Modeling video ad effectiveness on mobile

Posted on December 12, 2016 by Eric


In September, it was revealed that Facebook had been supplying video advertisers with inflated metrics around average advertisement view time: it had counted only views that lasted at least three seconds, but in calculating average view time, it didn't divide that number by the total number of people that saw any of the video (in other words, it was supplying average view time for people that had spent at least three seconds watching a video).

This mistake sounds egregious, but it's also probably insignificant: average ad view time is a meaningless vanity metric that doesn't, on its own, explain whether an ad campaign is economical or not. The same goes for number of video views, or cost per video view, or even, when it can be calculated, cost per app install from a video ad. Without context, these metrics don't provide any insight into ad performance.


(From a study by VentureBeat and on television ad spending by mobile gaming companies, September 2016)

In 2013 I argued that brand marketing and performance marketing can both serve mobile marketers, but that brand marketing is most valuable as a complement to the former that serves as a force multiplier and / or a bridge to unreachable audiences. The various strategies that have played out in television advertising for mobile developers -- such as Supercell's Superbowl ad starring Liam Neeson -- aren't generally, at least in an obvious way, designed to expand a "brand": they are very deliberate and, ostensibly, constructed to produce installs.

The CEO of Machine Zone made this point at the Re/Code conference in February of this year: television ads are generally commercially opaque in a way that is very unfriendly to advertisers, but they can be useful in increasing the performance of direct response mobile ads that are much more transparently measurable. That is, if a mass-market, high-budget television ad creates a level of awareness that improves the performance of direct-response ads to the extent that the television spend is recouped with positive ROI, then it is a worthy expenditure under those conditions. This is the performance marketing approach to spending money on non-direct response video ads: building a framework for quantifying the impact of spend, and not simply spending money apropos of nothing with the vague justification of building a "brand".

Video was the big story in mobile advertising in 2016, and it'll probably continue to drive increasing advertising spend well into the foreseeable future. Facebook especially has benefited from the rise of the video ad format on mobile, but as more mobile developers take their video ads to television, the distinction between brand and performance marketing is being purposefully muddied by advertising networks and advertising agencies to try to convince advertisers that all video spend is good spend.

It's not. A model of video ad effectiveness must tie the spend on a campaign to performance (ie. income generated by that campaign that wouldn't have otherwise materialized), not vanity metrics. Ad campaign performance is a function of return on investment: building the links between spend (especially television spend) and user monetization is incumbent on the advertiser (eg. did the campaign boost direct response ad performance? Did the campaign generate installs that could be tracked by a time series analysis of installs during and after the spot being aired? Did the campaign increase general virality above a pre-defined baseline?), but those links don't consist of metrics like average view time, which is more trivia than a performance metric.