Tracking the traffic an ad box gets seems like a straight-forward problem, right? Mark whenever you get a display, and call it a day. If you see the same IP address (roughly equivalent to the same person) then you know they've hit the site twice.
Turns out it's not that easy, and that's one of the reasons you're never likely to get matching data from two different hit-tracking applications.
First off, how are you recording this information? JavaScript hit tracking won't get people who have JavaScript off. Server-side tracking can keep track of direct image accesses, but these are completely invisible to anything at the HTML level. Project Wonderful is an advertising network, so we're going to records hits with both JavaScript and with fallback straight HTML. Most ad networks only use JavaScript, but we're not most ad networks!
Next, WHAT are you tracking? Do you care about people who load the site entirely, or who just access it before clicking away? Our answer is simple: we care about who sees the ads. So if you put an ad box at the top of your page, and one at the bottom, those two ad boxes will likely record different hits, because not everyone will always wait for the full page to load. You'll see more traffic on your top ad box, and that's great, because that ad box is actually seeing more traffic. More people are looking at it!
Finally, there's the matter of volume. If you're recording every hit on a page, that means at the minimum recording what page it was on and which IP accessed it - for every single hit across the entire network. And you need to be able to process, analyze, and summarize this data in real time. Imagine all the traffic from every single site on your network being funneled into your servers. You need to be keep up!
In the past, Project Wonderful's hit tracking scaled up as sites did: if you got 100 hits a day, we tracked every one of those, but if you got 50,000 then we would randomly sample the traffic stream. It's sort of like how surveys work: you don't call up every single person in the country to ask them what they think about Doritos, you call up a representative sample and extrapolate from there. But the problem with random sampling is that there's that pesky "accurate to within 1 percentage point, 19 times out of 20" phrase you've often heard afterwards. When you're randomly sampling, you're never 100% sure you're capturing everything, because you are by definition missing some things. It's really unlikely, but something could slip through!
Because of this, we've been investing in infrastructure behind-the-scenes to move away from random sampling. Very soon, Project Wonderful will be turning on our new hit tracking code. Every single page view is now logged, processed, and analyzed in real time, and random sampling is a thing of the past. It's an impressive feat (one that's normally only possible when you have Google-level infrastructure) but it was something we wanted and it lets us support some new features, so it's been something we've been working towards for months. This code has actually been running in parallel with our current hit code a while, which has let us debug it and ensure everything's working as it should, but very soon we'll be switching over and displaying this new data exclusively.
We'll make a new post when it's online, so you'll know. Smaller sites will notice no difference, as random sampling doesn't affect them. Larger publishers won't notice much of a difference in their numbers: page views will be almost identical, but unique visitors will likely be smaller (now that we're tracking every single display, we can give more accurate numbers here).
Stay tuned, and we can't wait to share the exciting new features we've got coming down the pipe!
prguitarman
baratheonpilled
2087
vaporwave
amandla
titsay
