocity instead of AUM

There are only two ways he has seen to scale this craft: grow AUM and move up the stack into leading seeds, As and Bs, or scale the number of early deals under a collaborative structure [1]. BoxGroup chose the second, and the cost is workload rather than capital. "If you took a look at any one of the people who work at Box Troop's calendar, the sheer number of net new deals is ridiculous," and he suspects most investors would be "truly shocked" by an audit of how few net new meetings the average partner at other firms takes in a week [1]. The reason more small funds do not scale this way is simply that everyone doing this job is ambitious and wants scale, and the AUM path is easier [1].

The numbers he attaches: about 5,000 to 6,000 qualified or referred opportunities a year, converting into roughly 70 to 80 investments [1]. He is unbothered by the index accusation, pointing to the historical YC arithmetic of batches around a hundred companies producing two or three billion-dollar outcomes, one to three percent hit rates from an elite pool [1]. BoxGroup has to sift far more than a hundred deals to reach the same place, but the logic is the same: take more shots on goal, and the odds of landing a thirty-billion-dollar company rather than a three-billion-dollar one go up [1]. The bar stays high throughout, framed internally as always asking whether this will be the most important company started over the next decade [1].

On adverse selection as the thing to solve

Pressed on whether the model works because of volume or because of selection quality, he does not hedge: "it's both but it's primarily not getting adverse selection" [1]. That is the whole game. His implicit premise is that quality at seed is somewhat knowable, and that the harder problem is access rather than judgment. Proprietary deals barely exist anymore, and where they do they last "like literally a blink of an eye like not even a day anymore" [1]. The consequence is that being able to invest alongside the best leads, rather than having to beat them, is worth more than the extra points of ownership.

On check size and making it easy to say yes

Check sizes have moved from 50 to 75K in the early days to 750K to a million now, tracking the inflation of round sizes as a fully loaded seed went from around $2 million a decade ago to $5 million [1]. The design principle is founder optionality: "we want to make it as easy as possible for a founder to say yes to us" [1]. If BoxGroup gets there first and there is no lead, they can commit and be the first yes, then help find the lead. If a lead is already in place, they can round out the syndicate quickly. What they will not do is write the $3 or $4 million lead check [1].

On getting there before there is a deck

The firm's internal shorthand is blunt: "if there's a deck it's too late" [1]. A deck means a team has coalesced and real work has gone into presenting it, which means the party started without you. So the outbound effort aims at a stage earlier than a company exists: "before there's a company there's a human being how do we get that human being often times before they've even left the company and that's what earns us the right to be that first check that first. Yes" [1]. Waiting for someone to announce themselves does not work, since nobody posts on Twitter that they are starting something, and by the time they do there are usually already two rounds in [1].

The tactics are unglamorous and he says so. Events across the country, intern events, campus events, engineering leader gatherings, none of it unique alpha, all of it dependent on curation. "It is very easy to do a bunch of events that return nothing because you didn't curate great people and do all the hard leg work" [1]. They built software to scrape LinkedIn for job title changes [1]. Inbound he describes as icing, delicious but insufficient: "you can't just eat icing right? You need the actual body of the cake and that is outbound" [1]. Underneath the tactics is a mentality point he considers more important: "there is no substitute for your time," and most check writers refuse to spend it on someone who has not left their job and may never start a company [1]. That means accepting a high volume of meetings that go nowhere, an hour with a director of engineering who turns out to simply love being a director of engineering [1].

On archetypes and why incoherence is a good sign

Rosen resists blanket judgments about founder quality because the assessment only makes sense against what someone is actually building. Internally the team has vocabulary for the patterns: the amazing big-company person who is not a founder, the extremely technical founder who is not commercial, and so on, and each archetype has to be lined up with the work in question [1]. His broader advice on archetypes is to identify which one you are, whether hyper-competitor, philosopher or specialist, find where your advantage sits, and "just focus all your energy and effort on being amazing there" [1].

The corollary at the earliest stage is that you have to stop looking for polish. "When you go super early and nothing is formed you cannot not look for perfection. You cannot look for cohesion even in ideas" [1]. He describes passing on founders who were compelling as people but were juggling ten different ideas, and treats those passes as a mistake since corrected: scattered thinking "that's a feature, not a bug. That means we're getting to them before they've actually figured out exactly what they want to work on. And ideas are not precious. You can work on anything" [1].

On AI pulling the founder age curve down

New technology always favors the young, because "it's always the younger generation that figures out the most native ways to use that technology," and with AI the rules of company building are being rewritten fast enough that the advantages of experience decline [1]. He points to codegen as an area where something new lands seemingly every day [1]. The practical consequence for sourcing is a sharp shift toward students and very young founders, and he estimates the number of 18 and 19 year olds BoxGroup funds now is at least one and possibly two orders of magnitude higher than a decade ago [1].

On why they do not hold partner meetings

BoxGroup does not run them, and Rosen thinks "partner meetings make no sense" [1]. The stated purpose is to get several decision makers' opinions on a founder, but the format forces the founder to regurgitate the same pitch for a fresh audience, few of them enjoy it, and the exercise is performative [1].

Takeaways

• The binary rule of neutrality: once a collaborative firm leads even two or three rounds out of 80, every counterparty assumes it is withholding its best deals, so the trade-off cannot be made case by case [1].
• Rosen deliberately trades ownership percentage for access, on the view that being in the right companies is harder than owning more of them [1].
• His "venture humility score" says the achievable improvement is in seeing, not picking, since he considers it a fool's errand to expect to become 2x or 10x better at picking [1].
• The funnel runs roughly 5,000 to 6,000 qualified or referred opportunities a year into 70 to 80 investments [1].
• Check sizes rose from 50 to 75K to 750K to a million purely to track round inflation, with the goal of being able to say yes whether or not a lead exists, and never writing a $3 to $4 million lead check [1].
• "If there's a deck it's too late": the target is the person before they leave their job, which requires accepting a lot of meetings that lead nowhere [1].
• A founder pitching ten unrelated ideas is evidence you arrived early, not evidence they are unfocused, because ideas are not precious [1].
• AI has pushed the number of 18 and 19 year olds BoxGroup backs up by one to two orders of magnitude, because new technology rewrites the rules and erodes the advantage of experience [1].