Basically what I developed was a local engine that could use characteristics like browser history, programs installed, computer usage stats, etc. to target ads. It had a scripting language so that instructions to implement new targeting criteria to scan the computer for could be sent down without changes to the underlying software. For example, if Netscape wanted advertisers to be able to target Microsoft Office users, we would include instructions for the software to search the computer for Office and watch for future installs of it. No data ever actually left the user’s computer, as is the case with your model - the fact that they matched a given characteristic was simply stored in a local database.

Websites that wanted to place targeted ads on their site could then do so with some special markup indicating exactly what criteria they wanted to use. They could specify an almost limitless number of different creatives and corresponding targeting criteria right in their HTML. My thought was that they could pay Netscape for the privilege of enabling highly targeted ads.

The privacy downfall here, at least in Netscape’s opinion, was that sites could extrapolate whether or not someone matched the criteria by watching to see which banner their browser downloaded. I countered this by having an option for the software to download 5 or 10 other banners in random order in addition to the one that actually met the targeting criteria in the background, but that wasn’t good enough to overcome the perception of privacy issues in their eyes.