ATT is killing advertising performance. Six tactics for adapting to the new advertising landscape.

As of last week, over 70% of iOS devices have upgraded to a version of iOS that requires compliance with Apple’s App Tracking Transparency (ATT) privacy policy. Ascertaining the impact of ATT was difficult for most marketing teams early in the rollout, since Apple very slowly distributed it to users’ devices initially; this low adoption rate meant that a small proportion of any advertiser’s traffic mix was actually SKAN-enabled. The flat iOS 14.5 adoption curve muddied most advertisers’ metrics:

  • The CPMs for devices on iOS versions lower than 14.5 skyrocketed as advertisers fought to reach those users;
  • Android CPMs increased precipitously for the same reason;
  • CPMs for ATT-mandatory versions of iOS 14.5 and later decreased;
  • The widespread application of fingerprinting camouflaged the true percentage of installs that were SKAN-enabled and thus un-attributable, although fingerprinting only ameliorates a lack of install attribution and can’t replace the event stream that allows for platforms to optimize targeting.

The impact of ATT is much clearer now than it was two weeks ago, with the vast majority of devices having an ATT-mandatory variant of iOS installed. And the impact is meaningful. According to Tenjin, mobile ad spend on iOS has decreased by about 1/3rd since the beginning of June. Singular reports roughly the same, while also having observed a 50% decrease in 90-day LTV for acquired iOS users.

Perhaps more telling than these broad market surveys are two recent announcements from Facebook about how they are adapting to the new ATT environment. The first was that, in the coming weeks, Facebook will transition its Facebook Audience Network (FAN) ad platform to contextual targeting, versus the user-centric targeting it employed using Facebook’s behavioral data set. FAN is Facebook’s DSP: it facilitates advertising between third-party advertisers and publishers, and it allows advertisers to use Facebook’s own tools for targeting and segmentation.

Facebook announced last August that ATT might force the company to shut FAN down for iOS altogether because of the difficulty it would face in maintaining ad efficiency on behalf of advertisers without access to the IDFA. My understanding is that FAN was seen inside the company as something of a loss leader: the margins on DSP traffic are low, but in driving incremental conversions on behalf of advertisers, FAN improved the performance of those advertisers’ campaigns on Facebook’s owned inventory, which sees much higher margin. The fact that Facebook is altering the way it targets users on FAN, which puts it into direct, head-to-head competition with other ad networks that primarily focus on the gaming market and target primarily contextually seems to imply that it can’t risk any further impairment of its advertising business.

And the second of Facebook’s recent revelations that intimate at degraded iOS advertising performance in the post-ATT era was that a dearth of conversion values being delivered with SKAdNetwork postbacks has caused problems for its event-optimized AEO and VO campaigns. Facebook sent an email to advertisers recommending that they 1) consolidate AEO and VO campaigns, so as to concentrate the number of campaigns that are generating the conversion values needed to optimize to down-funnel goals, and that they 2) consider shifting budget into campaigns that optimize for installs. Facebook advised advertisers in this correspondence that they should aim to achieve 128 installs per day with any given AEO or VO campaign in order to receive enough conversion values to deliver acceptable performance.

These recommendations are significant: for most advertisers, the AEO and VO campaign strategies are the bedrock of Facebook ad spend. That Facebook would need advertisers to consolidate campaigns in order to drive sufficient captured conversion event volume, or even suggest to advertisers that they transition budget away from these important strategies, is a chilling admission that ATT is dampening advertising efficiency.

Of course, most advertisers already knew this: talk to anyone who spends significant money on Facebook advertising and they’ll tell you that the performance of their iOS campaigns has suffered materially in the weeks since iOS 14.6 adoption accelerated. It’s no secret that pain is being felt by advertisers across the spectrum, for both mobile app and mobile web advertising. The advertising landscape has changed, and advertisers must adapt to that change: the best time to begin planning for ATT was last June when it was announced, but as the unattributed adage goes, the second-best time is right now. It is worth acknowledging here that a great deal of volatility should be expected after any foundational change is made to an advertising optimization platform, and surely some of the degradation that is being seen currently can be attributed to dollars flowing out of the system as advertisers simply sit on the sidelines. But the core infrastructure of mobile advertising has been permanently altered, and advertisers must evolve to accommodate that. Below are six tactics for preserving advertising efficiency in the post-ATT environment.

Overhaul the creative production process

As I outline in Creative paralysis: ad creative production and testing in iOS 14 and in the above-linked Twitter thread, the capacity for creative testing is substantially constrained by the campaign ID limit and timer system imposed through SKAdNetwork. The creative production strategy that I describe in this article from a few years ago is simply non-viable in the ATT environment: creative can’t be mass-tested because enough “surface area” in terms of audience target to creative pairings simply doesn’t exist. The ability of ad platforms to pair finely-defined audience targets with creative allows ad campaigns to optimize at the lowest possible level of granularity. That ability has been stripped of ad platforms with ATT: because so few campaign IDs are available for use in targeting any given audience, only a small number of pairings of creative and audiences can be tested at any given time.

Advertisers will need to completely overhaul their creative production processes, focusing on smaller volumes of asset delivery with more implicit audience relevance. “Throwing a lot of mud at the wall to see what sticks” can’t work any longer: more research and non-advertising measurement data will need to be invested into creative production in order to captivate and motivate audiences that are mostly defined by demographic features and interests and not by recent purchasing behaviors. This is real audience development work. But a deep understanding of the intended audience, along with better insights around how ad presentation, narrative format (for video), etc. contribute to outcomes will be needed to empower a vastly reduced volume of ad creative to perform as well as when essentially everything could be tested.

My sense is that ad platforms will only be able to accommodate a tenth of the creative assets that high-budget advertisers were testing weekly prior to ATT. Production will need to transition into a “less, but better” operating model that incorporates more information into the production of fewer creative assets.

Use interest group targeting

This tactic is likely obvious to most advertisers, but it’s also a deceptively difficult and time-consuming exercise. Campaign strategies like AEO and VO on Facebook, tROAS on Google’s UAC, and other event-optimized tools on other ad platforms allowed advertisers to become “lazy” in terms of audience targeting: advertisers simply cast a wide net and the platforms used that freedom to build relevant audience sub-segments based on past behaviors and similarities to other users. Advertisers then cycled their best segments back into the algorithm through custom audiences and other types of audience lists and let the platforms build segment associations with user features that were almost totally unseen and even unknowable by the advertisers.

The idea of targeting broadly and scaling infinitely was always a myth, though. As I wrote in Broad targeting on Facebook took me to $500k / month in spend. Now what?:

This often becomes apparent at the level of $500k per month (or thereabouts) in spend on Facebook: delivery stalls and conversion costs rise because Facebook’s optimization algorithm has segmented the very broad audience definition to conversion events that are ultimately superficial for a high-scale business: first purchase or 1- or 7-day ROAS. It is at this point that the advertiser needs to actually t ake the reins and direct Facebook (and the other channels it works with) with respect to targeting in order to scale spend for particular audiences via audience-specific ROAS modeling.

The techniques I cite in that article are even more relevant and imperative today, when broad targeting supported by conversion-optimized audience construction simply can’t be done as efficiently and seamlessly as it previously was by ad platforms. Advertisers need to undertake the tedious, time-consuming work of triangulating their audiences on advertising platforms using the contextual cues and guideposts that they’ve mostly been able to ignore.

Bid on top-of-funnel conversions

Again, this might be an obvious reaction to ATT for most advertisers, but the effort involved is concealed in the implementation. Bidding on an up-funnel conversion event — an install, or registration, or even just click — requires a fundamentally different apparatus across predictive modeling, audience construction, creative direciton, etc. than does bidding against a purchase.

With ATT, the qualified intent of the users that platforms are able to target on behalf of advertisers is diluted: targeting purchases will necessarily become more expensive because the users being exposed to ads are less relevant. This change has already happened; advertisers can’t do anything about it. By moving the conversion objective up-funnel, an advertiser acknowledges the loss of precision and salience in terms of targeting and decides to pay for more of the conversions that can be reliably delivered as a result.

Revisit the product’s FTUE, user onboarding, landing pages

This tactic is more of a corollary to the previous tactic, but it bears explicitly acknowledging that all of the optimization effort that was invested into curating the user’s journey needs to be revisited when the scope and composition of traffic into the product has radically changed. Again, this change in the composition of traffic has already happened: ATT forced it, and advertisers must acknowledge that. What product leaders must do now is adapt their products to the broader, less-focused, less-intent-galvanized traffic they receive in the ATT environment such that the product creates conversions to an acceptable degree relative to marketing costs.

One problem with modern consumer technology product development is that A/B testing is used as a crutch to settle philosophical disputes around design. My antipathy towards A/B testing and the problems it incubates in consumer products are well-documented, but this particular issue — related to the expiration of A/B test results as the profiles of new users for products transform over time — is covered in Two fundamental problems with product A/B testing:

It is not rare for significant — sometimes extreme — differences to exist between the engagement and monetization profiles of organic users and those acquired via paid channels. If a user base changes to the extent depicted above, which isn’t uncommon, then the results from the A/B test conducted in February are obsolete by December. This isn’t to say that a team couldn’t conduct a new test every month or even every week to accommodate the above change, but in my experience, product teams very rarely think about the composition of DAU when planning tests and are loath to revisit already-tested mechanics.

The users being acquired into products in the post-ATT environment may look different from those acquired in the pre-ATT environment. They will react to different stimuli; they will likely have, on average, less of a proclivity toward making purchases; they may require more education and hand-holding through the onboarding; they might be less familiar with the product category, etc. The journey that guided users to success in the product previously might simply fail in this new environment, and it needs to be revisited if not wholly re-imagined.

Invest requisite resources into a conversion value model

This topic represents the bulk of the content in my course, iOS 14: How to prevail in Q2 2021, and I also cover it extensively in How to scale and optimize marketing spend with SKAdNetwork, but my sense is that it’s the single area that almost every advertiser has underinvested into, owing to a number of factors:

  1. A lack of clarity around when conversion values are included in SKAdNetwork postbacks (I discussed this in episode one of the recent ATT: One Month In podcast series);
  2. A lack of clarity around how advertising networks will relay SKAdNetwork postback data (this is mostly resolved through Apple’s decision to send postbacks directly to advertisers in iOS 15);
  3. Over what observation time period most ad networks will converge for capturing conversion values (Facebook’s recommendation / soft requirement is 24 hours, and this seems to be emerging as the standard);
  4. General uncertainty around how the various features of SKAdNetwork will be implemented and whether ATT would be rolled out at all.

Much of this uncertainty has now been alleviated. Conversion values (and the on-site events that are captured with web campaigns) are the only campaign-level signal that advertisers can avail themselves of in assessing the value of traffic. Big projects like media mix models and incrementality testing are important and are certainly becoming more popular as the aperture of marketing measurement expands, but conversion value modeling is table stakes.

Given the more limited amount of data available in the ATT environment across all other dimensions, it’s irrational for any advertiser to not have built a conversion value monetization model, even if it is only used as one input for the broader measurement program.

Get (much) better at modeling from restricted samples

Understandably, most advertisers aren’t good at building projections from small samples of behavior, because they have never really needed to be: every user’s origin could be attributed, and groups of users could be constructed of sufficient size that analysis was straightforward and painless.

But this is changing, and the ability to draw conclusions and make decisions from limited data sets, and to be able to know when a sample can provide proper guidance for a wider cross-section of the user base, is becoming necessary. While ATT opt-in rates are mostly irrelevant with respect to managing marketing measurement, opted-in users can provide helpful guidance around user behaviors and monetization patterns, and this insight is useful in optimizing advertising campaigns.

Attribution on mobile has never been truly deterministic. Advertisers must recognize that the new environment in which they now operate means that some marketing activities can be measured directly, but most can’t, and in actuality, things have looked like this — or at least, the environment has been lurching in this direction — for a long time.

The difference is that now, advertisers must now master the probabilistic techniques that were erstwhile used by ad platforms on their behalf. For every advertiser, some ensemble model exists that takes into account opted-in user data, SKAdNetwork postbacks, and overall spend and revenue data that capably guides profitable ad spend. Ad platforms can model conversions and do some basic forms of fingerprinting, but advertisers shouldn’t accept that those activities are serving their best interests. Internalizing these capabilities is a critical task for advertisers in the post-ATT era.

Photo by Luke Jernejcic on Unsplash