What I look for in a founder

The corridor outside the main room in Tokyo is always too narrow for the number of people standing in it. Paper cups, nobody sitting, the next session about to start. A founder catches me against the wall and she has four minutes, and she knows it.

She does not open with the product. She says her cathode coating fails at a temperature she cannot explain, that she has three theories about why, and that she will know which one is right by the end of the month. Then she asks me something. Then the session starts and she is gone.

That was the whole meeting. I remember it better than most hours I have spent in a room.

I am an impact angel investor. I write small cheques out of my own money into founders using AI to move faster on hard problems, mostly climate. I am early at this and my record is short. So what follows is what I pay attention to, not a method that has been proven on anybody.

How long is one turn

The first thing I want is a number. How long from asking a question to knowing the answer.

Founders find the question slightly annoying, and the annoyance is informative. Some answer in a sentence. Eleven days last time, four days now, and here is what we cut. Others describe a process. The people describing a process are usually describing a company that has not learned anything in a while.

Where AI sits inside that number is the part I care about, and it is not the part the deck shows me. Almost every deck has AI in it now. Usually it is at the end, as a feature, a chat box bolted onto the thing they already built. That is fine and it changes nothing about the company.

The version I want is in the middle of the loop, at the place where the waiting happens. Two thousand candidate structures simulated overnight instead of six chosen by hand and run for a month. The whole literature of a field read instead of the forty papers a person can hold in their head. A sensor stream that used to need two analysts, answered on Tuesday.

None of that is visible on a slide. It shows up as a smaller number, and the number is the thing I am buying.

Whether they can be told something

This one I cannot test with a question. I have tried, and everyone says yes.

So I watch for something small instead. Somewhere in an hour, does the person stop halfway through a sentence and say, no, wait, that is not right. Not about anything load-bearing. About a figure they had wrong, or a competitor they had dismissed too fast, or a word they had been using loosely for a year.

A founder who can do that in front of a stranger can do it with their own data at eleven on a Sunday night, when nobody is watching and the data says the last nine months went the wrong way. A founder who cannot will defend a dead branch of the roadmap for a year, politely, while the money runs out.

I have met brilliant people who cannot do it. The cheque does nothing for them and neither does anything I say. That is the part of this where I am least useful, and I know it.

Saying no

There is a kind of founder who has never turned anything down.

Their roadmap is a list of other people’s requests. Every line on it came out of a meeting where somebody important asked whether it would also do X, and the true answer was no, and what came out of the founder’s mouth was yes, by the third quarter. Three years of that and the product is a corridor of half-built rooms.

Once in a while a founder tells me instead about a pilot they refused. Real money, a real logo, a customer whose name would have made the next round easier. It would have taken four engineers sideways for six months to build something only that one company wanted. They said no and spent the six months on the furnace.

The good ones are never smug about it. They tell you it was horrible and that they went back and forth for a fortnight and still are not certain.

Saying no to an investor is the same muscle. I would rather be turned down by someone who has a reason than taken on by someone who takes everything.

The paragraph that starts with the broken thing

I ask people to send me a paragraph before a deck. What I am reading for is whether the broken thing is in it.

Most paragraphs are the good news. The pilot, the grant, the professor who said the technology was a decade ahead. Then you meet them, and forty minutes in something else comes out, and it turns out that is the actual company. The co-founder who left in March. The customer who has not paid. The number that has not moved all year.

If that had been the first sentence, I would have read the rest faster.

I do not think founders hide it to deceive anyone. They hide it because they have spent years being told to stay positive in rooms where honesty bought them nothing. Fair enough. I have done the same thing, and I still do it when I am tired.

The last two of these are really one thing wearing different clothes. Turning down a customer and naming your own broken part both need you to be, for a minute, less impressive than you could have been. Nearly everything I actually want from a founder sits underneath that.

What I get wrong

The confident calls. It is almost always the confident ones.

I am also pulled toward people who look like they had to fight to get into the room, because that is my own story and it bends my judgement. I carry a Korean surname. I grew up a weird kid in Kawasaki and I have been the wrong sort of person in a lot of meetings. Noticing a bias is not the same as correcting one, and I have not corrected it.

And I fail my own tests. My loop is slow. I answer emails from four weeks ago. I have said yes to things I should have refused, mostly because saying no to a person’s face is hard and I am bad at it.

When I lost a game as a kid, my friend Yuma used to tell me not to waste time on what I could have done and to work on what I could do now. He was better at that than anyone I have met since. I think about it most when I am holding someone’s paragraph and deciding whether to write back.