This is one of the most common reactions, and the answer is, there's no real difference: lossy compression is the same thing as lossless. You simply store some extra bits to correct the errors or lost data (typically in an arithmetic encoding framework where you can plug in arbitrary predictors or combinations of predictors), and now it's back to the lossless setting. So all real compression algorithms are effectively 'lossy' already, and making the distinction gains you nothing. You instead spend your time thinking about how to predict better at all.