YouNoodle Is A Terrible Idea

If your co-founder is a child hood friend chances are you know them very well. If you just met them at the WWDC a couple months back chances are they might be an ass.

-- I'd like internal details of my startup to NEVER be released without my knowledge (including ownership breakdowns, investment amounts, number of employees, strategy/expected launch date, revenue/traffic numbers, etc.), under any circumstances - even anonymously. (If you released info on a "video startup in Minnesota", but left out the name, that would be potentially be enough to be traced back to http://zencoder.tv.)

-- If you want to use my information in the aggregate, that doesn't bother me as much, and I understand that you may use information about my company as a part of your predictions.

-- Obviously, you should take data security very seriously and go above and beyond 99% of companies in this area.

-- I should be able to cancel my account with a click of a button, and this should remove all of my private information from your system.

These would make me comfortable with YouNoodle, and if I felt certain that you'd follow these standards, I'd probably try it out.

I'm guessing it's unlikely you guys will do that. But if someone else does I'd be happy to try it out :-)

> "Before the 1950s, lending depended on the wisdom and judgment of loan officers. Then, a company called Fair Isaac developed a way to use four simple variables to develop a credit score that reliably predicted the risk of lending to an individual."

That is a very misleading analogy. YouNoodle is not the present-day analog of Fair Issac. This goes back to the idea that while it's very difficult to predict random events individually, you can predict them well in aggreggate. Fair Issac isn't really betting on customer X defaulting on a loan -- they're really betting on a thousand customers like X. Same with insurance premiums. A model won't be a great predictor of one person, but if you have hundreds of thousands of customers, you can have a much more accurate estimate. This is the law of large numbers. For a simpler example, you wouldn't want to bet on the outcome of a single coin toss, but it's a pretty safe bet that if you toss a (fair) coin a million times, the heads percentage will be very close to 50%.

Fair Issac played the game by lending to lots of customers. YouNoodle is trying to play this game for each individual case. They're bold enough to mathematically estimate the value of one individual startup, and say that they have some confidence in it. It's the analog of making a loan to 1 person, insuring 1 person, or betting on a single coin flip. Probability tells you this is very risky. VC don't make thousands of bets like insurers or banks do. For the bank, my individual outcome doesn't really matter to the big picture. VCs on the other hand invest in a very small number of startups - each startup matters.

Exactly my point. Given 100 startups with that profile, you can say with some confidence that the average outcome will be 500k. But you're ignoring the distribution! It's very risky to make the bet that one individual startup will be worth 500k.

I'm not saying you should make better predictions. I'm saying that it's a very risky game to make mathematical predictions on individual valuations at all.

Most machine-learned models aren't better than humans, they're just much, much faster than humans. It's a bold bet you're making if you're trying to be better than humans at predicting valuations. Models don't allow for intangibles like "killer sales ability" that may not be reflected in your degree or past job, but that a VC can pick up on 5 minutes after meeting you.

See my reply to Sam's post here: http://news.ycombinator.com/item?id=272015

Edit: Someone put in the relevant numbers for WebVan or Pets.com. I think those two failed within ~3 years.