Wow. Some hacker was a Netflix subscriber, and started noticing that he was having longer-than-usual waits for the hit movies on his list. Yet when he opened up a new account for his wife, almost every movie she asked for, she got — instantly. What’s up?

He did a rather fascinating study: He took five accounts of various ages (pre-existing ones his friends used), then wrote some Perl scripts that would automatically add and subtract movies from their Netflix queues.

The result? Active accounts — ones that rented a lot of films — quickly found that their service was degraded: They’d start getting long wait times for the films they wanted. But new customers, or people who only rarely rent movies, would get what they wanted instantly. Netflix’s principles thus appears to be:

1) Giving trial and first month customers a great experience out of the block. Would you go past the 10 day trial period of you had to wait a week for a title Blockbuster has on the shelf now?

2) Keeping profitable customers happy. Wired magazine’s Dec 2002 issue reported that Netflix loses money due to postage costs if a customer rents more than 5 movies a month. Keeping your customers who produce the best profit margins happy is a no brainer. The customers who take a lonnnnggg time to watch their movies and then stick their discs back in the mail are incredibly important to Netflix.

3) Letting unprofitable customers drop off. I think this may be more of an unintended but acceptable consequence rather than the goal. Certainly many heavy users have in the past and will in the future quit Netflix due to frustration over the availability issue. In the 2003 Q1 conference call Netflix stated that a “there is a very, very, very small percentage of our subscriber base that is uneconomic.” That could very well be those customers who are yielding $2/disc costs or less. Subtract postage, handling, mailer costs, infrastructure, licensing fees, etc. and that heavy user leaves little if any room for profit.

This is really amazing stuff. Go check out his full findings here, including really cool charts of his data!

(Thanks to Slashdot for finding this one!)