3 ms·

A Machine To Pick Startup Winners

24 points by raheemm 14y ago

ovi256 14y ago
And that's a pretty clear "submarine" [1] article, without a doubt pitched by PR to the journalist. Don't get me wrong, it was a good and informative article, it's just that I'm training myself to recognize these. [1] http://www.paulgraham.com/submarine.html http://www.paulgraham.com/submarine.html
- akldfgj 14y ago
  Ah, yeah, it's Technology Review. That's what they do. It's not a rabble-rousing publication.
stcredzero 14y ago
The problem with analytics I've experienced, is that they often aren't actually good at picking out the something they're targeting. They're actually good at picking out something like what they're targeting. When I use Pandora, I just get hollow dopplegangers of bands I like, not bands I actually like. That said, and no doubt this will be controversial, I think there's great potential for Moneyball style statistical analysis of programmers. (Though there are significant organizational and cultural issues involved.) The reason why stats work in baseball, is that there's ample empirical data: baseball players are up to bat, or have a chance to catch a ball again and again. Like baseball players, there are certain actions that programmers do again and again -- namely commits going to production. If one could get accurate stats on a large enough population, with automated metrics for function points and bugs, it should be possible to produce meaningful stats on programmers working with a given technology.
- hooande 14y ago
  This sounds interesting, but I think the difference in the two situations is a repeatable experimental environment. In baseball every at bat is the same...a batter, a pitcher and 8 defenders. This lends itself well to statistical analysis. In programming, every commit is different and under wildly varying circumstances. I think the same logic might apply to the OP. Looking at a startup investment through the lens of a credit score makes some sense, but it seems more likely to avoid predictable losses than it is to identify the next facebook or google. As an aside, there's something I've always wondered about companies who predict the success of other companies...what score did they get from their own algorithm?
  - stcredzero 14y ago
    This sounds interesting, but I think the difference in the two situations is a repeatable experimental environment. In baseball every at bat is the same...a batter, a pitcher and 8 defenders. This lends itself well to statistical analysis. In programming, every commit is different and under wildly varying circumstances. You could argue this both ways. The situation at bat varies quite a bit. You can have different pitchers, different lighting, different wind speed and direction, different umpire, different situation on-base, etc... However, this all comes out in the wash because there are so many iterations. Programmers commit a lot. If you can get reliable data including bug rates for a given programming environment, I think you'd have good data.
Yaa101 14y ago
Computer says no...
jakeonthemove 14y ago
Hmm, since this is based on analyzing existing data, the system will most likely be useless for companies that want to do something new or something niche-specific. Not a good thing for an investment company, unless they always want to play it safe.
asanwal 14y ago
Interesting challenge, but it seems they look for a lead investor as the first signal of quality for the investment. As a result, it's questionable how much additional value their algorithm is uncovering/adding to their decision as a follow-on investor. Said another way, if Sequoia or Kleiner or Union Square are investing and Correlation then gets to see the deal, I wonder how many deals their algorithm would pass on. If this was truly algorithmic startup investing to be the lead investor, it'd be much more interesting. But this method of using social proof as the first filter seems like business as usual. That said, I imagine this is a great approach from a marketing perspective when raising the fund and reaching out to investors (LPs) as it clearly differentiates the fund from a thesis/approach perspective.

24 points by raheemm 14y ago

And that's a pretty clear "submarine" [1] article, without a doubt pitched by PR to the journalist. Don't get me wrong, it was a good and informative article, it's just that I'm training myself to recognize these. [1] http://www.paulgraham.com/submarine.html http://www.paulgraham.com/submarine.html

akldfgj 14y ago
Ah, yeah, it's Technology Review. That's what they do. It's not a rabble-rousing publication.

stcredzero 14y ago

The problem with analytics I've experienced, is that they often aren't actually good at picking out the something they're targeting. They're actually good at picking out something like what they're targeting. When I use Pandora, I just get hollow dopplegangers of bands I like, not bands I actually like. That said, and no doubt this will be controversial, I think there's great potential for Moneyball style statistical analysis of programmers. (Though there are significant organizational and cultural issues involved.) The reason why stats work in baseball, is that there's ample empirical data: baseball players are up to bat, or have a chance to catch a ball again and again. Like baseball players, there are certain actions that programmers do again and again -- namely commits going to production. If one could get accurate stats on a large enough population, with automated metrics for function points and bugs, it should be possible to produce meaningful stats on programmers working with a given technology.

hooande 14y ago
This sounds interesting, but I think the difference in the two situations is a repeatable experimental environment. In baseball every at bat is the same...a batter, a pitcher and 8 defenders. This lends itself well to statistical analysis. In programming, every commit is different and under wildly varying circumstances. I think the same logic might apply to the OP. Looking at a startup investment through the lens of a credit score makes some sense, but it seems more likely to avoid predictable losses than it is to identify the next facebook or google. As an aside, there's something I've always wondered about companies who predict the success of other companies...what score did they get from their own algorithm?
- stcredzero 14y ago
  This sounds interesting, but I think the difference in the two situations is a repeatable experimental environment. In baseball every at bat is the same...a batter, a pitcher and 8 defenders. This lends itself well to statistical analysis. In programming, every commit is different and under wildly varying circumstances. You could argue this both ways. The situation at bat varies quite a bit. You can have different pitchers, different lighting, different wind speed and direction, different umpire, different situation on-base, etc... However, this all comes out in the wash because there are so many iterations. Programmers commit a lot. If you can get reliable data including bug rates for a given programming environment, I think you'd have good data.

Yaa101 14y ago

Computer says no...

jakeonthemove 14y ago

Hmm, since this is based on analyzing existing data, the system will most likely be useless for companies that want to do something new or something niche-specific. Not a good thing for an investment company, unless they always want to play it safe.

asanwal 14y ago

Interesting challenge, but it seems they look for a lead investor as the first signal of quality for the investment. As a result, it's questionable how much additional value their algorithm is uncovering/adding to their decision as a follow-on investor. Said another way, if Sequoia or Kleiner or Union Square are investing and Correlation then gets to see the deal, I wonder how many deals their algorithm would pass on. If this was truly algorithmic startup investing to be the lead investor, it'd be much more interesting. But this method of using social proof as the first filter seems like business as usual. That said, I imagine this is a great approach from a marketing perspective when raising the fund and reaching out to investors (LPs) as it clearly differentiates the fund from a thesis/approach perspective.