the problem is in the connection-less nature of http protocol, which only connects to the server to retrieve a web page, but there is no option to tell, when the user closes the page or the browser - only if he executes another action which might indicate, that he is now on a different page (which might be in a different browser window though).

Due to the stateless nature of the Internet, any results you get will be completely unreliable. You will find extremely short periods where someone logs in gets the info they are looking for and either logs out or goes somewhere else. On the other hand, you will find extremely long periods where there is no user activity until some sort of timeout period (say 20 minutes). The results represent that the user was either reading your web page or someting else in the bathroom.

The bottom line is that you will never know if the user was using your website, mulitasking or sleeping.

A much better metric is to use a unique coupon tracking system. Regardless of the time spent on your site, when the coupon is used, you can measure the results of your effort and your return.