Why no one can honestly tell you what your fund will return

The question behind the question

A model that told a General Partner what their fund would return would be worth considerably more than one that tells them whether their fund is internally consistent. The reason the Colibrí Architecture model does the second thing is not modesty. It is that the first thing cannot be done honestly from the inputs available, and a model that pretends otherwise does damage in proportion to how much it is trusted.

What a configuration can and cannot support

A fund's configuration is a complete, precise, and knowable object. Fund size, portfolio count, ownership target, stage, scope, reserve share, team size, and the rest are all declared, all available on day one, and all internally checkable against each other. Whether they fit together is a question the configuration itself can answer.

Performance is not in that object. It depends on which companies the fund happens to meet, the market it deploys into, the co-investors who show up, the hires the founders make, and a decade of accumulated contingency. None of that is in a configuration, and no amount of regression on historical funds puts it there. What historical funds can tell you is how configurations like this one have been distributed, which is a statement about a population rather than a forecast about a fund.

So the model reads what is legible and declines the rest. A configuration the model reads as coherent can still return poorly. One that surfaces tensions can still return well. Both statements are true, and a model that could not say them would be overclaiming.

Why this is the more useful reading anyway

There is a practical argument as well as an honest one. A prediction, even a good one, is difficult to act on. Being told a fund is likely to land below its target does not say which decision to revisit.

A configuration reading is actionable by construction, because every input is something the firm chose and can change. A tension between ownership target and stage names two decisions and asks which one moves. That is a conversation a partnership can actually have on a Tuesday, and it is available while the fund is still being designed, which is the only point at which the answer is cheap.

A question this raises: why don't VCs follow Warren Buffett's 20-slot rule

Buffett's rule, offered to students, is that if you had a card with twenty punches on it representing every investment you could make in a lifetime, you would think far harder about each one and end up concentrated in your best ideas. The question that follows naturally is why venture funds, which face the same logic, build portfolios of thirty companies instead.

The answer, per SaaStr, is that they largely do follow it, and the appearance to the contrary is a counting error. A firm makes many investments over its life, and a fund typically holds twenty to thirty companies, but the unit that matters is the partner: each partner usually does something like four to six investments per fund. Measured where the judgment actually happens, venture is already concentrated, and most partners will make well under a hundred real investment decisions in a career.

The more interesting question is the one underneath: does concentrated-portfolio theory from public markets transfer to venture at all. There are two structural reasons to be careful.

The first is that Buffett can size a position after forming conviction, and can add to it for years at prices he chooses. A venture investor sizes the position at entry, before most of the evidence exists, and can only add on terms someone else sets. The concentration is front-loaded into the least informed moment.

The second is the shape of the return distribution. Public equities are bounded on the downside at total loss and rarely produce hundred-fold outcomes; venture routinely does both, and the entire return of a fund can come from one position. That makes breadth do something in venture that it does not do in public markets: it is not only risk reduction, it is exposure to a tail that cannot be identified in advance.

Both of which is to say the rule does transfer, but it transfers to the partner rather than to the portfolio, and it is already being followed more closely than the headline count suggests.

What the model claims

The claim is narrow and it is worth stating exactly. The architecture of a venture firm can be evaluated against a principled framework. The framework surfaces tensions and configurations worth examining. The outputs are consistent across firms at the same lifecycle stage. And they are diagnostic: they tell a General Partner something true about how the firm is built that they can act on.

The model does not predict returns, does not predict success or failure, does not rate firms, and does not assert that any threshold separates the funds that work from the funds that do not. It is an instrument that supports judgment rather than a substitute for it, and the value of the reading depends on a General Partner interpreting it against context the model cannot see.

Sources

• Why Don't VCs Follow Warren Buffett's 20 Slot Rule?SaaStrAnswers that they largely do, once concentration is counted per partner rather than per firm.