“We get really great analysis and good market mapping and good work. We actually find that the recommendations from AI are, no matter what we do, really disappointingly low conviction. And so we have a scoring system that we sort of play with internally where we like to score deals on a one through nine, no fives.”
This Week in Startups · Startups & Venture · November 2026
Context
A co-panelist asked how far VCs are willing to go against an AI's opinion when they feed it a pitch. Ben says his firm's AI scores come back in the middle almost every time, which is useless for a business built on conviction.
Transcript
…to be a company that is going to you're comfortable trading on your merits and like what you really are versus some like weird manufactured, like, you know, don't look in that room story. And so like, I think this is a good thing. I don't think it's like changing the world, but I'm all for it.
…presentation materials into our AI and respecting to some extent its opinion, I'm curious for you guys, like how far are you willing to go against the AI? Are you at a point where you've fed stuff in and it said, this is a terrible investment, never do it. Are you ever going to supersede it?
So, two things. One is we've all heard that like that adage about the contested deals are the great deals. I think Sequoia has also talked a lot about this in the past. If that's the case, then we should be concerned about this vintage for late stage companies because, you know, the companies that are getting access to the real late stage capital are all these like highly, highly consensus deals. Now, I can't speak to what actually happens in the partner meetings of these mega funds, but right now we're operating in a market where there's this belief that every that like things are obvious. And so that's just like an interesting little tidbit around the edges of that. As far as the advice AI gives, we get really great analysis and good market mapping and good work. We actually find that the recommendations from AI are no matter what we do, really disappointingly low conviction. And so we have a scoring system that we sort of play with internally where we like to score deals on a one through nine, no fives. And
one through nine, one through five. No fives because cowards pick fives. I do one through 10, no eights because cowards pick eight.
Normally it's a world of no sevens, but we could talk about this separately, Jason. But I do think that there is we so we go and we put it in and we get scores between three, four, six, and seven almost 100% of the time, which is just like a pretty weak willed, you know, it's like a yes, but with…
hint. Don't give us the whole algorithm. I know it's proprietary. No, no, no. Do you have some variables? Because I'll give up a couple of variables for us.
More from This Week in Startups
