10 insights from Rovio on maximizing SKAN for user acquisition
Fewer than 10% of mobile marketers in a recent Singular webinar rated their use of Apple’s SKAdNetwork framework for mobile attribution as “good.” Just over 40% said their efforts were “passable.” A full 50% said “we really have work to do.”
That’s one reason we recently released a really, really good guide to high performance user acquisition on iOS. (OK, I might be a little biased.) It’s also exactly why we partnered with Rovio on a very special webinar, One year of SKAN: The Rovio Journey.
My first and best advice: don’t delay, go sign up for that webinar, and enjoy it on demand.
My second: check out the following insights from that very special hour, where we talked about:
- Conversion models
- Predictive life-time value
- Campaign strategies
- Creative strategy
- Brand advertising post iOS 14
- First-party data
- Privacy thresholds
- Testing strategies
Rovio supplied an all-star and all-female group of experts, including Ann-Marie Pelkonen, Senior Product Manager, UA, Anastasia Nikitina, Senior Performance Marketing Specialist, and Erlin Gulbenkoglu, Senior Data Scientist. Here’s a selection of the insights they shared …
1. Pay the price and build a team
Everyone wants to achieve success with a minimum of effort, but that’s just simply not realistic in most areas of life and work. And it’s especially not realistic with SKAdNetwork.
“One key to our success has really been that we took SKAN seriously from the start. We really resourced, we researched and we tested really early on, so that has benefited us in the long run.”Ann-Marie Pelkonen
If you’re looking for success on iOS, you simply have to become an expert on SKAN. There’s just no way around it. Only perhaps 20-30% of the supply is accessible with probabilistic attribution, and that’s going away as well. You need to pick a team, resource it, and take some time to figure SKAN out. (Singular can help.)
2. Start simple and iterate
So many tried to do too much with conversion models and timers in the early days of SKAN. And that’s going to continue to be a problem with SKAN 4, coming out later this year, because SKAN 4 is even more complex. It’ll have much more opportunity and data, sure, but many more challenges as well.
“Regarding the conversion value schemas … we started simple and we have since adapted our approach to something a little bit more complex, but then keeping in mind still, the network requirements and then what’s best for our revenue attribution model that we built for iOS.”Ann-Marie Pelkonen
It was pretty hard to mess up IDFA-based attribution. It was simple, it was accessible, it was durable, and it was industry-standard. SKAN is much harder. You may dislike it or even hate it, or even think it’s garbage, as one of our recent webinar guests said.
The reality is as simple as it is harsh: it’s what there is.
Unless you’re going to ignore iOS and go Android-only (which is losing the GAID in 18 months or so), you simply have to figure it out.
3. Understand your own app’s signals
Unless you know the signals in your own app that predict user quality and profitability, you’re in for a world of hurt. After building up from simple steps, ensuring they understood SKAN, and building a working revenue attribution pipeline, Rovio worked on predictive lifetime value.
“Then we moved with more advanced things like pLTV, then we built the technology to support the pLTV conversion models. And then we have always been talking with stakeholders … we try to build something that gives us the best revenue attribution accuracy and also supports the networks that are using the signals from conversion values.Erlin Gulbenkoglu
Notice there are two things that Rovio’s using pLTV for:
- Long-term strategy: understanding the value of newly acquired players to continue to iterate marketing effectivness.
- Short-term optimization: feeding growth partners the signals that drive near realtime campaign efficiency.
4. Widen your data funnel to get more reporting signal
In your previous life as a mobile user acquisition specialist, IDFA was pretty much all you needed. Now that SKAdNetwork is the source of insight, you need more than one signal, as I’ve blogged about incessantly over the past year. (See “MMP in 2030: marketing measurement from the future.”)
Rovio’s doing exactly that.
“Now I need to use many more of different data sources than I used to use before. And I still cannot be 100% confident in the decision that I’m making.”Anastasia Nikitina
Now Rovio is looking at, among other sources, pre-install data and post-install data, and using both to ladder up to overall ROI performance. That includes CPIs and CPMs derived from SKAN plus post-install data including conversion and retention metrics. And Rovio is comparing all of this to organic performance and matching against expected results.
Truth is a little fuzzier post-IDFA. Rovio uses confidence intervals to guide decision-making, and makes decisions based on probabilities rather than certainties. (Let’s be honest: marketers have always done this. But now it’s a little more necessary.)
5. Demand DSP transparency
One of the key components of Apple’s mobile attribution framework is privacy thresholds to achieve crowd anonymity. Because of that, one of the non-Rovio participants in the webinar, David Phillipson — CEO and co-founder of Dataset — shared that transparency with your ad partners is critical.
“It’s absolutely key that … DSPs are absolutely transparent with the advertiser and with the MMP … by actually sharing [sub-publisher IDs] with Singular and Rovio, we’re actually able to establish what spend level per publisher we need to get to, to make sure we get the optimum amount of granular data out … [so] that we can put the best budgeting and pacing in place per sub-publisher within their budget, within their goals, to make sure that we get as much data out.”David Phillipson
In other words, by sharing which publisher is driving every single impression that leads to an app install, DSPs and ad partners can ensure that advertisers maximize installs per source, which maximizes signal. Without that critical step, there’s the chance that a given budget could be spread among so many apps and websites hosting the ads that Apple’s privacy thresholds simply obliterate a massive amount of the data you need to optimize on.
6. Run experiments on historical data
If I had a dollar for every time a mobile marketer told me “testing is the key” in a webinar or podcast, I’d have a free European vacation. Testing is essential, but tests are expensive too.
There’s one way you can test almost for free, however …
“We have run experiments on the historic data, like, when we could do a revenue attribution with IDFA and pLTV has given us lower revenue attribution errors, and we decided to go with that approach. And one advantage of that approach is basically we can use as many predictors as we want. We don’t need to encode our predictors into the 64 buckets we have [with SKAN].”Erlin Gulbenkoglu
Running tests on historical data allows you to pinpoint signals of high-value users in scenarios where you have all the data you need, so it’s definitely something to try.
A few obvious caveats, however:
- If your product has changed significantly, be aware this could skew insights.
- If people have changed significantly (Covid times versus “post” Covid, etc.) this can also result in insights that are not generalizable to today.
7. Consolidate campaigns
Privacy thresholds are essential to obscure individuals and achieve crowd anonymity. However, at the extreme, they block virtually all usable marketing optimization signals.
The solution: consolidation:
“The biggest change is probably that we had to consolidate our campaigns and increase the scale for those, and I would say this is the key for success basically on SKAN.”Anastasia Nikitina
Spreading too little budget over too many campaigns loses too much data. Consolidate campaigns so that privacy thresholds work as intended in achieving crowd anonymity without also taking marketing signal along with them as collateral damage.
8. Rethink targeting
Clearly, SKAdNetwork can work as a foundation for mobile marketing success. Clients using Singular’s SKAN Advanced Analytics are literally achieving 87% D7 revenue prediction accuracy on average. But that doesn’t mean that everything works. Retargeting largely died with the IDFA, and targeting isn’t that much better.
“We do not have that much of [an] opportunity to do targeting anymore.”Anastasia Nikitina
This of course could vary for different app publishers and advertisers. Rovio makes casual games, and the audience for casual games is large and diverse, which impacts the degree to which they need precise targeting. But the point remains: behavioral targeting without IDFA is hard if not impossible.
A few caveats:
- Contextual targeting can get you some value.
- Audiences assembled by third-party providers who own content fortresses (think Facebook, Reddit, Snap, Twitter, and others) could still work for targeting, but you won’t get specific data on them, of course.
- Privacy-safe targeting alternatives like Topics API in Google’s Privacy Sandbox for Android are additional potential options (though, obviously, not in this case on iOS).
9. Don’t (necessarily) pull learnings from Android
It’s tempting to test on Android, where you still have full GAID access, and use those learnings on iOS. This is potentially interesting, for example, in terms of creative strategy.
However, don’t assume it applies to iOS:
“We were relying a lot on Android to start with, but then since we’ve noticed that the same hero creatives don’t necessarily work both on iOS and Android. So that used to be something that historically was true to us, but we noticed that it’s no longer necessarily that way, so we’re trying to move creative testing more into iOS as well.”Ann-Marie Pelkonen
Looks like you’ll need to use some of those campaign identifiers in SKAN for creative, or lean on partners’ use of them. The good news: SKAN 4, with more granularity, is coming.
10. Leverage opt-in data
It may be scarce, but don’t abandon IDFA data just because it’s more challenging to get. You should still be asking for permission to track — and providing good reasons why — simply because this is an important additional source of data to help you check assumptions.
“We try to use opt-in data as well. We try to use it to validate if our revenue attribution is looking reasonable.”Erlin Gulbenkoglu
Don’t expect too much: note that Erlin says “try to use.”
But any data which is both deterministic and granular can be a big help, and some apps achieve opt-in rates exceeding 60%. That’s probably not your experience, as it is an outlier. Any data you do get, however, is helpful, and even if it’s only 10%, that can be a good survey sample size to test assumptions and check SKAN data.
Watch the whole webinar (and check out our upcoming SKAN 4 deep dive)
The entire webinar is available on-demand, for free, right here. I highly recommend you check it out.
In addition, as you know, Apple teased SKAN v 4 at WWDC just recently. This week I’ll be doing a deep dive into SKAN v4 with Singular CEO Gadi Eliashiv. We’ll be talking about:
- Crowd anonymity
- The new Source ID
- New conversion value types
- Multiple postbacks (3 in all)
- New timer behavior
- Web to app support
- Conversion values for the postbacks #2 and #3
- Modeling missing data
- Ecosystem changes
- Testing set-ups
- And more …