View-through attribution is bad for the app economy


One of the most significant trends in mobile marketing in 2015 has been the impressive growth of mobile video ads, with spend having increased 70% year-over-year from to $2.62BN (projected) in 2015 in the US alone. Mobile video ad spend is projected to reach parity with desktop ad spend by 2018.


It’s not hard to understand why advertisers would pile into mobile video ads: generally speaking, they perform far better than other ad formats on mobile. In a joint study, AppLovin and AppsFlyer determined that video ads perform 40% better at driving “retention” than non-video ad formats. Business Insider found that in-stream video ads perform three times better than other formats (although this includes both desktop and mobile video ads), and Millennial Media observed that video ads performed between 1.7 and 3.5 times better than correspondent standard banners.


It’s not hard to imagine why mobile video ads perform better than other mobile ad formats: video ads allow for a more instructive and nuanced introduction of an app to a user, which helps in communicating a product’s unique selling proposition and differentiating features. Clearly, the development of video ad formats and the continued innovation with respect to how they’re integrated into mobile publishing platforms is good for the app economy.

What is questionable is the worrying trend toward attributing installs based on mobile ad views.

For the uninitiated, a view-through attribution window is the amount of time an ad network may claim that any installation of an app resulted from an ad view. For many networks, this attribution is one day: if a user sees an ad for an app and installs that app within 24 hours, the network gets to “claim” (ie. charge the advertiser) for that install.

The reason networks would want attribution to base based on view-through windows (as opposed to clicks) is obvious: the networks have the opportunity to claim more installs, and will probably force anyone spending large amounts of money on video ads to only run them on a single network. The larger a network is, the more leverage it has to demand view-through attribution (usually with a one day window), which forms a vicious circle: the largest networks demand view-through attribution in their IO terms and become even larger by virtue of forcing developers to only use them.

View-through attribution is inexact, not friendly to advertisers, and is at fundamental odds with the mobile paradigm. Because of the install process, all mobile app usage is, in a sense, transactional: a hurdle must be cleared (install the app) before a curious onlooker becomes a user, and that hurdle is tied to immediate need and interest. Since most apps are free and the platform operators control distribution, brand equity has little relevance on mobile; advertising campaigns must be engineered to deliver network effects rather than brand recognition. This transactional, network effect-chasing model doesn’t benefit from views, it only benefits from clicks, and that is the dimension on which advertisers should be able to evaluate campaign performance. Indeed, in a study, TubeMogul found that video ad views had a negligible effect on product awareness:


But perhaps most worrying about the increasing pressure on advertisers to accept view-through attribution is that it underlines the fact that advertising networks are looking for ways to innovate on value extraction (pricing schemes) rather than value creation (product). This isn’t good for the app economy. What developers want isn’t to deal with unctuous salespeople or onerous attribution terms; they want clear paths to building engaged audiences.

As mobile advertising has evolved from banner ad units to static and video interstitial units to now more native formats, developers have used their wallets to voice their support and commitment to spending lots of money on advertising units that perform well. There’s still so much value to be unlocked in product innovation that burdensome attribution specifications seem unnecessary.