iOS 14 Update: Facebook’s Data Challenges Show the Need for Alternative Measurement in the Privacy Era

With Apple’s announcement of iOS 14.5 and App Tracking Transparency (ATT), we knew there would likely be big changes ahead for advertisers on Facebook, much more so than other major paid acquisition channels. Those assumptions were all but confirmed when Facebook announced the sunsetting of longer attribution windows at the beginning of the year. More recently, those assumptions turned into reality with a series of changes to data access and analysis tools after their tandem release in April.

Those changes included:

• Reducing performance tracking time from 28 days post-ad click or view down to a max of 7 days post-click
• Introducing modeling for a small portion of conversion data
• Limiting the number of actions you can optimize for
• Disclosing that view-through performance would be harder to collect under ATT
• Removing performance by demographic segmentation

That’s a lot of change for any platform, but especially for Facebook, given how performance attribution has historically worked for most of its advertisers. Now, data is starting to emerge that shows just how much conversion attribution is changing on Facebook. That changing data access begs real questions about how marketers will account for it, and suggests the urgency with which they need to be exploring new measurement strategies now in order to continue making successful budgeting decisions in the future.

Facebook Ads Attribution Post-iOS 14.5

First, an important caveat to this: Facebook Ads having a tough time attributing conversions is not the same thing as it having a tough time generating conversions. Marketers need to be accounting for this when evaluating performance and making budgeting decisions.

Now, let’s look at how the reported data is changing. This is aggregated across our agency’s full set of Facebook advertisers, which spans all manner of conversion goals, industries, and audiences.

Aggregate data across Metric Theory clients on the Facebook Ads platform, inclusive of all Facebook placements (e.g. Instagram, etc.).

This breakdown covers three important periods: All of 2021 prior to the release of iOS 14.5 and App Tracking Transparency, then the roughly 6 week period where global iOS 14.5 adoption increased to about 20%, and finally the period after Apple chose to force update phones to iOS 14.6, after which adoption surged to around 60% in just over a week. Adoption of iOS versions enforcing ATT currently sits over 75%. All of those data points are attributed to the excellent work being done by Alex Bauer and Branch Metrics.

What we see is the cost of ads growing before falling back slightly, while the rate of attributed conversions drops 63% from the first tracked period to the last. This happens despite Facebook modeling conversions to attempt to account for some of the lost iOS insight. Counterintuitively, as ad performance shows declines, ad cost actually rises and stays relatively stable.

So what might be happening? Likely, Facebook ad cost at the beginning of the year was suppressed by externalities that reversed in Q2, like the typically expected Q1 slowdown for ecommerce coming off the holidays, reduced advertising around the US Presidential inauguration, and the impact of COVID-19 on the economy prior to vaccines becoming more available in the US. But the fact that ad costs stayed higher despite conversion rate declines suggests more advertisers were accounting for an attribution loss than one might expect.

Compare that data to the same metrics on LinkedIn Ads, which while catering to a different audience than Facebook, is an operationally similar network.

Aggregate data across Metric Theory clients on LinkedIn Ads, inclusive of all placements.

What we see with LinkedIn is no material change to ad cost, with a huge difference in its ability to maintain performance attribution compared to Facebook. To that difference, we offer the following explanations:

• More of LinkedIn’s conversion total happens in the first day after an ad is shown (True)
• LinkedIn is less reliant on iOS traffic on the whole (Somewhat True)
• LinkedIn is better able to model conversion data (Unlikely)
• Facebook and LinkedIn use incredibly different data collection methods (Very Unlikely)

With the most likely contributing factor being that LinkedIn’s B2B audience lends itself to quicker conversions (lead form fills) and a higher share of desktop and non-Apple traffic unaffected by ATT, that would mean that Facebook’s data drag is due to some combination of attribution loss and how its optimization algorithms operate in the face of less actionable data.

What This Means for Marketers & The Future of Measurement

What Facebook is experiencing has implications for its ability to reach new prospects and how you’ll need to manage Facebook Ads in the future. But if you’re relying on Facebook to report performance, it more plainly shows how big of a gap there is between what Facebook can tell you about its performance and its true impact on your business.

Facebook may be feeling the heat the most right now, and some ecosystems will sustain more deterministic data than others, but the privacy era threatens the efficacy of every ad channel that’s viewed as a performance channel. Modeled data already makes up a larger portion of reported performance now than ever before, and it is only likely to increase as more platform and policy changes come into effect. You’ll want another vantage point to allow you to compete much more confidently in the coming decade.

At a minimum, you should be comparing channel data with your website’s analytics program to see how ad networks are reporting performance against your own website over time. Like with any conversion tracking system, that will provide an initial gut check to find inconsistencies that you can account for when presenting performance internally.

That alone isn’t nearly enough, however. Web analytics programs rely on some version of the same thing Facebook and other ad networks rely on to track data, cookies amongst them. To really have a durable alternative to platform measurement, you must look at solutions that take two disconnected data sets and estimate causality. Running geographic tests to understand incremental business impact of creative, budgeting, or ad channels is one simple way. Running these tests has other benefits, too, like the ability to measure the incrementality of your marketing. Measuring this way doesn’t need to require more than advance planning and ongoing tracking.

Many other brands will need to go even further, exploring software or managed solutions to operate perpetual lift and incrementality testing at scale. For those that need more direction and guidance, working with agencies that can audit and implement new measurement and reporting practices will be important. Those that are most committed or have more business critical data needs might look at embedded agency partners, or find a partner to in-house these functions, from hiring statisticians and data science roles that specialize in marketing to building a center of excellence around marketing measurement tailored to your business.

This wide range of solutions may seem daunting to pick through, but consider that even now, some of your competitors are plotting their path forward to make this a competitive advantage. You can start very simply and then only move to deeper or more costly ones as you reach limitations and are more clear-eyed about what you need. For those relying entirely on platform measurement, though, Facebook’s current challenges have demonstrated that it is important to do something, and there’s no time like the present to start.