Ever since 2010, a lot has been written about PPC keywords cannibalizing one another. As the term “cannibalization” suggests, this Google AdWords behavior of deciding which keyword to trigger is considered dangerous for your PPC campaign. But is it really as bad as public opinion suggests?
I would like to share an analysis with you that shows what impact match type cannibalization really has on an account.
The basic concept of match type cannibalization
If two different keywords trigger the same search query, the ad will be displayed for the “most relevant” one. Most relevant in this context means, the keyword which Google considers to generate most profit for Google will be triggered. If you are familiar with Google’s ad rank concept (consisting of max CPC and quality score), that’s precisely what we are talking about here.
Example: Imagine you have the following two keywords in your account:
- mustang blue jeans (broad), max CPC: 0.24, quality score: 4
- mustang blue jeans (exact), max CPC: 0.20, quality score: 10
Whenever someone enters the search query “mustang blue jeans” in Google, AdWords has to decide which of the two keywords to trigger, in other words: the two keywords will compete one against another. As the exact keyword has a higher quality score it is most likely that this keyword is triggered even as it has a slightly lower max CPC. However, be aware that the quality score (QS) 4 is an average whereas Google calculates the quality score for every combination of keyword and search query! So even as “mustang blue jeans (broad)” has an average QS of 4, it might have a QS of 9 or 10 for the search query “mustang blue jeans”. In this case, the broad match would be triggered more often than the exact match due to the higher max CPC.
Setting up the experiment – part 1
The easy way to get an insight how severe match type cannibalization affects your account is by analyzing your search term report and compare it with the active keywords in your account. What was most interesting for us was how exact keywords were cannibalized by other keywords, so we focused on exact keywords in our first experiment. All search queries that (1) matched with phrase or broad matching according to the AdWords search term report and (2) were identical to active exact keywords were considered as “cannibalization”.
The following points might also be important to mention:
- Be aware that in addition to match type cannibalization some cases of word cannibalization will occur in the test: for example the keyword “adidas shoes (broad)” can not only cannibalize “adidas shoes (exact)”, but also “brand new adidas shoes (exact/ phrase/ broad)”.
- Our bid management system ensures that always the more exact match type of one and the same keyword gets the higher bid.
- As this approach won’t work for dynamic accounts, we only used accounts where no keyword changes were committed during the evaluation period.
Cannibalization rate based on clicks: 5%
The cannibalization rate was a lot smaller than we expected: on several large accounts of apparel and shoes retailers we witnessed an overall cannibalization of 5% (+- 1%) based on clicks.
Demystifying the myth
Ok, only 5% of our exact keywords are cannibalized by other keywords. But what we still don’t know: is this good or bad for our PPC account and does it keep our wallet happy? As mentioned before, public opinion tells you that cannibalization is bad – but I never came across any convincing argument why it should be so. Let’s just have a look at the common misconceptions:
Cannibalization increases your PPC cost because Google always picks the higher bid.
Ok, let’s say the last part of this statement is true in most cases. Still, it won’t increase your overall PPC cost, because it is you who decides how to set your bids. Google picks the higher bid, but your bid management will then lower this bid until it has reached whatever your bidding strategy is up to.
Google will always take broad matched keywords first.
Wrong. Google will always pick the keyword/ ad combination that is likely to generate most profit for them.
Cannibalization gives you a much harder time gathering statistics.
This is indeed a good point. As the statistics for one and the same search query will show up divided into two (or even more) different keywords, you’re automatic bid management will sometimes come to queer conclusions. But given the fact that all statistical data is shown on search query level as well, statistic tools could easily be able to gather the right information. So still this isn’t a valid argument that cannibalism is bad – it just means that you need to be aware of cannibalism and use the right tools in order to manage your account well.
Setting up the experiment – part 2
The first experiment helped us to get a first impression to the extension of match type cannibalization. Now we wanted to know how cannibalization influenced important KPIs such as CPC or click-through rate (CTR). So we started another experiment. With AdWords Campaign Experiments (ACE) we chose a set of keywords and duplicated it. The test campaign was modified by uploading negative keywords that prevented all kind of match type cannibalization. Different to the first test, broad keywords that cannibalized phrase keywords were taken in account as well.
Same CPC, same CTR
Two weeks and 2000 clicks later, we got the following findings: there were no statistical relevant differences between the original and the test campaign. CPC was exactly the same, so was the click-through rate (CTR). The test campaign’s conversion rate was a little higher while its earnings-per-click was a little lower – normal statistical dispersion considered the fact that only 150 sales were tracked during the evaluation period.
As it was lots of manual work to keep the max CPCs identical for both original and test campaign, we decided to end the experiment. An important question that is still open for analysis would be what effect keyword cannibalization has on conversion rate. As long as we strictly speak of match type cannibalization it would be really surprising to have any effect at all. But as soon as word cannibalization (see the explanation in part 1 of the experiment) and different ads with different accuracy enter the stage, things could become interesting.
Conclusion: Extend of match type cannibalization is highly overestimated
After the second experiment, we came to the conclusion that match type cannibalization is highly overestimated by most PPC experts (always under the premise that that the more exact match type of one and the same keyword is priced with the higher bid).
Match type cannibalization does of course take place in a certain amount: As our first experiment shows, about 5% of all clicks in an account were triggered by phrase or broad matched keywords, although the exact keyword matching the search query existed in the account. However, the effect on both CPC and CTR were tending towards zero.