On August 11 Google Analytics changed their attribution model and it’s had a bit of a dramatic effect in some accounts. Visits are way up, and metrics that depend on visits such as e-commerce conversion rate are consequently affected as well. There was a bug that was fixed on August 17 that did inflate visits even when the campaign source did not change, but visits seem to be back to “normal”.
Some people seem quite upset about the change, but I am not sure whether this is about the temporary bug or about the underlying change, which in my opinion is great news. I totally disagree with comments such as this one:
As a data analytic, the most important thing is data consistency. The ability to measure changes in traffic, conversion, bounce, etc. is the most important aspect of an analytics effort.
As with any model, it’s not so much about being “right” or “wrong“, but whether the model is useful. Does your model help you understand your customers a little bit better and make smarter decisions? If there is a better model then you should change to that one.
Ironically, the data causing higher visit numbers was there all along! It was just hidden due to the way Google Analytics used to count visits, pageviews, etc. Anyone remember 0 visits?
Here is what is happening by way of an example.
1) Grab this custom report for the new Google Analytics and pick a date before Aug 11
2) Filter the data to only show keywords where visits = 0
I see something like this:
Almost 6% of visitors have such 0 visit keywords. Now go try and find any of these keywords in your regular keywords reports. You won’t be able to find them because they have been taken out.
Now pick a date after Aug 11.
Almost all of the 0 visit keywords have disappeared because they are now attached to a visit. Hence they will show up in your regular reports. You should therefore find that you actually have more keywords now.
So there you have it. 6% more keywords in my example, which indeed opens up the possibility of having better attribution management as was pointed out in the post. The bigger the change, the more you should worry about attribution.
Finally, I am not a huge fan of visit-based metrics anyway. Who cares if your dear customers take 1 or 2 visits before placing an order? The main thing is that they accomplished their goals during each visit, i.e. browse in the first visit, have a cup of tea, then buy in the second visit.
Follow me on 
{ 21 trackbacks }
{ 5 comments… read them below or add one }
Do we need nooverride=1 now ?
You say, “As with any model, it’s not so much about being “right” or “wrong“, but whether the model is useful. ”
I have to take exception with this statement. All questions of web metrics always boil down to questions of epistemology. Epistemology is that branch of ancient Greek philosophy which asks the question, How do we know that we know what we know?
You can talk patterns and trends all you want, but as soon as you start running reports you have moved beyond trends and concepts and smack into hard numbers and counts. You can extrapolate back out to trends from numbers and counts, but if the numbers and counts are not auditable and verifiable and if they’re not accurate, how “useful” are those trends? Not very.
Hard numbers and counts are exactly where web analytics users want to be. And they don’t want models, they want accuracy and consistency. Now, if you believe that therefore basically all web analytics users are “wrong” because they “don’t understand” the limitations of web analytics, then perhaps you don’t understand what we really want – an auditable, verifiable trail from web log to analytics report, from page serve to page load to navigation click to conversion and transaction.
I mean at the end of the day, it’s really very simple. Either it did happen on the web or it didn’t happen. Which is it? If it did happen, count it, if it didn’t happen, don’t count it. And prove how you got the numbers. I don’t see where that is asking too much of an analytics platform.
It’s really a simple accounting problem. Let me explain it this way. Imagine getting a note from the accounting department on payday. “We deposited your pay check into your bank account today. With some slight deviations, it’s trending consistent with previous paychecks. However, you may notice a discrepancy in the actual amount which may range from several percent lower than you expected to a magnitude of 10 times that amount less. Not to worry, this is merely due to an inconsistency between our accounting session timeout and the bank’s session timeout. Although we may never know what happened to the missing amount, we’re certain that the remaining amount of your deposit will be useful. And oh by the way, the full amount you expected was reported for tax purposes.”
Hi there. Thanks for the great comment.
I totally understand where you are coming from and that having accurate data is super important, but I would say that web analytics data is not that accurate to begin with. You have to account for technical errors, cookie blocking and deletion, etc.
And Google Analytics does have a data model. According to their model the session timeout is 30 minutes, time on site is 0 seconds for bounces, campaign cookie length of 6 months, etc. You could change for example your own model to set the session timeout to 24 hours if it yields more useful data for you, but the data would then be different. Which model is right?
Now Google Analytics has changed the model and it is unfortunate that they had a bug for a few days when they did that. I also agree that changing the model can make comparisons harder, but I think that the new model is better than the old one and that it allows us to get a better understanding of visitor behavior going forward.
Hey Mike,
Steered over here from Occam. . . wanted to ask why we’re only considering keyword visits here and if there’s a more comprehensive way of considering all potential zero visit sources. Didn’t get great/clear results in GA when I tried modifying the report via sources or some other dimension.
Nate
I was looking for such changes and where they appear,, you have clearly guided me through,,thanks for your help.