Two tactics used more often in top-performing Amazon audio campaigns
By: Michael Wilson, Sr. Analytics and Media Manager
Measuring the results of audio advertising can often be a challenge for advertisers because, unlike display or video ad campaigns, audio customers are usually further than one click away from a purchase. This means that to attract customers, advertisers first have to motivate them to perform another, non-purchase action, before they can make an actual purchase. Advertisers often attempt to do this by first inspiring consumers to view product pages. In this article we explore the insights of a 2020 Amazon Ads study on audio campaigns. Specifically, we explore actions that differentiate top-performing audio advertisers from other advertisers.
We quantify the key actions by detail page view rates (DPVRs), which is the calculation of the number of page views from ad-exposed audiences in relation to impressions delivered. Using this metric, along with a machine learning algorithm, we are able to identify insights that may help improve DPVR.
For more information, see the Methodology section at the end of this article.
Top-performing audio advertisers were 24% more likely to use Amazon lifestyle audience segments (feature is only available in Amazon audio’s free, ad-supported tier) than other advertisers. Our analysis indicates that lifestyle and in-market audio segments are the most commonly used audience segments by top-performing audio advertisers. These segments also perform higher than demographic segments in terms of reach and DPVR.
Top-performing audio advertisers, on average, ran audio ad campaigns 10 days longer than other advertisers, month over month. By running longer campaigns, advertisers may reach increase unique reach. Not only did the top-performing campaigns run longer, but they were also heard more often (top-performing audio advertisers used a 5-6X frequency cap on average compared to an average 4X frequency cap for other advertisers).
Our analysis shows that top-performing audio advertisers ran ongoing audio campaigns for a minimum 30-day campaign period and ran them at either a 5X or a 6X frequency cap as a way to help increase consideration (DPVR). This longer campaign duration time period is important because it allows more time for algorithms to analyze campaign performance and provide recommendations for in-flight and future campaign optimization. The greater frequency associated with these campaigns is also important because it may help improve brand awareness.
The methodology in this study is comprised of five components: campaign selection, creation of a success metric, identification of effective advertising actions, advertiser ranking (based on segmenting advertisers into four clusters), and then a comparative analysis to identify the attributes that separate the most- and least-performant advertisers in terms of success and strategies.
We analyzed 176 US audio campaigns in 2020 to determine which strategies are most effective in making campaigns successful. The success metric chosen for this study is DPVR for advertisers who sell in Amazon’s store. Using this metric, we then used a machine learning algorithm (e.g. Pearson Correlation and XGBoost) and subject matter expert suggestions to assign feature weights to identify the top campaign actions tied to higher DPVR. After identifying the top campaign actions, we ranked campaigns based on how often customers performed these actions and for which campaigns customers performed these actions the least, and provided those results in this article.
How are advertisers distributed across the clusters?
We used machine learning algorithms to automatically classify advertisers into clusters based on their advertising and retail attributes.
How does the clustering work?
We created a binary composite score based on DPVR, and then applied an XGBoost classifier to identify which features at which weights best predict these labels. Features are advertising or retail actions such as ad product usage intensity and mix, timing of advertising support, tactics of targeting, creatives and placements, customer review counts and ratings, percentage of products with quality product pages, and the types of products promoted in ads.
Using the identified features and weights above, we then applied a k-medoid clustering algorithm to classify advertisers into clusters. Note that we classified advertisers by their actions rather than by the components of their composite score. Next, we ranked the final clusters by their composite scores from high to low. Cluster 1 is the most successful cluster with the highest composite score, and Cluster 5 is the least successful.