Abstract

The coronavirus pandemic resulted in both an immediate decline in digital ad spend and a shifting of that ad spend to direct response campaigns. As a result, publisher ad revenue declined as well.

Currently, buyers and publishers use ad viewability as a measure of inventory quality. However, it is a flawed metric. Buyers need more accurate ways to predict campaign performance and deliver higher return on ad spend. This study demonstrates that buying inventory based on reader engagement delivers higher performance and higher return than buying based on ad viewability.

—

CONTEXT

Coronavirus industry effects

As a result of the COVID-19 pandemic, we have seen immediate and noticeable changes to buyer behavior across the Sovrn exchange. We examined those macro trends, and this is what we found:

Across the Sovrn exchange, advertising spend decreased ~40% from Jan-Feb 2020 to March-April 2020.
During the same period, online traffic simultaneously rose an average of 17%.
Reader engagement (see definition below) also rose 9% on average.
A significant portion of the remaining advertising spend shifted to direct response campaigns.

As a result, publisher revenue dropped precipitously. It became clear that publishers needed a way to offer more valuable inventory to buyers.

At the same time, buyers want better metrics that allow them to both assess the quality of supply and predict campaign performance.

Tracking engaged time

In order to measure reader engagement, we identify and track reader behavior. We achieve this through a proprietary DOM event listener that tracks 40+ reader actions, such as page scrolls, mouse clicks, or keyboard activity. This allows us to effectively identify every single interaction between user in page.

We simultaneously compare these reader engagement events with both user time-on-page and the amount of time that a given ad unit is viewable. Whenever the reader is demonstrably active and the ad unit is 50% in view, this constitutes engaged time.

Engaged time ends after 10 seconds of reader inactivity, or when the reader performs a disengagement event, such as changing tabs or typing into the address bar.

Methods

We deployed our tracking beacon across 2,238 domains within the Sovrn exchange. We then measured the amount of time that each impression we tracked was in-view and compared it to the amount of time that each impression was in-view with an engaged reader. In addition, we tracked whether or not a given impression was clicked by a user. Finally, we compared CTR between engaged time and viewable time.

In order to ensure a direct comparison, we measured both engaged time and viewable time on every impression we tracked. This ensured that the sample set was the same between both measurements. Our study surveyed a wide range of sites, from national newspapers (e.g. mirror.co.uk) to tiny cooking blogs (e.g. anerdcooks.com).

In order to verify our click tracking data, we compared our results to those provided by Google Ad Manager.

Discussion

These results support our hypothesis that engaged time is a better predictor of campaign performance than viewable time. Engaged time therefore demonstrates a number of applications for both publishers and advertisers, and offers higher marketplace efficiency.

The most noticeable effect of this performance increase is return on advertiser ad spend. Buying 100 hours of engaged time supplies buyers with a 2.71x higher CTR than buying 100 hours of viewable time.

Similarly, studies have shown² that brand recall rises sharply when an advertisement is in view for approximately 30 seconds. Therefore, buying based on engaged time has the potential to boost the performance of awareness campaigns, a topic that deserves further study.

Publishers with a highly-engaged audience are therefore in a position to increase their revenue. Segmenting and packaging this highly-valuable inventory delivers better results to advertisers, and should therefore command a higher price.

While these results are incredibly exciting, there are myriad opportunities to increase both the uses of engaged time and the adoption of an industry standard.

For example, initial internal studies have already shown that targeting engaged users with on-page notifications (e.g. subscription prompts) leads to improved performance. There are a number of next steps we plan to take in order to build on the results of this study:

Measure brand awareness campaigns with eye-tracking studies

Having demonstrated the value for a single performance metric (CTR), we hypothesize that reader engagement has similar effects on brand awareness and recall.
Grow publisher revenue through an engaged ads marketplace

Building an engaged ads marketplace is the first step towards standardizing the practice of buying on engaged time. While we have publisher participating in an early version of this marketplace, publishers should be able to transact on engagement across the marketplace.
Expand use cases for engagement signals

As previously noted, engagement signals have myriad potential uses. We plan to partner with publishers to better understand how this technology can improve performance across more aspects of a publisher's business.
Push for standardization of engagement measurement

Currently, there is no shared industry definition of engagement. Building one will allow the growth of trading on this new metric.

Advertising performance

Using engagement to predict CTR

Abstract

—

Coronavirus industry effects

PROBLEM

Hypothesis & definitions

—

Engaged time

—

Viewable time

The problem with viewable time

Tracking engaged time

—

340+

—

40+

—

Measuring response rate

—

Initial method & results

Methods

—

2,238

—

3,650,882,131

—

6,789,688

Results

Discussion

—

Advertisers

—

Publishers

—

Moving engagement forward

Let's talk

Authors

—

Brian Bouquet

Babac Vafaey

Owen Naughton

Jasper Lipton