Finding the Negatives in AdWords

Unless you only use Exact match, you will get traffic you did not want through Google AdWords. Using either Broad or Phrase match types means giving up some control over what your ads appear for in exchange for assistance from what is arguably the largest database documenting human behaviour on the planet. There is a tool to help you manage this: negative terms.

Google is on record saying:

“Twenty percent of the queries Google receives each day are ones we haven’t seen in at least 90 days.”

Broad and phrase match makes it easy for an advertiser to deal with this and keep the size of keyword lists under control. Both of these match types can also produce irrelevant traffic. The terms added to the campaign may be similar or frequently associated with the keywords in normal search. This does not mean they are always relevant.

Irrelevant queries are not the only reason traffic might be unprofitable. Issues such as competition, pricing, geography or other circumstance can render some queries too expensive to be profitable, or even accessible for your campaigns.

Setting negative terms is the easiest way to reduce a campaign’s exposure to bad traffic. A negative term prevents an ad from showing for search queries containing it. In campaigns using terms relevant to multiple industries, negatives make a difference to the cost per acquisition.

Finding the Z for X and Y

Finding negative terms for a campaign is an interesting exercise. Unproductive clusters of terms can be found using any tool that allows for filtering by words and analysis of relevant metrics.

For example: A campaign is over target CPA, and features two keywords, ‘A’ and ‘B’. Both keywords are using broad match. Historically these terms have generated sales and a consistent level of traffic. A search term report showing all search terms for keywords ‘A’ or ‘B’ reveals a number of other commonly occuring terms for which the ads display. Each of these terms above a certain threshold can then be analysed.

A quick analysis of other terms that account for a certain percentage of impressions (or other metric of choice) has revealed that traffic with the term ‘Z’ frequently exceeds the maximum allowed CPA, whilst other significant terms like ‘X’ and ‘Y’ are on target. So what now? Should it produce minimal sales, it should become a negative term. If it does produce some sales, the traffic for the term can be analysed again, to identify if there is another term that should be removed from the campaign.

As the number of significant terms in a campaign increases, so does the complexity of the analysis. In the example attached, the campaign’s targeted CPA on reporting period has been revised, and it needs to be adjusted to match the new target.

The dataset used in the example contains a range of descriptive terms that relate to locations, product description, usage and alternative products. In this example, hierarchical clustering reveals a few terms that appear across multiple nodes. Each node is split until it reaches a threshold, after which the data would cease to be statistically significant.

The Intersection of Clusters

What you are looking for is the groupings that have the greatest impact on the metrics the campaign is measured on. If it is return on investment, the CPA is used for assessing the clusters of terms; if it is time on site, or even raw volume, those would be used instead. While this can be done within Excel, across multiple worksheets, the information is best dealt with as a network, and visualised as such.

And how to get it all wrong

There is one huge error that can lead to otherwise productive terms being discarded: decisions made without enough data, either as a result of a small dataset, or overfitting of terms. Search Engine Marketing traffic is very prone to volatility at low volumes, and the problem becomes more pronounced the lower the chance of a sale per click is at the ad.