SKAdNetwork gets real: your conversion, privacy, and ad monetization questions answered
After a year of talk, talk, talk, we’re finally living in an iOS 14.5 and SKAdNetwork world. And while it’s spring in the northern hemisphere, the sun is shining, and life is good, work is totally insane because everything is upside down in the world of mobile marketing on iOS. Shockingly, just shockingly, huge chunks of the mobile ecosystem were not actually prepared for iOS 14.5 and SKAdNetwork.
So are you completely and totally set up for SKAdNetwork in your iOS campaigns?
We recently brought together an all-star cast to get ground-level insights on critical components of a winning SKAdNetwork setup and best practices for conversion value management. We also chatted about fingerprinting (of course!), App Store rejections, supply side challenges, fraud prevention, and reporting. (Check out the on-demand show here.)
But as per usual, there were more questions than time.
So I’d like to take this opportunity to answer all the questions from webinar attendees that we couldn’t answer live. (Note: answers are provided by Thomas, Gadi, and Alon, with some additions and details from me. Any errors are entirely my fault, so hunt me down if you find them.)
Your top SKAdNetwork questions, answered
Q: What are the top 5 platforms that are most prepared for ATTmageddon?
A: We see more than 5 platforms that are ready, and we’re adding more to the list every week. Readiness is measured both by a working setup which includes SKAN campaigns, API and data access, reporting, and of course publishers supporting the demand side. You can find our integrated partners here.
Q: Has anyone received the source app populated on SKAd postbacks?
A: Yes, you should be able to receive the source app for postbacks. If you’re not receiving this, it could be because you haven’t crossed the privacy threshold.
Q: If users opt out ATT consent, is there a way we can send them an email and request opt in?
Thomas Petit says: You can send users to settings to change them manually. There’s a deeplink for that, and it can be from an email, or via an in-app message, etc. Don’t expect a very high rate of completion though: few users are expected to go through that process, at least in the first few weeks.
Q: What are your thoughts on the privacy threshold on the conversion values? Only a small number of players convert, and those are the ones we really care about …
A: No one knows exactly what Apple’s privacy thresholds are. What we do know is:
- Privacy thresholds are not reliant on conversion values. So using a smaller pool of possible conversion values will not make more conversion values appear in your postbacks. This is easy to tell by the fact that conversion values are not signed and not sent to Apple’s servers.
- As far as we know, privacy thresholds are determined by campaign-id and source-app-id, but the exact thresholds are unknown.
- Technically, Apple provides a device with a signed SKAdNetwork postback and a flag telling it to send out a conversion value. That means that technically it’s possible that different thresholds affect the inclusion of the source-app-id and the conversion-value. This can probably be determined by looking at the data and seeing if there are cases where only one of the two fields appear.
Q: How does SKADNetwork impact log level data access for companies that build custom algorithms?
A: There is an impact because the data is different … but you still have data. These companies need to completely rework how they look at quality campaigns. You have to look at your SKAdNetwork data in aggregate. You need to extrapolate quality scoring from limited data which will be a challenge, but isn’t impossible.
Q: Do you think it will be feasible to guess at user attribution based on conversion values? For example … two users with a given conversion value from a particular country … let’s say the UK. Then, since one campaign had two such conversion values from the UK, those users must have come from that campaign. Feasible?
A: Thomas Petit says: Some companies work on probabilistic modelling like this, from internal data (unlike fingerprinting). Check, for example, the work from Algolift on this. Don’t expect to replicate what you had before. The insights from such methods differ from previous ones, but can be directionally helpful.
I’ll add: This particular method is clearly unlikely to work at scale. It’s also an additional layer of complexity. And, you get campaigns (yes, just 100, or much less on some platforms) in SKAdNetwork. I’d be willing to bet that if Apple caught wind of serious threats to the privacy levels it thinks are necessary, it will refactor SKAdNetwork to add more obfuscation.
Q: Can you talk about how time based metrics (D3 retention, etc.) can be incorporated in this new world for optimization?
A: You can use some conversion value bits to encode time since install. At Singular we call this a retention bit as part of our measurement periods. We support up to seven days.
However, it’s important to note that the longer your measurement period is the less accurate your reporting will be due to inherent randomness built into timers. If you want to measure longer retention, you need to balance that with your partners who ask for shorter measurement periods. For example, Facebook doesn’t support long cohorts today and recommend 24 hours to ensure they can optimize as quickly as possible.
Q: What is your model preference? Is it better to go simple to start, especially since the industry is not quite there yet?
A: There are a slew of conversion model options for marketers to choose from. The key is to first define what events and conversions can be used to predict the future value of an activity for your specific app.
For example, if you monetize via in-app purchases, you may want a revenue-based conversion model that helps you optimize towards high-revenue users. Or if users need to make a sequence of actions for a significant conversion, you could use a funnel model. Leveraging early growth indicators in your conversion model will help you to understand which campaigns are performing and what they drive.
One thing to note is that different business models are impacted differently by SKAdNetwork limitations. If your app normally optimizes on D1 revenue because your users take action on the first day, you won’t see a big difference in KPIs.
But if you’re an app with a 30-day trial, or if you’re a game that monetizes over time, you need a longer attribution window, and since that isn’t available, you have to pick an earlier event that is a good predictor of future growth.
Q: What do you think the impact will be to IAA (in app ads) monetized games? Do you expect revenue to be heavily affected?
A: We do expect that Q2 of 2021 will be challenging with eCPMs on iOS dropping by 20%-40% initially. Having said that we do not foresee a catastrophe and most advertising spend will be retained, though it may be distributed differently. We believe that in Q3 we will start to see numbers picking up again and getting closer to 2020 levels.
Q: Not sure if I heard correctly, did Gadi say filtering consent will not meet the review requirement. What happens if you have a primer gating the ATT native pop up, with two CTA buttons … for example Allow, and Ask Me Later?
A: Thomas Petit says: Apple updated their guidelines specifically banning such filtering. Your pre-prompt screen can have only one option and has to trigger the native prompt.
“If you display a custom screen that precedes a privacy-related permission request, it must offer only one action, which must display the system alert.”
Q: We are seeing 10% accuracy of SKAN vs our MMP (IDFA based). Do you think MMPs will go towards modeled conversions for opt-out users to compensate for this under-attribution?
A: 10% shouldn’t be the case. We’re seeing some customers with 80% accuracy, but keep in mind that it’s highly dependent on the ad networks you’re working with and making sure targeting is working. But we have seen customers working with partners quoting 80-90% accuracy.
Q: What about GDPR in Europe? Do you have to trigger a GDPR notice first, and then ATT?
A: Alon Golan says: Apple’s policy and GDPR do not necessarily define privacy the same way, and for that reason there may be a need to show two prompts. You may decide to show the GDPR notice first and if the user rejects consent then perhaps you can avoid showing the ATT popup as well. Having said that … this is open to interpretation for every app developer.
Q: Can someone touch upon fingerprinting vs. probabilistic determination? Can we still operate on a CPI model given this change?
A: There are two parts to this answer …
Gadi Eliashiv says: Fingerprinting is about identifying a specific device and associating the device to a specific group (campaign, creative, or publisher app) using non-deterministic identifiers … mostly IP address and User Agent.
The fact that attribution using these identifiers is not 100% accurate, and hence probabilistic, doesn’t make it legitimate. Even if you do such attribution on the server-side and only share this data in aggregate, it still does not comply with policy.
Probabilistic attribution which does not rely on associating a specific device to a specific group is a different topic. Given that this does not rely on device level data (IP included) and is mostly done to form aggregated reporting, it is allowed. This is somewhat more similar to MMM and other classic methods for aggregate data analysis.
Thomas Petit says: While the terms are sometimes used for each other, I see a clear distinction between them: fingerprinting is one of various types of probabilistic attribution, and it’s not allowed.
- Fingerprinting is used to try and identify a specific device using user-level data (such as IP) for the purpose of attribution, AKA “tracking across apps & websites.” As accuracy can fluctuate, it had been used when other methods (such as IDFA) aren’t available, as a fallback method. It’s now been unequivocally banned by Apple from April 2021.
- Other probabilistic attribution methods remain valid andallowed. They are sometimes called media mix modelling and use anonymous first-party data only such as internal ID and funnel behaviour. This does not match device info from other sources, and as such is legitimate under Apple’s new policies. (Example: Algolift. MMM is a complement, not a replacement.)
Q: As an incentive-based business we need to reward our users, so getting user-level data is a must. In addition, our traffic is mobile-web based. What solution does Singular have for such cases?
A: There are certain workflows where you can still collect user-level data. Mobile web-to-app flows is one such example. We go beyond mobile attribution, and also provide web and cross-device measurement, so I definitely recommend you reach out if you’d like to discuss your current setup.
Next steps: maximizing SKAdNetwork data for growth
If you’re looking to learn more, watch the on-demand webinar for a huge amount of insight and guidance from experts. If you’re ready to improve your SKAdNetwork set-up and engineer for growth on iOS 14.5, book some time with a Singular expert to talk through what you’re doing, what you need, and whether we can help.