Fix it now: The effort to fix a defect as soon as your developer types the wrong code is pressing Ctrl-Z (UNDO!). Cost is essentially ZERO. And if you're pair programming, you're very likely to catch the defect at this point.
We'll fix it next week: The cost of fixing a defect one week after it happens is a fix on the incorrect code, plus the time refactoring one week of code developed on top of the bad code. Probably not too much, we're talking about a few hours max.
We'll fix it at the end of our iteration: The cost of fixing the defect after a two-week iteration is probably about twice as much as after one week. Cost is up to one day.
Don't worry, we'll get to it later, we don't have enough QA budget for it now: Here we are 6 months later at the release date. The cost of fixing one error in the code is now exponential if you have to refactor 6 months of code based on the initial defect. Cost: Potentially HUGE!
Now, imagine you're one of the bean-counting managers when it's time to release the buggy software. Here's your argument: "What! We have a ton of fixes to do? Marketing already ran the ads announcing the release date! We have to beat Foo, Inc. to market on this or we're dead in the water! Plus, it's going to cost us way too much to fix everything now! We have no choice, release it now, we'll fix it later".
Fast forward to your customers using your new buggy software. They're disappointed. They start ripping your product (and maybe even you) on their blogs. Word gets out, your stuff sucks! You do your best to put out patches and service packs. It doesn't matter, word of mouth has spread the bad news, your product still sucks!
Now ask yourself, was it worth it. Releasing low quality software based on the argument above makes no sense. Your company probably lost market share to Foo, Inc. even though you beat them to release because of the bad press. And you still had to spend the money to fix the defects through patches and service packs. By now, you're probably way in the hole, much worse than if you had taken the time and money to fix the defect when it happened or very shortly thereafter. I've already made the argument for the value of pair programming and continuous QA/testing, but it's worth repeating. Spend the money, spend the time, use your resources and build defect free software NOW! The longer you wait, the more it costs. And you may end up releasing buggy software intentionally, and that's the worst thing you can do. Maybe David Rice, author of Geekonomics: The Real Cost of Insecure So