The vanishing marketing data lake

Two weeks ago, Facebook communicated to advertisers via email that it would end its Advanced Mobile Measurement (AMM) program, which allows the companies that participate in its Mobile Measurement Partner (MMP) program to share user-level attribution data with their advertising clients. In April 2020, Facebook altered the terms of its AMM program to prevent MMPs from sharing user-level data for view-through conversions with advertisers. At the end of October, the program will be shuttered completely, meaning MMPs will only be able to share aggregate attribution counts with their advertising clients for both view-through and click-through attribution across iOS and Android.

Three things should be considered when contextualizing this news. The first is that MMPs will continue to receive user-level attribution data for both view-througn and click-through attributions: what changes is that MMPs will no longer be able to then relay that user-level data to advertisers, having instead to provide only aggregated counts. I was incorrect about this point in a tweet that I published with some commentary on the change. The data to which MMPs have access will not change and so the cross-channel attribution that MMPs conduct will not, either.

The second consideration with this change is that Facebook, being what is known as a “Self Attributing Network” (SAN), manages attribution for the traffic it delivers to advertisers and invoices on the basis of its own reporting. I spoke about the dynamics of SANs and how they undermined the notion of deterministic attribution in this podcast from July 2020: the fact that SANs can claim (and demand payment for) attributions independently of MMP analytics created a two-tiered reporting system. Within this system, advertisers could be charged multiple times for the same attributions because they paid Facebook, Google, Snap etc. on the basis of what those companies reported and paid everyone else on the basis of what their MMPs reported. An MMP can only really be an objective “referee” for the players in the game that recognize it as an authority: everyone else (the SANs) simply invoices advertisers for the conversions they claim to have generated.

And the third consideration with this change is how user-level attribution data was used by advertisers: to build campaign-level monetization models using internal revenue data in order to price bids (I use the word “campaign” here broadly to refer to any hierarchical metrics group on Facebook, ie. Ad Set or Ad). This becomes impossible with the AMM program being retired because attributions will be passed from the MMP to the advertiser as broad, aggregate counts and not user-level attributions. Within its own data environment, the advertiser won’t be able to associate user revenue to the campaign of its provenance.

This is a substantial change to the way that advertisers measure campaign performance and build bid models. And it further amplifies advertisers’ dependency on Facebook and Facebook’s version of the “truth.” In this new dynamic, MMPs will report discrepancies between the attributions claimed by Facebook and those measured from other channels, but without user-level data available to audit those discrepancies (eg. “which user did Facebook claim that should be attributed to Channel X?”), an advertiser can’t adjust its bid model for the true source of the users that Facebook double-claimed.

In other words: Facebook and other SANs had an unfair advantage in claiming attributions previously, but at least advertisers could use their own data (coupled with MMP attributions) to credit revenue to deserving channels. Now, without user-level attributions for Facebook, advertisers can’t do that: advertisers won’t know which users Facebook is taking credit for providing.

Advertisers should expect their marketing data lakes to continue evaporating in this way in the wake of ATT and ahead of other, forthcoming privacy policies. Every ad platform, but especially each of the walled gardens, is incentivized to silo its data as severely as possible so as to limit its interoperability with competitors and fully exploit the advantage of its size.

As an advertiser thought experiment: can you maintain the viability of your business if you stop working with Facebook? Is that true for any other ad channel? If the answer to both of these questions is “No,” then Facebook is advantaged by reducing the granularity of performance data that it shares with you.

The marketing data lakes are vanishing: objective, “communal” data isn’t a reality in the ATT environment. Trustworthy measurement is only really possible when an ad platform serves content in a first-party setting — in other words, in a content fortress. The largest ad platforms always begrudgingly contributed their data to data lakes, and now they are even further incentivized to sequester it.