Eleven coffees with people who get it, and none who can build the half you cannot.
It starts well every time. You meet someone sharp, you finish each other’s sentences about the market, you both think the incumbent is asleep. By the third coffee it feels like the search is over. Then you notice that you have both spent an hour describing the product, and neither of you has said who is going to build it.
Or you get the other version, the one that arrives in month nine. The co-founder who seemed perfect turns out to have meant evenings and weekends where you meant everything, and to have assumed a split you would never have agreed to if either of you had said a number out loud.
There are two questions almost nobody asks on day one. How many hours will you really work? What split do you need? They feel too blunt for a first meeting, so they wait, and they keep waiting, until they are too expensive for any meeting.
Meanwhile every product that could help is pulling in the wrong direction. Same school, same interests, same vocabulary: the more you overlap with someone, the higher they rank. For almost any other relationship that is a sensible default.
Every matching product rewards similarity. For a founding team, similarity is the failure mode.
LinkedIn is very good at what it is. It will help you find people like you, and it lets people who want a job say so in real detail. But co-founding is not a job search. There is no field for “I need someone who is not me”, and no way to say “full-time from January, and I need forty percent”. It was never designed for this, and that is a fair thing to say about a product rather than a complaint.
In Deals On, co-founder fit is the heaviest single signal in the engine. It weighs more than how well your stated goals line up. It is also the only signal that punishes overlap instead of rewarding it: the same machinery that rewards a shared function when somebody is hiring is deliberately inverted here.
And it refuses to be charmed.
Two people who each need 60% score zero. No split satisfies both, and rapport does not change arithmetic.
Here is one search, as it actually runs.
On day zero he writes four constraints most people never write down until the argument: I am a product person, I need engineering, full-time from January, I need 40 to 50 percent, and I will only build with someone I can sit next to.
That night, everyone who is also a product person scores badly, not neutrally. Everyone on nights and weekends while he is full-time scores badly. Anyone whose minimum split, added to his, comes to more than a hundred percent scores zero, because no split satisfies both of them.
In the morning there are two names. Both engineers, both full-time, both in the region he named. One of them has written, in their own words, what they need the other person to bring, and it is what he brings.
Three weeks later they talk. The hard questions had answers before the first coffee, because the system refused to match on anything else.
The filter did the argument for them, on day one instead of month nine.
None of this makes the decision for you. Two names is not a shortlist you are obliged to like, and a score is not a marriage. What it does is stop you spending nine months discovering, slowly and expensively, something both of you could have said in a sentence at the start.
Say your number before the coffee goes cold. Then go and find the person who can build the half you cannot.
