Mobile marketing ecosystem and SKAdNetwork: your questions answered about DSPs, SSPs, MMPs, and any other P’s we can think of
Getting ready for iOS 14.5 and SKAdNetwork has been a massive job for marketers and publishers. But it’s not just about those two absolutely key components of the mobile marketing ecosystem. If demand-side platforms, supply-side platforms, and mobile measurement platforms don’t also prepare, the result will be complete chaos.
Apple’s iOS 14 privacy changes are essentially forcing an entire ecosystem to rewire itself.
And that doesn’t happen easily. Or quickly.
Recently we offered a webinar with Liftoff, Fyber, and InMobi to talk about the state of industry readiness, the essential components that each party has to prepare for SKAdNetwork readiness, and actionable steps marketers can take for a (relatively) smooth transition. We also chatted about how to test a SKAN set-up, optimizing consent, and much more.
At the end, we were flooded with questions and could not answer them all live. So, as per usual, we’re answering them here on the Singular blog.
Our great panelists helped provide some of the following answers:
- Eran Friedman, CTO and cofounder: Singular
- Andry Supian, senior product manager, Liftoff
- Natalie Binns, senior product manager, Fyber
- TK, Ram Krishnamurthy, VP, InMobi
Extra questions on SKAdnetwork, ecosystem readiness, and marketer challenges
What does MMP stand for?
Thank you for a starter question to get us going!
Mobile measurement partner (what Singular is). MMPs have access to data from sources like Facebook, Google, Apple Search Ads, Twitter, Snap, and many other platforms to help advertisers understand how their campaigns are performing.
Where can we get the SKAdNetworkIdentifiers for our iOS app? In the documentation some sample ids are present … do we use them as our SKAdNetworkIdentifiers or do we have to get IDs specific to our iOS app?
Publishers have to add the list of SkAdNetwork IDs provided by each SSP (on behalf of the DSP partners) to their info.plist file to support attribution via SkAdNetwork. Check with your partners for the right IDs, but for example, you can find InMobi’s here.
What do you think will be the impact of SKAdNetwork on the attribution of mobile web to app traffic?
There are few ways to look at this.
First, the lack of a device identifier in iOS14 will have a huge impact on user acquisition, since attribution as we’ve known it will be a challenge. Second, the more SKAN-enabled inventory there is to target, better the ecosystem’s ability to adapt for user acquisition attribution.
We know that Apple is also working on Private Click Measurement. Currently, that’s in WebKit for “privacy-preserving measurement of ad clicks across websites and from iOS apps to websites in iOS and iPadOS 14.5 betas,” Apple says. I think we can safely assume that Apple is aware of the mobile web to app marketing flow and will be looking for a way to safely measure that in a privacy-sensitive way also.
About redirects: to clarify, are redirects allowed in the click user flow, or are the users sent directly to the app store, and all downline links – to the ad network and MMP for example – all clicks have to be server-to-server posted to the ad network and MMP?
It’s not about deeplinks, it’s about click measurement.
Redirects for clicks are allowed, so you can send people to your MMP first, which will redirect the user to the App Store or Google Play. This enables the MMP to count clicks and powers advanced user flows like dynamic redirects, linking to the correct store based on the device type.
In the past, you could use the information from these clicks for measurement as well (based on IDFA, if provided, or fingerprinting). Moving forward, however, you are only allowed to use these clicks for user-level measurement if you know that a user/device has approved ATT for the app.
So … as things currently stand, redirects will not enable measurement. Even though deeplinks and deferred deeplinks will still function from an end user point of view, they won’t enable attribution or marketing measurement.
How do you see the chain work with an ad network buying traffic through a DSP? The DSP is doing the signing with their SKAN ID, and then relays that to the MMP. In that scenario, the advertiser is having to select the DSP as the “media” partner, and the ad network doing the buying has no direct connection to that transaction or reporting flow. Any thoughts?
Whoever provides the attribution needs a SKAN ID.
As a DSP, understanding supply availability is essential. Will SSPs display applicable (SKAN enabled) dimensions in their reporting UI tools?
Each SSP is likely to be a bit different, but most likely will. They are dependent on publishers, however, and publishers are quite late to the SKAdNetwork party.
What do you expect the impact will be on publishers? Do you expect UA costs to go up and ad revenue to go down?
I think it depends very much on what kind of app you have and what vertical you’re in. Publishers in vertical with broad appeal may find that contextual targeting works just fine for them, while app publishers in niches that require very specific users will likely have a harder time targeting them.
In addition, because tight targeting on premium users/customers via IDFA won’t be nearly as prevalent, ad revenue for apps with a premium audience could go up. Simply put, contextual factors will be much more important for targeting premium audiences, and what app or website they’re on is a big clue.
One other factor here: a minority of publishers are ready for SKAdNetwork, meaning SKAN-compatible ad inventory might be limited in supply early on. That could also drive prices up in the short to medium term.
InMobi VP Ram Krishnamurthy adds:
Publishers will see a short term impact on both user acquisition and ad monetization. UA will become more expensive as attributable traffic on iOS will shrink. From a ad monetization perspective there are steps you can take towards adapting to the changing landscape. We can find them here.
Have you been able to figure out either from the DSP, SSP or MMP side what is the minimum amount of installs required to receive conversion values in the postbacks?
I’ve heard of cases where marketers have received data from as few as 50 installs. Note, however, that privacy thresholds operate on many levels, not just total installs, so if you have complex conversion models with postbacks firing at different times, you could have a much harder time getting data back.
We’re working with a self-service DSP that wants to use their SKAN ID instead of ours but they’ll use our impression and click URLs when serving ads. Will the DSP company name or our company name show for those conversions in the Singular dashboard? Will this scenario cause any other issues?
It depends on the configuration on the MMP side – the SKAN postbacks would show under the name of the configured SKAN ID, while the impressions, clicks and attributed IDFAs for opt-in users would be based on the tracking link source. We would recommend keeping these two consistent since advertisers may be confused when comparing between the two result types.
Contextual Data … do you have any case studies related to the contextual performance you have been seeing? Are all DSPs able to accept/ingest the new contextual parameters being sent by SSPs? What legwork is needed for DSPs to ingest this new data?
InMobi VP Ram Krishnamurthy says:
In general, now is the time when publishers should think about the data they have, plus what and how they can share it with their SSPs who can then use that with DSPs for targeting. Dimensions worth considering include:
- Time spent in app
- Session depth
- In app purchases
- In app behavior
How are contextual signals (eg. session depth) being standardized? How can a DSP use them unless they are standardized?
There really isn’t a global standard, but what Singular is doing is providing a translation layer that decodes what you mean by a conversion with a value of 4, for example, to something that an ad network understands. Once you’ve set up your conversion model in Singular, our SKAdnetwork integrations with supported partners allow partners to decode the meaning of SKAdNetwork conversion values.
In other words, all you need to worry about is defining and iterating your models.
What are the creative changes if any for working with SKAdNetwork? For display, native, video, and playable ads?
InMobi VP Ram Krishnamurthy says:
We don’t think there are any creative changes required per se, but we do think from an ad unit/slot perspective, full-page interstitial will have a lesser impact on their eCPM as opposed to static banner slots.
For probabilistic web to app tracking, it’s our understanding that if the user does not accept the ATT prompt on user open, that any probabilistic or fingerprinting tracking will 100% not be allowed. Is this correct?
Yes. Get all the details here in our fingerprinting in iOS 14 blog post.
Has there been any guidance on what is the reattribution window in SKADNetwork? If a user installs and reinstalls twice in the same day, would we get two postbacks or is there a minimum amount of time that needs to pass between those installs?
There is a redownload parameter in SKAdNetwork version 2, so theoretically you can get multiple postbacks for the same device and/or user. However, given that SKAdNetwork has a delay mechanism in place, it is extremely unlikely. Also, there are no postbacks for re-engagement campaigns where the app is still installed on the device.
So the most likely scenario where this could happen is an install, app deletion, and reinstall days or weeks later.
How do you envision retargeting campaigns in SKAdNetwork? What are the solutions being worked on to continue retargeting efforts?
If you presented the App Tracking Transparency prompt and your users accepted tracking, providing their IDFA, you would be able to use it for retargeting, provided that both publisher app and advertiser app have permission.
If you don’t have that, you are essentially unable to do classical retargeting advertising. A potential exception would be if you received a first-party piece of data like an email address from your users during sign-up or purchase events, and use that in your own owned systems (like email marketing platforms) to reach out and attempt to re-engage.
Note: without ATT permission, you cannot upload any identifier to an ad network and use it for retargeting, even if it is a first-party identifier.
Can you evaluate how much CPM performance might decrease without IDFA? Is it close tothe monetization of Limit Ad Tracking – on (LAT) users now?
InMobi VP Ram Krishnamurthy says:
LAT versus non-LAT has huge variance. For example, in the static banner world, it’s almost half … LAT traffic sees half the eCPM of non-LAT traffic. In case of interstitials it is about 80%.
About conversion values: How do you recommend the advertisers find the right valuable events from their app to send back to the DSPs?
That is literally the multi-million dollar question. The easy answer is: test, test, test.
The slightly more useful answer that we’ve been hearing from top marketers lately is this: pick early events, even if they’re less predictive, because quick feedback is essential. Don’t pick too many conversion models because Apple’s privacy thresholds ramp for each variation. Try something simple and iterate.
Important note: check out these references on predictive LTV in SKAdNetwork and iOS 14 conversion optimization. (The second one includes information on how to simulate different conversion models on existing data, meaning that you can check what a different conversion model would do without actually having to implement it.)
When buying Limit Ad Tracking (LAT) traffic at scale, with no IDFA, MMP matchings would have all been done probabilistically. Could that have created a false increase in tracking rate?
It would all have been categorized as probabilistic, as least in Singular’s systems, and therefore would not have increased the percentage of traffic that was directly and deterministically trackable.
Probabilistic tracking, however, is only 70-95% accurate, if that’s what you mean.
Ultimately, in iOS 14, fingerprinting is of extremely limited utility.
Will advertisers see 2 performance reports from DSPs?
It depends on each DSP’s policies. For performance reporting in Singular, you’ll be able to see a merged report as well as separate SKAdNetwork and IDFA reports.
Is the conversion value update postback sent only after the rolling 24-hour window expires for that player? Are the intermediate values also available somehow?
You will only get one postback under the SKAdNetwork framework, so no: no intermediate values.
Note: you can delay that postback up to a week if you think your conversion model warrants it.
On the buying side of things, any strategies on buying with re-targeting and exclusions based on IDFA going away, and frequency caps also being severely restricted or going away without some creative solutions like frequency capping using WIFI ‘unique’ IPs, but do you hear any solutions from buyers on the above?
Retargeting is still available IF you get tracking permission in both the advertiser and publisher app. That is likely, however, to be rare in both, and even more rare in the overlap.
Similarly, exclusions would be possible with tracking permission, but is likely to be rare.
Frequency caps won’t be easily doable without IDFA. You might look at some IP-based solutions, but anything approaching fingerprinting without tracking permission is forbidden by Apple, so that’s likely to be a limited solution at best.
How will SKAdNetwork and low consent rates on ATT change the lives of brand advertisers? Does it impact measuring viewability, frequency, and reach of the campaign?
InMobi VP Ram Krishnamurthy says:
ATT does not affect viewability as much as it affects attribution. The bigger impact for brands will be the lack of targeting.
And on frequency capping:
As mentioned previously … that’s going to be tough.
Some ad networks are implementing short-lived IDs that they say will enable frequency capping and use a “non-persistent revolving identifier which is not connected to a user device,” and therefore doesn’t transgress Apple’s policies. Time will tell if this works, but even if it does — and it passes the Apple sniff test — it won’t be as effective as IDFA.
What are your recommendations for resolving data discrepancies in hybrid setup of probabilistic attribution and SKAN postbacks? Looking at a cohort which is reported to have different revenue using these two mechanisms, how do you decide which is correct?
My recommendation is that you are likely in contravention of Apple’s regulations regarding no tracking without permission. If you have SKAN data, you have deterministic attribution data. If you get ATT permission, you have deterministic IDFA-based data.
Mixing the second with probabilistic data seems counterproductive. Mixing the first with fingerprinting data seems to put you at risk of getting your app kicked off the App Store.
More questions? Of course there areun
Everyone has more questions because everyone’s situation is just a little bit different. Singular was the first MMP to embrace SKAdNetwork, and we’ve been advising our clients on how to prepare for months.
If you need help get in touch, set up an appointment, and get on the line with one of our experts.
We’re more than happy to help.