As more impressions become available on ad exchanges, B2B marketers are taking advantage of the additional available inventory. However, some of the inventory available is of very low quality and often below the fold or loaded at the bottom of a webpage. If the user doesn’t scroll down far enough to see it, that impression will not be seen, but will still count as an impression for purposes of the marketer's bill.

Viewability is an online advertising metric that aims to track only impressions that are actually seen by users. As inventory becomes ever more available in programmatic advertising, and with more than half of online ads – especially online video ads – never seen, viewability is an increasingly important factor in advertising purchases. The Media Rating Council (MRC) sets what is considered “viewable.” Desktop display is defined as viewable if 50% of the pixels are in view for a minimum of a continuous second.

Applying the MRC standard for viewability, above-the-fold ads are not always viewable, and below-the-fold ads sometimes are. Based on past research from DoubleClick, median viewability for above-the-fold ad units is 68% while median viewability for below-the-fold ad units is 40%.

‍

Multi-Touch Attribution Models and Viewability

B2B marketers are looking for tools and methods to optimize media buys to maximize the impact of media investments, sometimes taking into account viewability data. For that purpose, many marketers leverage various attribution methods and technologies.

When using last-touch attribution methods, if a marketer is interested in tracking the brand impression impact of a display ad – even if it's a direct response ad – s/he will assign 100% of the credit of a conversion to the last impression served to the user, even if the impression was not viewable. This attribution method is commonly used for display by advertisers using 3rd-party ad servers.

However, a not-viewed impression by definition has had zero effect on a buyer's behavior and therefore should be assigned zero credit for his or her conversion. Even in simple last-touch attribution methods, only viewable impressions should be assigned credit for the last touch, yet the ad server is not able to determine the viewability of all impressions.

More sophisticated, multi-touch attribution methods are designed to measure the impact of marketing touchpoints on buyer behavior. A good attribution model is designed to determine how much more likely a buyer is to convert after he or she engages with, or is at least exposed to, a particular ad impression.

Since buyers are exposed to multiple touch points across their journeys, marketers want to understand how each point influences their purchase. Knowing the touchpoints’ relative importance enables marketers to allocate resources properly across the different media channels.

Attribution experts have argued in various articles that attribution platforms need to exclude out- of-view ads so that credit is not assigned to non-viewable impressions and marketers and their media partners are incentivized to target higher-quality inventory. The challenge is that not all impressions or placements are measurable for viewability, so it is not currently possible to determine viewability for each impression and exclude non-viewable ads from attribution.

However, it is possible to estimate the average viewability of a placement and to insert that information into multi-touch attribution. With the adoption of sophisticated AI models for multi-touch attribution, the question is whether these models can estimate the true impact of display impressions without using any viewability data to determine the percentage viewability of each placement. Are attribution models smart enough and granular enough to assign credit to display impressions in a way that differentiates among their viewability rates? The recent advancements in AI learning models seem to indicate so, with the ability to determine the differences in impact per impression for different ad placements and that placements with higher viewability are assigned higher weight than placements with lower viewability.

This is extremely useful for marketers for various reasons. First, it eliminates the need to use viewability data in attribution models when the attribution models are sophisticated enough to measure the impact of each impression by placement. Note that many marketers have thousands and thousands of placements, so AI learning models need to deal with thousands of features without overfitting the data.

Secondly, it also confirms that viewability is a key variable in driving conversions, so advertisers do well if they optimize the impression buys to maximize viewable impressions for their target audiences. Demand Side Platforms (DSPs) enable advertisers to optimize impression bids based on likelihood that impressions are viewable. For instance, DoubleClick Bid Manager uses many signals (e.g. URL, time of day, page category) to predict, impression by impression, the probability that it will be viewable and dynamically adjusts bids up or down based on that probability to deliver the viewable CPM target.

Thirdly, this eliminates the need for media buyers to place minimum viewability thresholds within the DSPs or for them to become fixated on the need to hit a viewable percentage. Setting minimum thresholds for viewability not only makes the CPMs much higher than necessary (there is more bidding competition on impressions with high viewability), but it also eliminates the ability to place ads where viewability information may not be available (e.g., some mobile web or in-app placements).

If marketers know that attribution models will account for viewability, their media optimization can then focus on cost-per-conversion by placement, which takes into account CPMs and attributed conversions. Advertisers can bid on placements with any viewability, knowing that low-viewability placements are generally less expensive per impression yet may still have some impact on conversions. In fact, low-viewability placements may offer special opportunities to increase reach and drive conversions from users who would not otherwise have seen any ads.

In conclusion, advertisers should look for sophisticated attribution models that can take into account viewability rates as well other attributes, such as campaign, publisher, creative and placement, so that proper credit will be assigned to display impressions.