Dave Snowden cites this great story from a commenter to his earlier post on the perils of six sigma.
When I worked at IBM we were asked (in 1990) to 6Sigma our CICS development team. The gurus told us that the next release of CICS could only have 6 bugs (or APARs as we called them). This was ridiculous but luckily a colleague ran a report and showed that IBM program products had extremely strong positive correlation of profitability with APAR rate. That is, the products with the most APARs were the most profitable. This is because great products, like CICS, get used for lots of things we didn’t think of and for which we didn’t test. Mediocre products only get used for what the tests cover. Bad products don’t get used at all and so generate almost no bugs.