In my last post I talked about a method for testing the value of using AdWords on your brand.
The results are now in:
- Average Hourly Revenue from brand keywords when Running AdWords: £3471
- Average Hourly Revenue from brand keywords without AdWords: £3278
- Average Hourly Cost of brand keywords: £0.86
WIN!
Methodology
Traffic in this account varies quite a lot depending on the day of the week so I chose to segment by hour rather than day of the week. I wanted to keep the slices as fine as possible to reduce the variation caused by the time of day so I used the following day parting scheme:
I waited a month before analysing the data; for this account a month is plenty of time for the result to be statistically significant.
To get the information I needed I opened the Total Revenue part of the Ecommerce Report in Google Analytics.
After exporting the information to Excel I could easily find the averages that I needed. To make the test as fair as possible I only looked at the hours between 1600 and midnight. This may mean that my results are invalid during the rest of the day
I always found it hard to do a correct A/B test on this. That’s a great post. I had an appointment with Google the other day for our campaign, and they were pushing us to use site links on our Brand Terms, which I said I was not interested because we are all over the place on Page 1.
Adwords Site links on Brand terms is in my opinion only a way for big G to get some more clicks on their sponsored links :-/
Hi Richard,
brilliant idea for testing brand terms for google adwords!
I tested something similar recently, but my results weren’t as actionable as yours…
cheers, ulf
@Phillipe
Unfortunately my method is not a true A/B test. As far as I know there is not way to randomly assign people to the “see AdWords ad” vs. “don’t see AdWords ad” bucket. My methodology is just the best I could come up with given the tools I have.
The client in my case study does not own the organic results for their brand name so I guess that your results for a similar experiment may be very different.
@Ulf
Do you mean that any result from your test would not have been actionable? Or that the result you got was not statistically significant or not conclusive in some other way?