In 2016, Apple introduced a hidden privacy feature that became colloquially known within the mobile marketing ecosystem as “IDFA Zeroing.” Prior to IDFA Zeroing, when an iOS user had activated a device setting called Limit Ad Tracking (LAT), advertising technology vendors could access that device’s unique advertising identifier, the IDFA, but they were expected to not utilize it for the purposes of advertising targeting. With the introduction of IDFA Zeroing, when a user activated LAT, their device’s IDFA was replaced with a string of zeroes, rendering it useless.
The IDFA is a unique device identifier on iOS that is designed to be used for advertising purposes; the IDFA was introduced in 2012 as a replacement for the UDID, or Universal Device Identifier, which was deprecated in 2013. Detail about how the IDFA is used for advertising measurement and targeting can be found here. IDFA Zeroing complemented Limit Ad Tracking by giving it teeth: prior to the introduction of IDFA Zeroing, developers were asked to attest upon app submission that their app, as well as any third-party SDKs integrated therein, would honor a user’s Limit Ad Tracking setting, but enforcement was challenging and irregular. With IDFA Zeroing, Apple didn’t need to depend on willful compliance from developers: Apple retrofitted the existing LAT device with IDFA Zeroing setting to simply restrict access to the IDFA.
In IDFA Zeroing is the massive change to mobile advertising that no one is talking about, published shortly after IDFA Zeroing was rolled out, I wrote:
In the longer term, [IDFA Zeroing] could precipitate a shift in budgets to non-attributable inventory, at least for the bigger spenders: television, out-of-home, etc. Attribution may be the sine qua non of mobile advertising, but it’s not essential for advertising apps: LAT iPhone owners are still iPhone owners (and they’re potentially the most valuable variety, being savvy enough to know that the limit ad tracking setting exists in the first place).
A steady drumbeat of signal reduction
The ability to observe conversions on iOS devices was not revoked suddenly and unexpectedly, in one dramatic policy change, through ATT. The chronology of feature releases related to the IDFA is instructive:
- 2012: Apple introduces the IDFA and the Limit Ad Tracking setting in iOS 6, after announcing the planned deprecation of the UDID in 2011 (the UDID was ultimately deprecated in 2013);
- 2016: Apple introduces IDFA Zeroing, which revokes access to the IDFA for all apps on a user’s phone when the device-level Limit Ad Tracking setting was activated;
- 2021: Apple rolls out the App Tracking Transparency consent dialogue and begins enforcing the restrictions mandated by the App Tracking Transparency policy. The ATT consent pop-up restricts access to the IDFA on an app-by-app basis, depending on the user’s choice.
And over this timeline, Apple introduced many other privacy features that aren’t connected to device identifiers: Hide My Email, Private Relay, Intelligent Tracking Prevention for Safari, Mail Privacy Protection, etc. And Apple isn’t the only force fomenting change related to privacy: Google announced in 2022 that it would deprecate third-party cookies in the Chrome Browser (although that timeline has been shifted to 2024), the GDPR went into effect in the EU in 2018, and various restrictions related to consumer privacy will be implemented into law soon in the EU through the Digital Markets Act and the Digital Services Act.
So if the notion that deterministic attribution for direct response advertising was jeopardized with the introduction of IDFA Zeroing, it’s even less viable in the wake of ATT, with multiple other restrictions poised to materialize soon. As I wrote in 2017 in Mobile’s post-attribution era:
And this is where the concept of attribution on mobile begins to whither as marketers diversify away from direct response…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. Modeling this is no trivial task: it requires comfort with uncertainty and the ability to use variability as a source of improvement.
If marketers and media buying teams are left with disparate, disjointed sets of advertising data, with the identity ligaments that once unified them broken, how do they adapt their decision-making and budgeting processes? And what elements of their now antiquated and obsolete functional tools and processes must be rebuilt for this new, privacy-conscious operating environment?
The marketing economist
I’ve written previously about the value and operational management of media mix models in the new privacy environment. From Media mix models are the future of mobile advertising:
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.
My point in the excerpt above: even putting aside conversion signal loss — that post was written in 2020, before ATT went into effect — marketers should have sought the directional guidance of probabilistic measurement solutions like media mix models because the clarity and certainty assumed from last-click attribution was dramatically over-stated. And again: anything that was true in 2020 likely represents an emergency now.
The use of any econometric, or probabilistic, model to map inputs (marketing spend) to outputs (conversions) requires something of a dual workflow: determining the commercial effects produced through different channels to allocate budget, and then actually managing campaign performance at the level of an individual channel to ensure that budget allocated there is used most efficiently. These workflows don’t operate on the same cadence: cross-channel econometric models are generally updated monthly or quarterly in order to smooth effects and capture as much meaningful data as possible, whereas campaign optimizations — for instance, creative adoption and budget allocation decisions across campaigns within a channel — tend to happen at least once per week, if not more.
Abstracting away the process of reconciling ad spend with commercial outcomes as “macroeconomic modeling,” whether through media mix modeling, incrementality measurement, or some other methodology, hides the complexity of those systems. Firstly, this approach differs fundamentally from the simple counting and database joining necessary with last-click, identifier-based attribution. Econometric marketing measurement models are difficult to build and challenging to maintain and interpret.
But secondly, and more importantly, these models cannot be connected to an existing marketing workflow or plugged into existing reporting infrastructure. Econometric marketing measurement is wholly distinct from deterministic measurement: it’s not captured in the difference in tools alone, but rather how those tools are used and what kinds of insights those tools produce. A mistake I see teams make in transitioning to econometric marketing measurement is thinking that the change is akin to plugging a new monitor into an existing workstation: the visuals will remain the same, they’ll just be rendered by new machinery.
This isn’t the case: probabilistic measurement models shouldn’t be used to populate an existing advertising dashboard or campaign performance report with bid suggestions or “Red, Yellow, Green” indicators of campaign health. The econometric approach to marketing measurement represents a total change of approach: the econometric model itself draws primarily from three data sets, but it can’t produce a pointed, prescriptive set of action items with total certainty (as deterministic, last-click attribution purports to do). The output of probabilistic measurement models requires interpretation: weaving a plausible story of how marketing spend generated conversions through an analytical framework. This is the job of a Marketing Economist.
The Marketing Economist’s responsibilities are to:
- Curate, clean, and normalize the three data sets identified above: Market Data, Product Data, and Channel Data;
- Operate and maintain the econometric / probabilistic model that attempts to reconcile these data sets;
- Provide analytically sound direction to the media buying team around budget allocation and channel-level targets.
These responsibilities might seem functionally equivalent to those of a marketing analyst or data scientist, which are familiar and commonplace roles. The difference is in mandate: unlike the marketing data scientist, the marketing economist is charged with interpreting data to tell a credible narrative. Data scientists and analysts seek clarity through systematic, quantitative computation; the marketing economist embraces uncertainty and attempts to model the interaction between complex systems with robust, analytical assessment as well as through deductive reasoning.
Only so much deterministic data is available as input to an analysis of marketing performance, and that pool of data is vaporizing over time. The marketing economist fills the deterministic gap with a model that attempts to describe the intersection of this loose and disconnected data.