For PPC managers, regularly analyzing keywords is decisively important when it comes to monitoring and directing AdWords accounts. However, due to the typically enormous volume of keywords, this can become especially tedious. When the goal is to figure out which “type” of keyword should be added, deleted, or modified, it’s only a matter of time before the limits are reached and performance has to be evaluated on a higher aggregation level, such as the AdGroup or campaign level.
Semantic Attributes Make New Aggregation Levels Possible
The average conversion rate on the AdGroup or campaign level doesn’t tell us much if the keywords therein are performing very differently. In the following example, the AdGroup “Adidas Running Accessories” consists of three keywords:
- “adidas sneaker”: CR = 2.0 %
- “adidas running shirt”: CR = 0.6 %
- “adidas shoe laces”: CR = 0.1 %
Assuming that these three keywords have similar traffic, the AdGroup’s conversion rate is at 0.9 percent. One could decide to ignore or deactivate this AdGroup, but then the important keyword “adidas sneakers” would be lost.
To avoid mistakes like this, we use a semantic attribute system which is able to categorize keyword strings. For example, the keyword “buy adidas originals shoes stan smith” is split up into four attributes:
- buy = Buy_Word
- adidas originals = Designer
- stan smith = Specific
- shoes = Category
As a result, a further aggregation level is created that gets “closer” to the keywords.
The Attribute System Enables New Insights into AdWords Account Performance
This attribute system can be used for the entire account. Attributes can also be combined with one another, and this can provide an overview like the following example:
|Attribute Combination||ROAS||CPC||CR||CTR||NC Rate|
Graphically, these relationships look like this:
The single attribute combination Designer contains keywords such as “adidas,” “nike,” or “jack wolfskin.” The combination DesignerSpecificCat, however, encompasses keywords like “adidas shoes stan smith.” The primary advantage of the attribute system is that we no longer have to rely on the Google structure alone. Based on the table, we can see that the combination DesignerSpecificCat has the highest conversion rate (3.71%) and the highest CPC (€0.52). In this case, it appears that Bid Management has already placed optimal bids.
Keywords can be Analyzed Specifically, Leading to Further Optimization Steps
In this example, we assume for simplicity’s sake that ROAS is to be optimized and that it should have a value over 4. First, we define a “Patient Zero,” an attribute combination on which to base our analysis. In this case, it is Designer.
ROAS is normalized in order to enable comparisons between the different attribute combinations. Formally, we calculate the normalization as follows: ROAS_normalized_a = ROAS_not_normalized_a * CPC_a / CPC_ basecombination. As previously mentioned, “CPC” here is the CPC of the attribute combination Designer. This is the base CPC, also known as the “Patient Zero.”
Example: ROAS_normalized_DesignerCat = ROAS_not_normalized_DesignerCat * CPC_DesignerCat / CPC_Designer => 3.24 = 3.75 * 0.38/0.44. This normalization method makes it possible to compare the ROAS of different attribute combinations, even though bid management prices every combination differently.
Next, it’s necessary to determine which attribute combination and/or which “growth pathway” can improve account performance based on the ROAS target most efficiently. Here, we present a procedure which can be helpful on the attribute level. Normalized ROAS values can be calculated using CPC:
|Attribute Combination||ROAS||CPC||normalized ROAS|
Assuming that shopping cart values are independent of CPC and that we would pay the same price per click for all areas, the long-tail area of the account should be built up: that is, the combination Designer + Specific + Category. The combination Designer + Category is not of immediate concern to the PPC manager.
Smaller Data Volumes, Faster Decisions
Using semantic attributes to analyze keywords is especially helpful for preventing overlap effects and it enables a closer keyword analysis than is possible on the AdGroup or campaign level. Additionally, the data volume required for analysis in Excel is substantially smaller. In a well-developed account, the amount of data to be analyzed is reduced by about 90%. Nevertheless, we can still describe the character of each keyword group, and in the case of new customers, we can create significantly better AdGroup- and campaign structures. This allows us to collect data more quickly and make decisions about the direction into which the account should be developed.