Month 1: Key lessons learnt about SKAdNetwork (SKAN) and App Tracking Transparency (ATT) so far

Month 1: Key lessons learnt about SKAdNetwork (SKAN) and App Tracking Transparency (ATT) so far
June 1, 2021 Simon Whittick

So, month 1 of Apple’s App Tracking Transparency (ATT) framework has passed. In this final lessons learnt post we review all the key things that have been identified and the lessons learnt with a full month of experience and discussion behind us.

In this post we’ll cover:

  • Realities of SKAN data by channel
  • Apple’s privacy threshold
  • ATT opt-in rates and adoption
  • iOS14.5+ adoption
  • Shifts in iOS media spend

We’ll also be discussing these lessons and shaping best practices on our 9th June webinar “Under the hood of ATT & SKAN: Lessons so far and best practices for the future” alongside industry experts Peggy Anne Salz. Thomas Petit, Alex Bauer and Paul Bowen. You can book your place here

Buckle up, here’s the review of lessons from month 1 …

Realities of SKAdNetwork (SKAN) data

After a month, we’re starting to understand the realities of relying on SKAN data. Here we summarise some of the most important landmines.

SKAN data by channel

One of the key lessons we’ve covered over the last few weeks is that data treatment is not consistent across media channels and MMPs. It’s important to understand what you’re seeing from each channel and also any funky logic in your MMP.

We’ve worked on summarising this information below based on some important landmines we’ve spotted in SKAN data for the biggest five channels that we report on. The information we have here is taken from live SKAN implementations and publicly available documentation*. 

It’s important to also note that data availability by MMP still varies. We’ve tried to present the best possible case of what the MMP is able to show in this.


A couple of important notes on this:

  • Geo breakdown: Whilst currently available using the IP address in the SKAN postback, it’s removed by SKAN 3.0, which will increasingly be used alongside iOS 14.6. This is likely why many of these channels have chosen not to support it and will leave advertisers relying on an already stretched Campaign ID to report by geo. Hopefully, Apple might also provide a privacy compliant way to expose this in the postback.
  • Google data: There’s still a lot of unknowns here because they haven’t yet integrated with MMPs (it’s imminent and we now believe it will be fairly similar to Facebook’s) and they are still building out their native reporting. They do have a SKAN specific native report but this is very limited. It shows conversion value count at a campaign level but not decoded. One key landmine is that if you do not select the SKAN specific report, then the standard Google Ads interface blends SKAN installs in the general “installs” column, which will cause discrepancies when comparing their UI, SKAN data, and MMP data. Also, for all levels the data shown in the Google Ads interface (unless selecting the SKAN specific report) are modelled conversions. The resulting numbers are based on a combination of MMP, SKAN and Google data. Again, this will make it hard to reconcile once your data is in an MMP.
  • Conversion Value decoding: For Facebook, Snap and TikTok the ConversionValue schema in your MMP is mapped to their own standard events. This can create discrepancies and means events can show up under the media channel event name e.g. app_custom_event.fb_mobile_content_view instead of the corresponding MMP event name.
  • Campaign name and ID enrichment: Some MMPs might enrich data from Facebook, Snap, TikTok and LiftOff with campaign names and IDs. This is useful, but requires additional advertiser configuration in your MMP.
  • MMP install date logic: Some MMPs have their own install date logic which will look at when the postback was received and remove some time to try and work around Apple’s random timer in a slightly basic way. This makes it impossible to reconcile data between the MMP and native reporting interfaces.

What do we learn from this? I think there are a few key takeaways:

  • Industry standards amongst media channels: We ideally need industry standards around how data is shared back from media channels to advertisers to make apples-to-apples comparisons easy. Or Apple could just share the raw postback directly to advertisers. From an advertiser’s perspective, ideally everyone would be more like Liftoff and similar networks who are transparent in their approach, with:
    • Raw postbacks shared
    • With un-modelled data
    • Include parameters around re-downloads
  • However, I suspect that the bigger players may be less willing to strip data back this much and remove their own modelling, so the ball may be in Apple’s court in terms of sharing raw postbacks with advertisers.
  • Industry standards amongst MMPs: Messing with the install date using questionable logic feels like a wasted effort that creates more confusion. That’s before we even get started on fingerprinting thinly disguised as probabilistic attribution / modelling. Ultimately, differing methodologies are creating more confusion rather than helping advertisers.
  • BI is going to become more important for mobile UA: All this noise in data from media channels and attribution tools is going to make the BI layer of your UA stack more important than ever. Removing noise, aggregating and normalising data and enriching SKAN data with probabilistic models will be an important layer to deliver a performance view that’s actually useful for optimisation. 

*Publicly available documentation by channel here:

Apple’s privacy threshold

We finally have some useful data on Apple’s privacy threshold and things appear more positive than initially thought. To recap, this is a privacy threshold set at the Campaign ID level by Apple where you need a certain volume of installs per Campaign ID to receive the Conversion Value data. When this threshold isn’t met the Conversion Value is returned as “null”.

Rich Jones, Head of Product at Dataseat shared some interesting research on SKAN Conversion Values. They’ve been analysing installs and what percentage are returning a null Conversion Value.

Source: Dataseat

Whilst initially, as reported by some, 80-90% of installs had a null ConversionValue. It appears that since the 22nd May only 40-50% had a null conversion value. Rich’s main hypothesis is that Apple has relaxed the privacy threshold around the 22nd May and thus more Conversion Values are now being returned. He shares more behind the numbers in this podcast.

The good news? Your SKAN data is getting a whole lot more useful.

Key lessons:

  • SKAN data is a bit inconsistent with different treatment from MMPs and media channels. There’s a need for industry standards or Apple to share the raw postbacks with advertisers to clear up the mess.
  • The BI layer in your UA stack is about to get a whole lot more important to clean, aggregate, normalise and enrich the mess of SKAN data and have a useful performance view for optimisation.
  • It appears that Apple has relaxed their privacy threshold which makes your SKAN data a whole lot more useful for predicting LTV by channel. If you’ve had a lot of null Conversion Values returned previously it’s worth checking back to see if that’s changed.

ATT and iOS 14.5 statistics

Lots of data has been put out there and we’ve seen plenty of conflicting numbers. Here’s some of the key research to pay attention to.

ATT opt-in rates

One of the more contentious questions in the first month of ATT has been: what are ATT opt-in rate averages? What seems like a relatively simple question has a very complicated answer. This is thanks to varying samples and methodologies for measuring opt-in rate across the reports released so far. 

Before we jump into those numbers, it’s worth identifying the app tracking statuses as identified by Apple:

  • ‘Authorized’: app users who tapped ‘Allow Tracking’ via the AppTrackingTransparency (ATT) prompt.
  • ‘Denied’: two groups of app users, which Apple does not separate:
    • Users who saw the prompt and tapped ‘Ask App Not to Track’. These users denied tracking for an app via the ATT prompt.
    • Users who went to iOS Privacy Settings and toggled ‘Allow Apps to Request to Track’ OFF. These users denied tracking for all apps at once, preventing apps from presenting the prompt.
  • ‘Restricted’: app users with ‘Allow Apps to Request to Track’ setting OFF and DISABLED. Apple disables this setting for devices with Apple IDs associated with minors, for example. These users never have a choice to select a tracking option and cannot be tracked by default.
  • ‘Not Determined’: app users who have not seen the prompt and who have toggled ‘Allow Apps to Request to Track’ setting ON. These users have not been asked yet to select a tracking option and are by default not being tracked.

Now that we’ve identified those it’s worth comparing the opt-in rate benchmarks from MMPs Appsflyer, Branch, Kochava and Singular as well as the widely cited and initially misunderstood numbers from mobile analytics tool Flurry:

Our aim here is to compare reported numbers that are calculated the same way. However, in many cases, despite the calculation being the same the sampling and methodology will have subtle differences under-the-hood. The main outlier is Appsflyer’s opt-in rate when comparing it to the numbers surfaced by Branch and Singular. The difference is likely a particular app, app category or geo skewing the figures in a sample. It could also be that their threshold for removing apps testing ATT is higher than others. It’s also worth noting that in all cases the reported numbers are for early adopting iOS users and early adopting apps, which does skew numbers.

So with the caveats out the way, what are the important things to learn here? You lose about ten percentage points off your opt-in rate at each stage of the opt-in funnel. Essentially, each of these calculations is showing the opt-in at different stages of the opt-in funnel:

  • When someone is definitely shown a prompt it’s around 40%.
  • When you include users who opt-out at the device level this number drops to just under 30%.
  • When you then include those who have been opted-out by Apple at the device level and those who haven’t yet seen a prompt (despite the app enabling ATT) the number is closer to 20%. This essentially tells you of the universe of possible IDFAs in your app, how many will you have access to?
  • However, as an advertiser, “Attributable Installs” is perhaps THE most interesting number. For attribution to happen outside SKAN you need dual opt-in on your side and the publisher side. Many MMPs haven’t been forthcoming in sharing this number yet, however, credit to Branch who did and found that this number stands at 9%. Whilst this number might rise slightly based on the sampling differences we’ve mentioned, it’s still low. At 1 in 10 installs being attributable outside SKAN, it does make you question if ATT is even worth the effort? Or should you now focus far more energy on SKAN?

Another important point is that these are global averages. The opt-in rate does vary by geo and app category. With smaller sample sizes and different methodologies across the reports it’s hard to compare geo and app category numbers but some directional trends appear to be developing for some segments across reports:

  • By geo: France is the only country that consistently appears at the higher end of opt-in rates across reports. Meanwhile, the UK, USA, Germany and Australia tend to appear to have lower opt-in rates.
  • By app category: Appsflyer has great breakdowns by app category in their report, which highlights that the top categories for opt-in are midcore games, shopping and food and drink. The worst opt-in rates were found in casual, hypercasual, hardcore and social casino games.

iOS 14.5+ adoption

When you average out the latest reported adoption numbers on the user side for iOS 14.5+ they sit at around 20%. The growth has been on a slow growth trajectory in recent days and weeks, with adoption a lot slower than most recent versions:

Source: Branch

Why? After a very short life. iOS 14.5 was rapidly followed by iOS 14.5.1 to iron out some bugs, including:

“Fixes an issue with App Tracking Transparency where some users who previously disabled Allow Apps to Request to Track in Settings may not receive prompts from apps after re-enabling it.”

However, 14.5.1 was reported to slash the performance of iPhone 11 and 12 models by as much as 60%. This may explain the slow rollout. We’re now at a stage where iOS 14.6 was released on May 24th, fixing “43 security vulnerabilities”

Many expect the release of iOS 14.6 to rapidly accelerate adoption of iOS 14.5+ versions with initial bugs ironed out. However, right now it appears that most iOS 14.6 adoption has come from those who had already upgraded to 14.5 or 14.5.1, not net-new. So iOS 14.5 still hasn’t moved beyond the early adopters. Eric Seufert also presents a counter option in this article that we may not see a massive inflection point of iOS 14.5+ adoption like we have with previous versions. Given the eyeballs on this around anti-competitive actions from Apple they might not pull the notification lever for 14.6 updates that Tim Cook’s hand is hovering over, because “the frog that’s being boiled doesn’t scream for help.”

We should have an answer in the coming weeks.

Importantly, iOS 14.6 does release SKAN 3.0 into the wild. The key feature of which is:

“Devices can now send install-validation postbacks to multiple ad networks that sign their ads using version 3.0. One ad network receives a postback with a did-win parameter value of true for the ad impression that wins the ad attribution. Up to five other ad networks receive a postback with a did-win parameter value of false if their ad impressions qualified for, but didn’t win, the attribution.”

Ultimately, this means one ad network will have credit for the last click, whilst upto five other ad networks will be able to claim influence. This will introduce further noise into SKAN data and if you’re being cynical could leave opportunities for those ad networks who aren’t sending raw SKAN postbacks to advertisers an opportunity to embellish their data. However, it would be hard for them to back this up.

ATT adoption

Given the slow user adoption of iOS 14.5+, it’s been easy for advertisers to take a wait and see approach to ATT. That’s exactly what’s happened. Appsflyer is reporting ATT adoption and overall it currently sits at 18% globally.

From conversations with advertisers there’s a few factors that are driving this low overall ATT adoption:

  1. Slow user adoption of iOS 14.5+ gives advertisers a chance to kick the can down the road and continue to operate as before for the majority of users without the limitations of SKAN and ATT.
  2. Given Branch are seeing only 9% of installs as attributable with dual opt-in many just don’t see enough upside from investing effort into ATT prompts and interrupting the user journey.
  3. For many app categories like fintech, where user privacy and data protection is paramount, they never intend to use an ATT prompt and are going to focus their energy more on maximising SKAN data.

Don’t expect there to be 100% ATT adoption across apps. Also, on that last point, Appsflyer share ATT adoption by app category, which reveals what types of apps might also be taking this approach:

App Category ATT Adoption
Hardcore Gaming 29%
Social Casino Gaming 29%
Shopping 27%
Midcore Gaming 23%
Food & Drink 21%
Social 20%
Hyper Casual Gaming 17%
Finance 17%
Casual Gaming 16%
Entertainment 15%
Lifestyle 14%
Photography 12%

Key lessons:

  • Current averages suggest that you’re going to get access to around 20% of users IDFAs in your app when using ATT.
  • However, in terms of attributable installs with dual opt-in on the advertiser and publisher side this number drops closer to 10%. 
  • At this point, it makes you question if ATT is even worthwhile? This is particularly a concern in app categories like finance where user data concerns are paramount. Just 10% of data isn’t going to be that effective for modelling. 
  • This is reflected by the fact that only 18% of apps have implemented ATT so far, with it lower in some app categories like finance. 
  • However, some ATT reluctance may also be down to iOS 14.5+ user adoption being around 18%. Many apps are simply kicking the can down the road and continuing as before while they still can.
  • In some geos and app categories opt-in rates are higher and may provide more value. For example, France, shopping, food & drink and midcore games seem to be trending higher with opt-in rates.

Media spend shifts

The expectation before ATT was that more spend would shift to Android and CPMs on iOS devices would plummet, hitting Facebook the hardest. So let’s have a look at data from the first month to see if those expectations have played out.

iOS vs Android spend

Looking at our own data (based on a sample of aggregated data looking at $500m annual ad spend), generally there’s not much to see here yet. Spend has remained pretty steady without any significant shifts between OS’s post-ATT, even when broken down by advertiser spend level according to our data:

There has been a slight decline recently for the smallest spending apps who we know more heavily rely on the duopoly of Facebook and Google as two out of three channels they run. Below it does appear that iOS spend share is starting to trend down for some of the major SANs as iOS 14.5+ adoption has increased.

This trend might not stick, or it could become even more apparent as iOS 14.5+ adoption increases. We’ll continue to share data in the coming weeks, so sign-up to our newsletter to get notified of new data drops. 

Other reports are either showing small dips for iOS spend vs Android or no change. It does appear that the iOS share of spend for major SANs like Facebook and Google is starting to feel the crunch. So for advertisers relying heavily there, we could see this trend become more pronounced as iOS 14.5+ adoption picks up.

iOS share of wallet and CPMs

In terms of CPMs on iOS, various anecdotal reports like this and this suggest there has been deflation. In general, reports have suggested CPM deflation of around 30-50% where the IDFA isn’t present vs where it is. Similarly, reports suggest that iOS 14.5 CPMs are about 10-20% down vs previous versions.

Most reports about Facebook performance seem to suggest that it’s been a rollercoaster ride since ATT launched, with Facebook testing their algorithms in this new environment. We’d expect this to settle as those algorithms learn and advertiser spending starts to stabilise.

From an overall share of wallet perspective, there’s nothing to write home about yet in terms of swings in our data.

If you were pushed to analyse the small (not significant or wholly indicative) movements, it does appear like Google’s share of wallet has declined slightly, whilst Apple Search Ads is up slightly. However, these are tiny trends that could fluctuate week-to-week. The real swings will likely come when iOS 14.5+ adoption reaches a critical mass.

Key lessons:

  • On SANs it feels like you’re going to be in for a rollercoaster ride over the next few weeks on iOS. However, it is worth continuing to experiment to get a short-term advantage when things level out.
  • Overall expect costs to increase on Android and decrease on iOS over the coming weeks. The impact will likely get bigger as mass adoption of iOS 14.5+ kicks in. SANs are likely to see the biggest drops in advertiser competition.

Wrapping up

It’s been an interesting month seeing ATT enforcement play out. However, it does just feel like we’re only reaching base camp in terms of the journey ahead. There will be plenty more lessons in front of us. If we were to summarise what advertisers should be doing now, they’d be:

  • Don’t invest too much energy in ATT, the effort vs reward doesn’t seem like it will pay back, unless you mainly monetise through ad revenue.
  • Focus more energy on understanding your SKAN data and discrepancies.
  • Focus energy on building a BI layer on your UA stack to aggregate, normalise and enrich SKAN data with probabilistic models so that you get a performance view that is actually useful for optimisation.. 
  • Prepare budget plans to account for bigger fluctuations in performance for iOS spend on major SANs like Facebook and Google. Also prepare them for bigger increases in costs on Android in the coming weeks. However, continue to experiment on iOS so that you can gain some short-term wins when the algorithms of the likes of Facebook work things out, before other advertisers flock back. 

We’ll be back with more post on specific SKAN and ATT topics in the coming months, so remember to sign-up to our newsletter to get notified when they go live. Also, don’t forget you can reserve your place at our  webinar on 9th June “Under the hood of ATT: Lessons so far and best practices for the future We’ll be joined by industry experts:

They’ll be helping us analyse early lessons and shape best practices for the future. We’ll also be revealing results from our proprietary analysis of media spend shifts in the first few weeks of ATT. Sign-up by filling out the form below. We look forward to seeing you there!