SKAN v4: 27 questions answered

How will SKAN v4 impact the mobile advertising industry?

App Tracking Transparency and SKAdNetwork have been the biggest changes in mobile marketing since the introduction of the IDFA. They have completely disrupted an ecosystem built on relatively stable, easily available, and very shareable device identifiers for the vast majority of smartphone owners.

As every mobile marketer knows, SKAdNetwork has been tough. It’s hard to set up, it offers limited capability, it broke attribution methods and BI systems, and it forced growth teams to change long-standing practices.

But there is light on the horizon. At WWDC 2022, Apple teased SKAN v4, and there’s some hope. SKAN v4 offers more data, more postbacks, and more opportunity to both measure and optimize marketing results.

So Singular CEO Gadi Eliashiv and a team of Singular experts dug deep into what Apple has announced so far about the fourth major iteration of SKAdNetwork, and we shared the results in a webinar with over 1,000 registrants: SKAN v4 Deep Dive: How iOS UA Will Evolve.

We had more questions in this webinar than ever before, and I’d like to take this opportunity to answer them.

27 questions on SKAN v4 answered

1. If UA campaigns become spread across thousands more campaigns, won’t we also just not end up reaching thresholds? And thus never getting beyond coarse?

A: Exactly. You’ll still have to concentrate significant numbers of installs inside each campaign (we’ll only know exactly how many when SKAN v4 is released and we do some testing). 

Upshot: advertisers with vast scale will benefit the most.

2. Can we ever expect Apple to allow MMPs to distinguish installs attributed via Apple Search Ads from all other channels? This fundamental flaw has most UA managers paying for installs coming and going.

A: That would be nice. We’ll ask Apple 🙂

3. Do you know if the coarse values will be customisable? Same question for the conversion window for multiple postbacks? My understanding is yes but is it?

A: Since the coarse values can refer to anything you want them to, they are (in a sense) customizable.

The conversion windows, however, are not. You can set values whenever you want, but they’ll only be sent after the conversion window expires:

1. 0-2 days
2. 3-7 days
3. 8-35 days

4. Has there been any speculation on whether the threshold for fine values could be increased vs. remaining as is? Or is it a given that coarse values will come in at a lower threshold? Same question for campaign IDs.

A: It is a given that there will be lower thresholds for course conversion values than fine, but Apple has not spoken on where the levels will be set, or on how the levels will compare to existing privacy thresholds.

We’ll simply have to wait for the spec, and probably also for some real-world testing.

5. Regarding the 3 postback windows,  does the conversion value schema definition need to be the same for all 3 windows?

A: Not really. The fine conversion values offer much more data; the coarse values just 3 options. Those 3 could be identical to some of the values you’ve defined for your fine conversion values, but you’re likely to have many more options there.

6. Are you suggesting that the 3rd postback would only be sent on Day 35 then?

A: Yes, or shortly thereafter, if there’s a random timer for privacy enhancement purposes.

7. Concerning conversion postback randomization … do you foresee this having implications on fingerprinting?

A: I think they’re different issues. (So essentially: no.)

8. How do you think it will be possible to connect all three postbacks in order to understand that this is a single install?

A: No. Think of them as aggregatable data for cohorts, but not for connected data points in a customer journey.

(I did initially when viewing the Apple presentation, but had my hopes dashed upon further investigation.)

9. Will the 2nd and 3rd postbacks at least allow us to see much more aggregate post-install data?

A: Yes, given that there was basically none available before SKAN v4 🙂

10. How does probabilistic attribution factor into the whole ecosystem? Why do ad networks need to run on SKAN postbacks? Why can’t MMPs make probabilistic attributions and post those to ad networks for decision making?

A: Apple has explicitly stated that fingerprinting is not allowed, whether ATT consent is granted or not.

All the major self-attributing networks support SKAN and make fingerprinting impossible, so that’s 75% of ecosystem volume. Only a quarter therefore even makes fingerprinting possible, but with more data available in SKAN v4, that’s the way to go. 

(Plus, it’s less risky. If Apple doesn’t allow fingerprinting, apps or SDKs that enable or provide it will run serious risks in doing so.)

11. I hadn’t seen anything on the randomized timer in conjunction with the set windows for postbacks. Is there just an assumption that there is a randomized timer to protect user anonymity?

A: Yes. We’ll have to see the final specification for confirmation.

12. Even if we don’t know that the postbacks are from the same person, we will know they are from the same cohort, right?

A: Exactly. Yes.

13. How is the coarse part of the second and third postback windows connected to the first postback? Let’s say we had enough installs per campaign , will the second and third postback be dependent on the activity of those installs then?

A: As far as we can tell, each postback opportunity gets sent or not based on its own merits in terms of achieving crowd anonymity.

See also Q5 above for some related details.

14. I got a little lost on the “fine/coarse” conversion value and the 3 postbacks. Can you give an example of how an advertiser (say, in rideshare) might set up a conversion model with SKAN 4.0?

A: Absolutely. Here’s one potential option: 

1. Installed, took some action
2. Took rides, value totalled either $0, $0-50, or over $50
3. Shared the app with a friend, posted to social, or didn’t

That is literally just one potential sequence of an infinity of possible ones. Insert your core KPIs for one that makes more sense for your specific business.

15. Will SKAN v4 increase the likelihood of a fingerprinting ban enforcement?

A: I want to say yes, but Apple has already banned fingerprinting … so no, not really. 

I think SKAN v4 makes it more likely that Apple will more aggressively enforce its already-stated rejection of fingerprinting, because now there is a more viable privacy-safe measurement solution than previous versions of SKAdNetwork, regardless of what people select in the App Tracking Transparency prompt.

16. Do you know if the range for low/medium/high coarse conversions will be customisable or if we have to assume that it will just divide the fine conversion in 3 equal tiers?

A: There will be 3 tiers. You will get to decide what you assign to each tier.

17. Have there been any updates on the timeline for SKAN v4 release (beyond “later this year”)?

A: No 🙂

If Apple makes its self-imposed “this year” deadline, my guess is October/November is the most likely time frame. That’s enough time from now for significant amounts of development, and it ensures that Apple doesn’t drop a new version of SKAN — even if it’s beta — right before the Christmas break.

18. Did I understand correctly that now the CV sending windows will not be dynamic, but static and will definitely be 0-2, 3-7, 8-35?

A: Yes, but we assume some randomness to the sending to preserve anonymity.

19. Is there any work around on retargeting with v4?

A: Not yet. That would require something like FLEDGE in Google’s Privacy Sandbox for Android, and we haven’t seen that from Apple yet.

We’ll have to file that in the would-be-nice-in-SKAN-5 file.

20. Can the coarse value be set differently based on the original fine value? Or perhaps in context of previous coarse values?

A: Yes, and yes.

It’s safest not to think of them as connected sequences but disconnected signals at different stages in the progression or aging of a cohort.

21. How are coarse values being defined? Is this something we can map on our end similar to how we map values and events to the 64-bit conversion values?

A: They are simply 3 values which you map in your systems to whatever events or activity you wish. (So that’s a yes.)

22. Does CV 0 still need to be the app first open? Or can you trigger CV 0 as another in-app event? (For example, start postback windows and attribution on a registration type event)?

A: It does not need to be the first app open, but you will need to trigger it (or lose your first postback opportunity) within D0 to D2.

So if you get a registration type of event within D0 to D2 (and your app or attribution SDK better be keeping track of timing), you could totally do what you’re suggesting.

23. What is the future of tracking owned media? (i.e. traffic from first-party mobile web, email, SMS?

A: Apple adding web-to-app is a very positive sign here. We don’t have the full implementation yet, but privacy thresholds still apply (which hurts a bit for owned traffic). Using campaigns effectively and even applying old-school methods like offer codes will be important for many.

24. Which platforms do you think will cope with SKAN 4.0 changes and which ones will not?

A: Time will tell 🙂 

Best guess: the big ones will get on the train. Some of the smaller and fringe players will delay and make errors. It will be messy. You’ll have to support users on lower versions of iOS with perhaps earlier implementations of the SKAdNetwork framework, and ecosystem partners who are at various stages of development as well.

25. How do you think Private Relay will be utilized?

A: We see it expanding over time, perhaps to all web traffic in addition to app traffic.

This article from late 2021 still represents my viewpoint on Private Relay: Almost every iPhone owner will be using iOS 15’s Private Relay soon

Of course, “soon” might be 2023 or 2024. There’s a lot for Apple to build (or partner with) in terms of capacity to support that kind of upgrade.

26. For an agency: what changes do we need to make in the postback which is updated by the MMP?

A: We don’t have the specification yet, so we’ll have to wait and see. We do know that Campaign ID is changing to Source ID.

27. With crowd anonymity thresholds applied to postback 1, would you agree that advertisers will likely be getting less information on user value in many cases than we’re currently receiving? Do you think it’s more likely that postback 1 might only get a coarse CV?

A: Some have speculated this, yes, but the truth is we just won’t know until we see the spec, start testing, and study early results.

You really, really, really need to watch the whole webinar

There’s much, much more detail in the full webinar, including discussions of:

• Crowd anonymity
• Campaign ID to Source ID and 4 digits from 2
• New conversion value type
• 3 postbacks versus 1
• Changes with timers
• Cohorts … funnels … customer journeys
• Web to app
• Conversion values going down as well as up
• Setting conversion values for postbacks 2 and 3
• Ad campaign optimization
• Figuring out conversion models in SKAN 4
• Modeling missing data
• Fraud
• Ecosystem changes
• Testing set-ups