Conquent: Without Limits
Conquent: Without Limits
Michael Bissell's Blog

Google Analytics, the cloud and missing numbers #fail

2010-01-26 08:19:38
Shortcut URL: http://t.conquent.com/D800

We ran the backend for a sweepstakes project for ABC/Lost this month -- it was one of those projects you simultaneously love and dread because a company like ABC/Disney can push a LOT of people to the microsite all at once. So, we set up the site on Rackspace's cloud (you can read the details about setting up the project here) and it all worked great.

As a bit of an experiment, we hooked up Google Analytics to the site to track visitors. Now, keep in mind, server stats are always a little tricky, what with bots and other automatic processes hitting your site all the time, but for something as tightly controlled as this sweepstakes site, this was a pretty good test of the technology.

And Google failed.

This conclusion wasn't from an arduous comparison of server logs and weighting what might be a real visit and what might be a bot -- that would be one of those vague, subjective conversations. The fact is that Google reported fewer page views than registrations -- which is impossible as a visitor would have to look at two pages minimum to register. One page to answer the questions, and the thank you page saying "You got the right!"

Google Analytics is run on the client browser with JavaScript; I know visitors had to have JavaScript turned on to register for the sweepstakes because they couldn't have gotten into the site from ABC without JavaScript turned on. So, either Google is overwhelmed and not picking up the visits as we flood the server, or people are blocking Google's scripts.

No matter what the reason, that kind of discrepancy can't be ignored. We usually create special tracking scripts (on the server side) to filter out all the noise in the server logs, which is kind of the thought behind Google Analytics. But our tests so far show that a client side, JavaScript driven tracker, even by the geniuses at Google, just doesn't work...





Next
The Shorty Awards Scandal -- Manual Spam is still Spam
Previous
Helen Klein Ross & Michael Bissell Interview at Adweek's Social Media Strategies Conference


russ: Re: Google Analytics, the cloud and missing numbers #fail
2010-01-26 10:30:16

were you using "vanilla" google analytics or the new(testing) asynchronous one?

http://analytics.blogspot.com/2009/12/google-analytics-launches-asynchronous.html

Note that one of the plusses it lists is "enhanced collection." I can't cite it because I just glazed over it but I read that using this one should get better matches with regards to numbers.



Michael Bissell: Re: Google Analytics, the cloud and missing numbers #fail
2010-01-26 11:46:39

Russ: We were using "plain vanilla"; but this does point out that they saw a similar problem and are attempting to correct it.

Thanks for the link!


Adam Chubak: Re: Google Analytics, the cloud and missing numbers #fail
2010-02-03 20:43:07

I am not so sure it failed. It is suspicious to say the least. What I want to know what kind of 'submissions' were they. For our clients we run two analytic packages, GA and another that connects the form submission itself to all of the contact points. When we get form spam, we see this all of the time. Too many 'submissions' and next to no page views. If its spam we just see a filled out form with no page views.

I would cross check the submission forms. Sounds like bot spam. The spammers are smart and they can fill the form out without even triggering GA.


Michael Bissell: Re: Google Analytics, the cloud and missing numbers #fail
2010-02-03 20:50:27

Adam: Good thought, but it would be difficult for a bot to get to these pages as the project was embedded in an Iframe on ABC.go.com and you had to log into ABC to get to the sweepstakes page. If the microsite wasn't given a unique ID the visitor would get kicked back to the ABC login page based on a server-side redirect (so not a JavaScript thing).

And again, there were MORE registrations than GA pageviews -- and considering we only counted unique user ids (you could only register once per ABC managed account), a bot would have only shown a single registration, or the database would have been polluted with random, fake user IDs, which also didn't happen.


Adam Chubak: Re: Google Analytics, the cloud and missing numbers #fail
2010-02-03 21:00:23

Hmm...Interesting. These are the types of issues, that I love digging into. They drive me nuts, at times ;).

I know GA, is no good with redirects, unless you throw the cookie info into the url. I know you mentioned the server-side redirect, not sure if there were any redirects at play, but may have caused some issues. But not with just raw numbers, only segmenting those numbers. So that probably was not it.

The only thing, that may have played into this that I can think off hand would be in regards to russ. The code may have not loaded fast enough to be triggered. Could have hung on something or people just zoomed by before it loaded. The new async code could help to resolve these issues if they do indeed exist.

<tinfoilhat>With a huge traffic surge from one UA id, it may have tripped alarms at Google and they may have slowed it down.</tinfoilhat>


Comment on this blog
Your name:


Your email (will not be displayed):


Subject


Message



Enter the text above to help us filter spam:


This article also appears on
Web Development