A recent theme that has seemed to reach critical mass within the milieu of the mobile ecosystem is that performance advertising on mobile is on an extinction course, with the meteor impact event being Apple’s presumed inevitable deprecation of its proprietary advertising identifier.
It’s certainly true that the deprecation of mobile advertising IDs would precipitate fundamental change within the mobile ecosystem — and do I believe that deprecation will absolutely happen, although COVID may have granted mobile advertising IDs a stay of execution. But mobile performance marketing will not disappear with the eradication of advertising IDs — its shape will merely change, but not as substantially as some might believe.
The truth is that mobile advertising identifiers are really just a crutch that many performance marketing teams lean on to feel confident in their work. Limit Ad Tracking, which sets a device’s advertising ID to a sequence of zeroes and was introduced by Apple in 2016, already renders roughly one third of iOS users totally anonymous: un-targetable and un-trackable (Google introduced a similar mechanic in the same year).
And perhaps even more confounding but less frequently addressed is the fact that every advertising platform effectively lords over the same device graph — the only difference between advertising platforms is their bank of historical monetization data per device. Advertisers think they are diversifying their traffic intake by operating campaigns across multiple channels, but in reality they’re just reaching the same users in the same apps via different inventory. Running multiple campaigns across different networks is akin to setting many fishing poles at various points on a river: if you don’t catch a fish at one point, you may still catch it downstream.
The mechanic that allows advertisers to believe they’re navigating this system efficiently is last-click attribution: the fishing pole that yielded the fish was clearly better suited to catching that particular class of fish! But in reality, Facebook and Google are more or less randomly rewarded with the vast majority of fish upstream.
Last-click attribution provides advertisers with a veneer of control, like a security blanket: budget is spent and attributed and everyone feels confident that they are systematically driving profitable revenue through performance user acquisition. But this measurability is really an illusion: some percentage of spend obviously produces incremental revenue, but the way that users swirl around within the lines of sight of various ad platforms means that, once an advertiser extends their spend beyond a single channels, they are guaranteed to be losing money to redundant, superfluous spend.
So the deprecation of mobile advertising IDs will really just accelerate a trend that has existed for the past few years: despite clear attribution and unit economic metrics, advertisers can’t trust that their spend on any given channel is totally incremental, and thus their measurement is really only valid at the broadest level of granularity. This macro-level framework is how many traditional companies evaluate their spend across advertising media, like television and radio, that is difficult to attribute.
The mobile ecosystem — with Limit Ad Tracking; the consolidation of inventory across just a handful of large advertising platforms; the wholesale shift to event-driven, algorithmically optimized campaigns by Facebook and Google; and the maturation of mobile and attendant decrease in smartphone sales — has been drifting toward such an environment for many years. Back in 2017, in Mobile’s post-attribution era, I wrote:
And this is where the concept of attribution on mobile begins to whither as marketers diversify away from direct response. TV, influencer, out of home, et al can all be integrated into performance marketing machinery with an eye toward profit at the unit level. But this is difficult; it requires building a top-down model of a company’s marketing schema that incorporates 1) uncertainty and 2) statistical robustness. And these channels aren’t attributable: the outcome of these campaigns, while capable of being evaluated broadly, can’t be specifically measured at the level of the individual user.
The notion that measurability is exclusively a function of attributable clicks has been evaporating for years, hastening for the aforementioned reasons, and yet user acquisition teams have persisted. The demise of advertising IDs won’t precipitate the demise of user acquisition on mobile, since the deprecation of advertising IDs will simply take a trend that already exists — of decreased reliance on click attribution in a shift toward more holistic, macro-level measurement — to its logical conclusion.
The structure and size of user acquisition teams has fundamentally changed as this trend has played out (see: The future of mobile growth teams, Are mobile marketing teams shrinking?), but user acquisition as a function is no less important than it was in previous stages of the history of the app economy — in fact, mobile ad spend is on a tremendous growth trajectory that isn’t poised to abate any time soon. User acquisition is therefore as critical to success on mobile as it has ever been — the question then is what form that function takes as measurement shifts.
Media Mix Modeling
Media Mix Modeling is an advertising measurement methodology that attempts to quantify the incremental business impact of spend on any given channel within the context of a multi-channel advertising environment. Media mix modeling is often referred to as a “top down” approach to measurement (versus “bottoms up,” click-based attribution) because it evaluates advertising performance through the correlation between broad inputs: media spend per channel and some conversion metric (usually sales). A media mix model often also takes time-based, economic, and exogenous factors into account, such as competitor spend around a new launch or seasonality.
Media mix models have been in vogue with large CPG advertisers for a few years: this Harvard Business Review piece from 2013 describes the media-mix-model-centric measurement system as “Analytics 2.0.”
This paper, titled Challenges And Opportunities In Media Mix Modeling and written by two data scientists at Google, does a very good job of explaining the underlying statistical methods used in media mix models, along with some of the challenges in constructing them.
One such challenge for mobile marketers operating across direct response channels is the inherent selection bias of utilizing ad spend data from targeted campaigns: the underlying demand is not captured in the model but influences both ad spend and conversions (this is known as a confounding variable). This issue is perhaps more acute with channels like Facebook and Google UAC, where targeting is largely controlled by the platform and optimized automatically over time. A detailed review of the mechanics of media mix models is beyond the scope of this article, but this white paper (with code) and this master’s thesis are very clear and helpful resources for learning about media mix modeling. This much more technical paper from Google introduces Bayesian techniques for capturing lagged response and diminishing returns in a media mix model.
A media mix model solution is writ large more relevant for advertisers that use a broad mix of offline and online marketing channels, such as a combination of direct mail, television, radio, paid search, etc. Most app advertisers and, more broadly, mobile advertisers allocate the majority of their budget to direct response channels, which have been seen as explicitly, precisely attributable. It’s clear that this is no longer the case: not only because of the impending removal of mobile advertising IDs, but for the reasons outlined earlier in the article. Direct response advertising is very much approaching a measurability environment like those of television, radio, out of home, etc.: bottoms-up measurement is an anachronistic performance assessment methodology that can’t effectively guide advertising incrementality-minded spend decisions in the contemporary environment. The veneer of credibility of bottoms-up, attribution-centric measurement has already cracked; it will completely disintegrate when the mobile advertising IDs vanish.
A quick note here about ad fraud. Ad fraud on mobile is a mostly exaggerated problem, the impact of which is overstated by mobile attribution companies to allow them to diversify away from their core business, which was always on a collision path with commodification. For the vast majority of mobile advertisers, which allocate their budgets primarily to the Duopoly and a handful of large, credible advertising platforms, ad fraud amounts to a trivial amount of wasted spend — certainly not enough to justify the use of expensive fraud analytics tools.
Most cases of ad fraud are really just cases of lack of incrementality: an advertiser is spending money on multiple channels to chase one core group of people, but because of last-click attribution and the disproportionate purview that self-attributing networks have, the advertiser can’t reconcile spend with acquisitions or make the case for incremental revenue through its ad spend. Marketing managers frame this incrementality problem as a fraud problem because it allows them to blame their lack of effectiveness on a nefarious bogeyman. It is telling that the most high-profile case of mobile ad fraud is Uber’s, which is more of an example of fraud powered by social engineering and manipulation than it is an example of systemic fraud powered by malicious technology.
The robots are coming
Marketing automation is often cited as one of the primary reasons that mobile marketing teams are shrinking: algorithmically-optimized ad platforms like Facebook’s and Google’s UAC can be scaled without much manual effort, and the large, matrixed marketing organizations of the 2014-2016 era are simply outdated. This was apparent in many of the high-profile COVID-related layoffs that took place in April and May at consumer technology companies: marketing teams were disproportionately targeted with the layoffs at Airbnb, Uber, and Lyft.
It is true that marketing automation is displacing media buyers: much of the work of campaign management has been ingested by the ad platforms, and the media buying and analysis work that remains beyond that is being addressed by tools from ad tech companies.
But it’s a mistake to confuse media buying with marketing. Modern marketing is incredibly technical and data-driven: all of the infrastructure and analytical tools that support media spend are only becoming more core to the broader marketing function for large advertisers. Yes, marketing teams are shedding media buyers as platforms subsume campaign management responsibility, but merely automating away the work of optimizing campaigns doesn’t alleviate the need for incrementality measurement.
In fact, incrementality measurement arguably becomes more important as automation proliferates because automated systems make it easier to onboard and operate new channels and media sources. So as media buying teams shrink, more and more resources are being allocated to the tools and infrastructure that power and measure the automated marketing systems. So long as digital products exist, companies will want to grow them. User acquisition as a function is not going to “die”; the composition of user acquisition teams is changing as the need for media buyers diminishes and the need for analysts, engineers, and data scientists increases.
Media mix modeling sits at the intersection of these trends: click-based mobile advertising attribution is already insufficient for measurement, and it very well may be rendered completely obsolete with the deprecation of mobile advertising identifiers. Media mix modeling allows advertisers to map spend to revenue on the basis of incremental value and diminishing returns: it sidesteps completely the complications caused by last-click attribution, limit ad tracking, overlap of audience, etc.
The mobile marketing team of the future exists to support a media mix model, which sits at the center of a Level 4-automated platform that directs spend across channels via predicted incremental contribution. This user acquisition team, staffed primarily by engineers and data scientists, looks very different from a contemporary team, but the functional focus and output is the same.