Much ink has been spilled litigating the particulars of iOS 14 and its anticipated disruption of mobile advertising. Most commentary has adopted a monolithic interpretation of the issue along with a bimodal expectation: at stake is the task of “attribution”, and either it persists in a weakened fashion under  AppTrackingTransparency framework or migrates to Apple’s SKAdNetwork API. With some noteworthy exceptions, analysis has largely landed on the latter understanding: opt-in consent will diminish the device ID’s utility beyond repair and attribution becomes a platform-level operation, with outcomes exposed to participants on an aggregated “need-to-know” basis. It’s Apple’s ocean we’re swimming in after all, and thats where the tide is headed.

To be clear, this is a compelling hypothesis. And those arguing in the other direction are grasping at murky hypotheticals. But both arguments mount their case on a leaky abstraction and in turn overlook critical details and a bigger picture. Attribution, after all, is an umbrella concept that represents a collection of distinct utility functions. To date, these functions have been conveniently bundled together in a generally available platform ID. But neither SKAdNetwork or AppTrackingTransparency are capable of subsuming them all together, and the fate each faces will be unique. As such, it’s not that iOS 14 will force the ecosystem to align on some new measurement framework. On the contrary, it represents an unbundling of attribution and a fragmentation of standards generally. To make a clear headed assessment, we need to unpack the concept itself and address its individual components.

What Device IDs Do

To borrow Clayton Christensen’s phrase, we currently hire device IDs to help us do specific jobs. The mobile ad ecosystem will be impacted to the extent that it changes how, and how well, those jobs get done. If you take the time to look, you will find the IDFA at work in three separate roles:

  1. Advertisers and publishers hire device IDs to mediate transactions as a neutral unit of account. The IDFA and GAID supply a means to verifiably correlate in-app behaviors with upstream ad click and view data, and in doing so make app install and conversion events “biddable” entities.
  2. Advertisers hire device IDs to help estimate return on ad spend. With a device ID, advertisers can attach channel source parameters to individual users. This lets them aggregate cohort retention and monetization behavior down to the ad level and dynamically model lifetime value.
  3. Publishers hire device IDs to help match “off-app” behaviors with specific user profiles on their platform. This effort serves two ends: it allows publishers to show ads to users who have already performed some behavior (e.g. retargeting) and to show ads to users they estimate are likely to perform some behavior in the future.

Predictions about what happens after iOS 14 depend on a number of variables and any specific expectation comes with a degree of contingency. But if we evaluate from the perspective of each job individually and what’s needed for it to be done, we can draw clear boundaries.

Job 1: Unit of Account

To serve as a unit of account, in the sense of providing a ledger of outcomes to price advertising transactions, any suitable candidate has to fulfill three criteria: it must have broad coverage, be available to each party in the transaction, and unbiased in it’s representation. And even in a best case scenario for adoption, IDFA under the AppTrackingTransparency framework will become unsuitable on all three counts.

Not only might coverage fall to as low as 10 to 20%, adoption rates and consent mechanics will vary by publisher and advertiser and, as such, the IDFA will fail to provide a consistent representation of outcomes. IDFA can act as a unit of account only under conditions of general availability. To be clear, it was on shaky ground in this regard already. Since Apple’s release of Limit Ad Tracking in 2016, the percentage of iOS users with a disabled IDFA has climbed to as high as thirty percent on some estimates. This has affected not just coverage and availability, but neutrality as well: users who manually opt out of ad tracking share behavioral and demographic traits, and their absence creates an inherent sample bias in any IDFA-based data set.

In this regard, SKAdNetwork presents a compelling and potentially superior alternative. As the device manufacturer and operating system provider, Apple is theoretically positioned to provide a complete system of record for cross-app advertising. They have practically universal visibility and can conduct attribution workflows without exposing user-level identities to any participant.

Of course, the operative word here is “theoretically”. Not only are the details ambiguous, the functionality as it’s currently described leaves advertisers with a very restricted set of capabilities: no view through measurement, no control over lookback windows, and no visibility into claim logic or matching methodology in general. It begs the question whether Apple actually yet understands what attribution is or what it entails. Nevertheless, given the paucity of alternatives, SKAdNetwork will be the only game in town.

Job 2: Value Measurement

The second job device IDs do is allow advertisers to attribute value to the events identified by the first job. For ROI based advertising, it’s not enough to have a record of what user actions followed from which ad interactions. In order to bid profitably for a specific action advertisers understand the monetary value that it represents, and that value can vary substantially across different channels and campaigns.  In this regard, IDFA pre-iOS 14 is perhaps as good as it will ever get.  

That said, unlike the first job, the ATT framework does not necessarily neutralize the IDFA as a tool for value measurement. Even at opt-in rates as low as 10-20%, the IDFA retains signal value. The magnitude of that value will be a function of the available sample size (itself a product of scale and opt-in rate) and the advertisers ability to probabilistically model population variables from it. To be clear, this is a categorical shift from it’s status prior to iOS 14, where it’s value was deterministic, easily extractable, and largely definitive. Under ATT, the IDFA becomes merely one signal in a broader exercise of triangulation.  

What of SKAdNetwork? It too represents a tool rather than a new standard. It’s value is the inverse of the IDFAs under AppTrackingTransparency: it promises theoretically complete coverage at the top of the funnel, but limited downstream visibility. Under current functionality, advertisers must extrapolate LTV from a combination of the 6-bit conversion value postback and whatever source parameters might be contained in the campaign ID (e.g. geo, gender, etc). This former is limited to a discrete event that needs to occur within a short window post-install. For apps with simple monetization mechanics, this could be good enough at low scale. But even subscription services with a 7 day free trial will face debilitating limitations. Freemium apps with a highly skewed longtail such as games are in an even worse spot.

Value attribution, then, is the first job to unbundle. What replaces a generally available IDFA is not singular or even straightforward. Advertisers will be left to federate a fragmented set of signals into a coherent understanding of ROI, and that exercise will be a messy and probabilistic one. It’s important to note though that, as Eric Seufert points out, deterministic attribution is somewhat of an illusion anyway (albeit an extremely useful one). After all, there is a fourth job of attribution that device IDs don’t do, and that is tell the advertiser what outcomes still would have occurred without any advertising. iOS 14 will bring that reality into a harsher light and force a renewed focus on incrementality.

Job 3: Identity Matching

So far, the debate about iOS 14 has largely focused on the impact to advertiser capabilities. This is somewhat expected: most commentary has come from participants whose business is selling software or services to advertisers, so it’s natural that they would focus first on their customer. But the workhorse driving direct response advertising performance over the last decade has nothing to do with advertiser measurement workflows. That distinction belongs to ML-driven targeting algorithms, and their sophistication has been a driving force in the explosion of long-tail internet businesses such as gaming and DTC ecommerce. And the ability to make user-level, cross-app identity correlations is critical to functionality: losing that would entail a reduction in advertising efficacy many times greater than any loss of measurement fidelity.

Facebook provides the canonical example here, and the loss / benefit calculus that they face with iOS 14 will be a fascinating one to watch. A score of user-level conversion probability is the heartbeat of the FB ad auction and drives performance across all of their downstream event optimization products. And even with a 3 billion user identity graph and an SDK and pixel footprint collecting event data from nearly every consumer internet property, they still rely on a third party identifier to close the loop on identity. On web, Facebook has a convenient workaround to the impending loss of 3rd party cookies in the FB Click ID, which lets advertisers use their own first party cookie to store and pass back a unique identifier. No such route on mobile is obvious.

So where does this leave Facebook and all other publishers relying on algorithmic targeting? Off the bat, it’s clear that SKAdNetwork offers little. While aggregate counts provide an optimization signal of sorts, the modeling that can be done with them is categorically different than what’s possible at a user-level and will not serve as a replacement. Not to mention that SKAdNetwork is unusable for any sort of custom audience targeting.

But as with ROI measurement, the IDFA under AppTrackingTransparency retains signal value in theory. And the calculus that quantifies that value is the same: it’s a function of sample size and extrapolation capability. Likewise, the device ID here becomes just one signal among many. The wildcard is of course Adjust’s proposal for an “attribution hash”. This is a clever solution that appears to comply nominally with guidelines, but assumes functionality that would require explicit modification from Apple to accommodate. It also conflicts with the spirit of Apple’s intentions in such a stark way that it seems at this point to be wishful thinking.

Presuming that channels are not able to obtain IDFA through single sided consent, the job of identity attribution is unbundled as well. Many have suggested that what replaces it is nothing, essentially: ad targeting will return to it’s pre-ML days and become a broadly targeted, context-focused endeavor. This is a naive assertion though, and one that, if true, would neuter the most efficient discovery mechanism for internet businesses in history. There is simply too much money to be made with precision ad targeting, and both advertisers and publishers are incented to use everything that’s left at their disposal to unlock it.

What likely replaces the pre-iOS 14 IDFA instead is a patchwork of partially available device-level data and explicitly captured first party identifiers. Email will perhaps be the biggest component, but in this world redundancy is king: we’ll see increased rigor across the board in first party data collection and identity management on behalf of advertisers and publishers alike. Facebook’s app event optimization products will end up incorporating functionality from it’s offline conversions API used by lead gen advertisers to match based on a collection of different IDs. Other SANs with scaled identity graphs will follow suit, although from a certainly weaker position.

What happens to SDK networks and programmatic–and by extension, all of the individual publishers that rely on them for revenue–is less clear. Both channels are less reliant on user-level targeting algorithms, but also less capable of doing anything useful with non-device level data. What is clear is that, like GDPR, iOS 14 advances consumer privacy in a way that consolidates power and reach to incumbent platforms.