Rovio SKAdNetwork questions answered: everything you wanted to know about Rovio’s iOS 14 mobile user acquisition infrastructure
If only you could ask one of the largest and most successful mobile publishers on the planet how they are planning to run user acquisition and growth campaigns on iOS 14.5 …
(You can. Sort of.)
In fact, a week ago 1,249 people did on a live webinar where Rovio data scientists, marketers, and executives shared their growth stack set-up, key research findings, strategy, and best tips for others on mobile marketing in a privacy-safe era with iOS 14 and SKAdnetwork. On the call: Rovio’s VP of Marketing Kieran O’Leary, SVP R&D David Mason, and senior data scientist Frederick Ayala, along with Vungle’s senior product manager Rina Matsumoto and Singular CTO Eran Friedman.
The only problem?
1,249 people can’t ask all of their questions in a single one-hour webinar.
So we collected them all, lovingly hounding the Rovians, Rina at Vungle, and Eran at Singular for answers. And, honestly, they were all incredibly generous not just with their time but their expertise. There’s no competitive reason for Rovio to share so much insight with the industry: they are literally doing this for the good of the ecosystem.
So we’ve collected all those questions and answers, and are presenting them here.
Suggestion: sign up for on-demand access to the webinar and watch it while you scroll through this incredibly informative recap of the questions we didn’t quite get to.
Ok. All signed up?
Let’s dive in.
Rovio and Vungle’s SKAdNetwork answers
Q. How is Rovio improving/maintaining targeting under SKAdNetwork?
A. When it comes to SDK-based ad networks, the current targeting capabilities shouldn’t be heavily impacted by SKaDNetwork. In other channels such as Facebook, it’s fair to say that on average we’ll have to go broader and/or leverage interest targeting since this parameter is here to stay – interests are Facebook’s first-party data.
– Kieran O’Leary
Q: How is Vungle adjusting their products to the delay in installs and moving to conversion values?
A. We’re making adjustments in our models to adapt to the new delays in installs. As for the conversion values, we’re working on the conversion value API integrations with MMPs to be able to ingest the data and integrate into our optimizations.
Q: What sort of impact do you expect on the efficacy of attribution data via SKAdNetwork as compared to existing methodology of IDFA?
A: It remains to be seen, but if we’re looking at our own take on conversion values, running an analysis on previously acquired cohorts, we can hope for an accuracy of about 90%. If we end up using Facebook’s take on conversion values, this accuracy might be affected.
Q: How do you compensate publishers when it’s not clear exactly how Apple decides to send a conversion event?
A: Many networks are modifying their advertiser billing method to a solution where there is no ambiguity in publisher revenue attribution (e.g. CPM). This will certainly impact the economics of ad monetization.
Q: What are your thoughts on the SKAN privacy threshold and the potential impact on conversions reporting, and any post-install events?
A: The privacy threshold has a big impact in almost everything, so it is important to monitor it and check that the proposed conversion value schema is not affected by it.
(One note I’ll add here: Singular CTO Eran Friedman has been doing some significant research on this, and we’ll be sharing that on the Singular blog and in video form shortly.)
Q: How do you think mobile growth marketing departments will change their portfolio budget allocation across channels based on what we currently understand about the limitations imposed by major ad partners?
A: If deterministic targeting disappears, then a variety of new upper funnel channels will look more inviting, such as Influencer Marketing, TV, Podcasts, or even programmatic. On the latter, it was never able to catch up with IDFA-based targeting, but now that the field is more even, there’s a fair chance that performance might get on par with SDK-based channels.
Now, we’re just testing all the new approaches offered by our existing partners, so it’s a little bit early to say that our media mix will be radically affected.
Q: Are there any estimates about % of users opting to accept tracking? Do you recommend to ask before the iOS alert to try to have a better acceptance ratio?
A: We have heard about very different opt-in rates, from 10% to 70% in some cases. In regards to the actual flow to ask for opt-in, we are just starting to test it now, and while we’re missing a lot of data points, it seems that to get the best results, the dialogue has to be at the start of the game.
Q: Have you done any internal tests on how the consents ratio may look like?
A: Yes these tests are currently ongoing.
A: Very early to say as ATT enforcement communication may influence results.
Q: Will Rovio ask ATT permission in its games? Can you talk a bit the thinking behind?
A: It’s considered and very likely (some tests are currently live) but we need to assess properly the pros and cons of doing it, be it for attribution, ad monetisation, or even targeting in some cases.
Q: What percentage of ATT opt in are you seeing?
A: At this point we are the early stage of running tests and we expect the communications around ATT enforcement to influence consumer acceptance so at this point it is difficult to measure.
(One thing I’ll add: we found in surveys that 38.5% would allow IDFA tracking. However: real life is different than surveys, and … as this becomes commonplace, that will impact consumer behavior too: think GDPR cookie notifications on websites.)
Q: Have you increased your Android spends significantly due to iOS 14.5?
A: No we haven’t, and we don’t think it’ll be our strategy. iOS users are here to stay, and we believe we’re prepared enough with SKAN to be able to keep a relatively similar split between platforms.
Q: A bit of technical question: How do you deal with the fact that the conversion values can only be updated when the app is open?
A: At this point this is reality and conversion values can only be updated for players active every day.
(If I can read between the lines here: it sucks, and you’re going to lose data, but it is what it is. Your daily active users are likely to be your most engaged and profitable, in any case. Non-gaming verticals like finance and insurance, however, might want to consider engineering their systems for very early data accessibility.)
Q: Do you see a privacy threshold on conversion values? In principle could these be used for user attribution (conversion value – campaign – country combo can be semi-unique)? Do you think that is feasible?
A: There is a privacy threshold and the rules are not communicated by Apple.
(Personal note: I would caution against trying to engineer a more granular user-level attribution solution out of SKAdNetwork. It will be a lot of work but it will be against the spirit if not the letter of the App Store guidelines … and if you succeed Apple will close the loophole and waste all your effort at some point down the line.)
Q: How does Rovio overcome the challenge to identify user quality early in the funnel (optimizing on conversion events early in the funnel in order to not extensively delay the conversion value information from SKAdNetwork being send back to Rovio/MMP)?
A: We have different conversion value schemas that rely on the first 24 hours. We compare them to check which ones give better revenue estimates and keep the best one.
Q: A question about Rovio’s paper and conversion value censorship: how do you identify that a certain conversion value was censored? I don’t mean on the postback, but on your user base. The solution presented in the paper seems to assume it would be able to identify which are the conversion values that a censored postback could have. And it is not clear to me: how do you identify it?
A: The paper relies on the total count of null values (those that fall under the privacy threshold) and not in knowing exactly which conversion values were not reported.
Q: Does the number of distinct conversion values affect the rate at which privacy threshold is applied?
A: From our testing as of now, we don’t believe so. We think it’s more tied to install density in a publisher. With that said, there’s still a lot more to learn about the privacy threshold!
(Obligatory shameless plug here: Singular CTO Eran Friedman has been doing some significant research on this, and we’ll be sharing that on the Singular blog and in video form shortly.)
Q: Given we want to optimise on spenders (and not all users), and given that these spenders are a small percentage of the player base, how does the privacy threshold affect the messages from these players?
A: If the conversion value schema places few spenders in certain values then it is likely that all those will fall under the privacy threshold. It is a good idea to backtest the schemas under various assumptions to see how realistic a schema is in terms of privacy threshold.
Q: What strategies would you follow If you didn’t have all this structure, and are a small dev company using a MMP?
A: Leverage what MMP has built out. We are building on top of Singular and we expect their offering and our own to develop as we all learn.
A: Automate backtesting pipelines and be pragmatic.
Q: What is your plan for optimization? Because we can not understand which users come from which campaign exactly …
A: With our conversion schema modelling we are expecting to be able to probabilistically attribute players with a high degree of confidence.
Q: Can you please share your calculation for ROI given the limitations?
A: In the short run, we’ll be looking at the larger picture since we won’t be able to distinguish between paid and organic traffic accurately. So our ROAS targets on iOS will have to account for the revenue generated by organic traffic – we’re using historical data about the paid/organic ratio to set them. Then the CV and other methodologies should enable a more accurate reading on our paid ROAS.
A: You can check the details of the method in our paper on revenue attribution. At a glance, the method has two steps. First, we calculate the actual average revenue per conversion value and multiply it by the count of each conversion value. Then, we distribute the revenue that could not be attributed from known conversion values (because of privacy threshold) based on the distribution of null values.
Q: What is the main KPI of creative performance (CTR, IPM, Spend, ROAS) at Rovio and how will your creative testing change moving forward without the level of granularity in data we receive?
A: For the time being, our North Star is ROAS while keeping a close eye on IPMs. In the future on iOS we’ll have to rely further on upper funnel metrics, and our testing will have to be done on Android – on proxy devices – and extrapolated to iOS.
Q: During your tests, did you see any difference depending on apps’ user size? I wonder if a new app having relatively small user size could have meaningful data from conversion value.
A: In our case we are using different conversion value schemas for games in, for example, soft launch where the audience is smaller. It is a question of evaluating what schema works for a specific game.
Q: How has the switch over to SKAN changed the way UA is run?
One of the main changes we expect are with the campaign types offered by networks but we’re seeing a shift to using contextual data and develops on this side as well.
Q: Are there any special impacts for apps in the kid section category?
A: This is not a category where Rovio is that active. However, I understand SKAN is good news for this category …
(I’ll chime in here: given that previously you had almost no information on campaigns for kids category apps, SKAdNetwork is actually a pretty significant increase in the amount of attribution and efficacy data for mobile marketers.)
Q: Do you feel that Facebook is still going to support your creative testing needs? If not, what will be your likely solution moving forward?
A: Yes they will, but we’ll switch our focus to Android.
Q: Is it possible also to change the design of the ATT prompt? For A/B testing what works better? Would you rather test time when a prompt is shown or design of the prompt at the first open event?
A: You can not A/B test the system prompt and the text you have to provide in the plist the description “NSUserTrackingUsageDescription.” You can however A/B test when you ask, e.g. at start-up or at first ad, and also test having your own warm-up screen before popping the system prompt.
Next steps: webinar, or chat with us
If you didn’t already sign up for the webinar, jump on the opportunity now. Rovio talks through the design of their solution and shares a schematic which will be very interesting to other growth teams.
And, if you have additional questions, or you’re ready to start implementing, reserve a time to chat with an expert at Singular about how you can keep optimizing your growth on iOS 14.5.