the model behind thesis-aligned scoring.
Cold decks land in your inbox at all hours. VCOS scores them against your thesis the moment they arrive. Here is how the model decides what surfaces and what stays in the queue.
A partner at a Series A fund opens her inbox on Monday morning and sees forty-seven new pitch decks. By Tuesday afternoon there are eighty-three. None have been read. None have been replied to. By Friday the queue has hit a hundred and twelve, and she is on a flight to a portfolio board meeting. She does not open the inbox until the following Tuesday, by which point the queue is two hundred and forty.
This is the story of every fund we have ever talked to.
the inbox is a coin flip.
Inbound deal flow looks like a steady stream. It is not. It is a binomial filter where the partner reads the first ten decks of the morning, makes ten quick judgements, and then defaults to no on everything that arrives after. The decks at the bottom of the queue are not worse than the decks at the top. They are simply newer, or older, or quieter, or sent at the wrong hour. That is unfair to the founders who sent them. It is also unfair to the fund, because the pitch you would have backed at the bottom of the queue does not get backed.
So we built Flow.
the thesis as a vector.
Flow's job is to make the bottom of the queue and the top of the queue equally legible. Every fund we work with starts by writing their thesis. Not the marketing version. The working version. The one the partners argue about on Sundays. Sectors, stages, geographies, founder profiles, deal sizes, and the hard nos. We translate that into a vector the model can read. Every cold deck that arrives is then translated into the same kind of vector. The model compares, scores, and routes.
The model does not say which deals are good. The model says which deals match this fund.
what surfaces.
The deals that surface are the ones that fit the thesis vector closely on most dimensions and have at least one dimension that is unusually strong. That last condition matters. A deal that is mediocre across every axis scores in the middle. A deal that is exceptional on one axis and decent on the rest scores higher. The model is biased toward conviction, not consensus.
The first fund that ran Flow on three months of inbox backlog found a Series A infrastructure company in week one. The deck had been sent in week ten. It had never been opened. The partner who passed on it the first time around led the round at the second time of asking.
score is a hint, not a verdict.
what stays in the queue.
The decks that stay in the queue are not deleted. They are not buried. They are filed against the thesis with a short mono explanation of which axes they did not match. A consumer-seed deck arriving at an enterprise-A fund gets routed with the note that says stage and sector are off but two warm intros sit in the CRM and the founder may be relevant for a partner connection. The partner can ignore that or act on it. The model has done the bookkeeping either way.
a hint, not a verdict.
The first thing every partner asks us is whether they should trust the score. They should not. The score is a hint. The model is good enough that the surfaced decks earn a five-minute read. It is not good enough to make a yes-no call. Conviction is still the partner's job. We are just trying to make sure the right decks reach her at the right time, and that the deck at position two hundred and forty has the same shot as the deck at position one.
Most funds tell us the most uncomfortable part of running Flow is realizing how many good companies they had been missing. The queue does not lie. Once the bottom of the queue has the same surface area as the top of the queue, the question becomes harder, not easier. That is the right kind of harder.
