YC W13 Will Be Smaller
We just finished interviews for the winter 2013 YC cycle. We funded a lot fewer startups this time than last. We funded 84 in summer 2012. We don't know yet exactly how many we'll have in winter 2013 because we usually add a handful after interviews and sometimes a few groups fall apart before we can fund them, but it looks like we'll have less than 50.
We meant to do this, but we didn't want to say anything about it beforehand, because our decentralized interview process means we couldn't predict how many startups we'd accept.
The reason we accepted fewer applications was that in summer 2012 we grew too fast. We had 66 companies in winter 2012, and that was fine, but for some reason more things than usual broke when we jumped from 66 to 84.
Things are always breaking at YC, because our strategy is to find bottlenecks by hitting them. That may sound irresponsible, but in practice it's the way most complex systems get optimized.
We're still a bit mystified about what happened. Why was 66 ok and 84 not? Is there some kind of hard limit somewhere between those two numbers? Or will we be able to morph YC to get past that bottleneck as we always have in the past?
My guess is the latter. We'll see. But we decided to buy ourselves more time to morph YC by decreasing the batch size.
The way we did it was to make a concerted effort to teach ourselves about the predictors of failure. In the past we spent all our time thinking about the predictors of success. And that is what you want to spend most of your time on; it's much more important to fund the startups that succeed than to avoid funding startups that fail. But we knew there were predictors of failure, and that if we made a conscious effort to catalogue them and to look for them during interviews, we could probably decrease the batch size by quite a lot without missing any Dropboxes or Airbnbs.
So that's what we did. For the last several months we've focused on identifying predictors of failure, both by looking for qualities that were missing in the top startups we've funded, and by looking at things the failed startups had in common. We even studied the times of day we accepted startups that failed (it turned out that, like judges, we were more tolerant after lunch).
During the most recent round of interviews, we applied all we'd learned. We were successful at being more selective. We won't know how well we did at the much more important task of not throwing out babies with the bathwater for a year or so.
We don't plan to stay at 50, or whatever exact number of startups
this batch ends up having. We've never had a predetermined batch
size; that's just the number we ended up with when we tightened our
filters as much as we could. Which means if we maintain the same
tightness and the number of applications continues to increase each
year, the batch size will increase with it. What will we do as we
approach the mysterious line between 66 and 84? We've been tweaking
our model, like we always do, and we hope some of the tweaks will
push the limit back. Or maybe we'll continue to get better at
identifying predictors of failure, so we'll be able to grow slower
than applications without missing any stars. Maybe as t approaches
infinity, we'll approach a state where each batch consists of 50
Airbnbs. That would be nice. But frankly we have no idea what's
going to happen. We're figuring this out on the fly, like we always