Welcome to WebmasterWorld Guest from

Forum Moderators: DixonJones & mademetop

Message Too Old, No Replies

Google Analytics conversion attribution is wrong


10:18 pm on Apr 10, 2012 (gmt 0)

Junior Member

10+ Year Member

joined:Oct 19, 2004
posts: 169
votes: 1

This headline is a bit misleading. What I really mean is that Google Analytics is limited to the technical limits of JavaScript and cookies. These limitations are very limiting. This is probably why there has been so much interest in persistent technologies like Flash cookies. Ive spent the last few weeks analyzing my sales and Ive found that an astonishing 77.4% of sales are misallocated by Google Analytics.

Ive used a proprietary tracking system since 1999 whereby I log session starts in one table and page loading in another table. All 13tb going back to 2003 is in a live production environment. In the session table I log referring URL, IP address, browser type So I have a way of comparing GA to my own tracking system one that does not rely on JavaScript. In the first wave of my analysis I found that my system was misallocating sales 62.3% of the time. So my tracking system was not performing terribly better than GA. The problem is the cookie. They just dont stick like they used to. Mostly because users have multiple devices (home computer, work computer, tablet, smartphone). Moreover, users have multiple email addresses. 8% of my sales went to users with 4 or more email addresses. 27% of my purchasers had 3 email addresses. I guess it makes sense.

My sites are very engaging and depend heavily on e-mail communication. We have a back-end user area that requires log-in. Much of my traffic was originating from my own emails. So two months ago we set out to marry user cookies together using email address. The results were astonishing in that 55.7% of orders got reallocated to other advertising sources (ie: PPC was allocated to SEO, SEO to PPC, and one PPC term to another). We found that PPC users had a propensity to use lots of PPC terms. Our expensive advertising got more affordable and our inexpensive advertising got more expensive. Bottom line, by using e-mail address to stitch together cookies were given a cleaner picture of our advertising efficiency. This is going to enable us to turn-on thousands of keywords that were shut down by flawed analysis.

Stitching cookies together time to order:

38% of all orders moved from organic/no-referrer to PPC.

Average days to order: 252.
23% of orders came within 1 day.
9.5% of orders came within 2-10 days.
24% of orders came > 60 days.

Google Analytics time to order:

Average days to order: 31.
66% of orders came within 1 day.
0% of orders came within 2-10 days.
9.5% of orders came after > 60 days.

GA is pretty good at allocating the most recent user session to advertising source.
5:56 pm on Apr 12, 2012 (gmt 0)

Full Member

5+ Year Member

joined:Oct 26, 2010
posts: 229
votes: 0

Nice analysis. Have you tried installing Google Multi-Channel Funnels? Should help you get closer to what's really going on.

You bring up a great point that I think isn't talked about enough: How many errors our analytics platform have. It's pretty shocking.
3:43 am on Apr 26, 2012 (gmt 0)

Junior Member

10+ Year Member

joined:Oct 19, 2004
posts: 169
votes: 1

Nobody cares. Just me looking for answers. Since GA is free people just use it and assume that it is correct. There's opportunity here.

Yes, I've used multi-channel funnels. Same basic problem. Cookies just don't do it anymore. Perhaps, someday G will use that great G+ data to stitch cookies together to get a true view of conversion attribution. I'm sure they know the weaknesses of GA. And I'm sure they're using their data in a smart way.