Campaigns that use Sponsored Display perform better

• In the first approach, we created and compared three brand groups based on ad product adoption:
• Sponsored Products only;
• Sponsored Products + Sponsored Brands;
• Sponsored Products + Sponsored Brands + Sponsored Display.
• In the second approach, we compared the actual sales of the advertisers that adopted Sponsored Display vs. the predicted sales had Sponsored Display not been adopted (over a 50-week period).
• In the final approach, we used machine learning to determine the next-month causal impact of adding Sponsored Display to ad campaigns.

In the first approach we compared the year-over-year (YoY) success of total sales, ad-attributed sales, and return on ad spend (ROAS) by analyzing three different ad product mixes:

• Sponsored Products only (baseline)
• Sponsored Products + Sponsored Brands (+6% total sales YoY, +10% ad-attributed sales YoY, +0.8 higher ROAS).
• Sponsored Products + Sponsored Brands + Sponsored Display (+16% total sales YoY, +25% ad-attributed sales YoY, +2.6 higher ROAS)

In summation, we find that the most successful ad mix in terms of total sales, ad-attributed sales, and ROAS is achieved when using a mix of all three ad products.

The final approach measured the next-month impact of activating Sponsored Display (product targeting), specifically, the next-month impact on new-to-brand customers, detail page views, ad impressions and total sales.

Brands that began using Sponsored Display product targeting for the first time saw, on average, 33.9% more impressions, 3.6% more detail page views, and 2.6% more new-to-brand customers the following month compared to those that didn’t. Similarly, brands that created a Sponsored Display product targeting campaign for the first time saw, on average, 28.8% more total sales, 4.2% more detail page views, and 2.6% more new-to-brand customers in the next month compared to those that didn’t.

Using multiple methods to test the effective of Sponsored Display, we conclude that brands that incorporated Sponsored Display on top of Sponsored Products or Sponsored Products + Sponsored Brands experienced increases in total sales ranging between 10% to and 29%, as well as increases in impressions, detail page views, new-to-brand customers, ad-attributed sales, and return on ad spend, compared to brands that didn’t adopt it. Because of this, we recommend that advertisers consider mixing Sponsored Display into their ad mix.

We also recently concluded another study that tested the effectiveness of adding Sponsored Brands video to campaigns and found similar, positive results. In the future, we will test the effectiveness of all four ad products to see if there is an optimal mix.

We analyzed just over 43,000 advertisers between April 2020 and June 2021 using three approaches:

• First, we performed comparative analysis on 11,000 of these brands to determine if there was any correlation between incremental increases and the inclusion of Sponsored Display with Sponsored Brands and Sponsored Products in the long term.
• Second, we conducted Sponsored Display activation analysis on 284 of these brands to determine if Sponsored Display audiences was the cause for any improvements over a 20-week period.
• Third, we employed a machine learning causal inference analysis to determine if Sponsored Display was the cause of any improvement in their KPIs the next month after activation, in the shorter term. We matched around 4,600 brands out of 43,000.

We found that a balanced strategy that incorporated all three products was correlated with performance increases among the 11,000 brands we studied using comparative analysis.

Turning to the individual analyses:

• First, while we executed causal impact analysis to measure the lift in specific performance indicators by Sponsored Display targeting tactic (product targeting and audiences), we complemented this analysis by creating a higher-level view of the benefit of using Sponsored Display by executing a non-causal analysis within a longer-term time period of one year. Specifically, we analyzed 11,394 advertisers in the US marketplace (both sellers and vendors), by dividing them into groups depending on their use of the three advertising products we were analyzing (Sponsored Products, Sponsored Brands and Sponsored Display). The three different ad-product usage defined groups we analyzed were:
• Sponsored Products only (which we considered the baseline)
• Sponsored Products + Sponsored Brands – this group may use Sponsored Display with a low (<4%) investment
• Sponsored Products + Sponsored Brands + Sponsored Display—this group used Sponsored Display with product targeting (>70%) and audiences (>10%) tactics
• Second, we conducted a causal impact analysis for advertisers using Sponsored Display audiences for the first time to infer the causal impact on total sales based on a diffusion-regression state-space model that predicts the counterfactual response that would have occurred had the advertising targeting action not taken place in a medium term of 20 weeks. For this study we selected 284 advertisers that satisfied specific conditions within a 50-week timeframe. Specifically, in 2021 they were using Sponsored Products and/or Sponsored Brands for at least 30 weeks prior to Sponsored Display audience activation, and, in the last 20 weeks of the analyzed medium-term period, the only action they took was launching a Sponsored Display audience campaign without any other advertising-specific action. We then calculated the sales impact for the 20 weeks following SD-Audience activation by predicting the counterfactual. The model is trained using the first 30 weeks to predict the counterfactual response for the following 20 weeks, had no action taken place. The lift in sales is measured by subtracting the observed sales (real value) from the counterfactual sales (predicted sales). The prediction uses 10 covariates including sales, units, glance views, etc. at a vertical aggregation level and compares the same covariates for 600 advertisers that never activated Sponsored Display within the same timeframe.
• Third, to measure the causal impact of advertisers who adopted Sponsored Display product targeting for the first time, we used machine learning to determine the effect of taking an action on advertiser performance in a shorter-term of one month. Our current methodology is based on a method called 2-stage GP (2-stage Gaussian Process) that shows improved performance on various causal inferencing performance metrics as compared to existing methodologies such as Double Machine Learning and Causal Forests when applied within the context of advertising.